Fashion_MNIST_AutoEncoder_kkoma2623.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Fashion_MNIST_AutoEncoder_20143109.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/kkoma2623/VisualComputing/blob/master/Fashion_MNIST_AutoEncoder_kkoma2623.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jJRtx8qLXn9m",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "54c91aef-ecd9-430b-9fd5-4a4d9c84821b"
      },
      "source": [
        "# Import tensorflow and additional libraries\n",
        "import tensorflow as tf\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "\n",
        "tf.keras.backend.clear_session()\n",
        "\n",
        "# Check version\n",
        "print('tensorflow version: ', tf.__version__)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "tensorflow version:  2.2.0\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tOk4j7VEYd1g",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 215
        },
        "outputId": "4e662814-6295-481a-9d83-d4b9ff7a0c5a"
      },
      "source": [
        "# Load : Fashion MNIST Dataest\n",
        "(x_train, _), (x_test, _) = tf.keras.datasets.fashion_mnist.load_data()\n",
        "\n",
        "size_of_train  = len(x_train)\n",
        "size_of_test  = len(x_test)\n",
        "num_of_class = 10\n",
        "\n",
        "print('The Shape of dataset:', x_train[0].shape)\n",
        "print('The number of train image:', size_of_train)\n",
        "print('The number of test image:', size_of_test)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-labels-idx1-ubyte.gz\n",
            "32768/29515 [=================================] - 0s 0us/step\n",
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-images-idx3-ubyte.gz\n",
            "26427392/26421880 [==============================] - 0s 0us/step\n",
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-labels-idx1-ubyte.gz\n",
            "8192/5148 [===============================================] - 0s 0us/step\n",
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-images-idx3-ubyte.gz\n",
            "4423680/4422102 [==============================] - 0s 0us/step\n",
            "The Shape of dataset: (28, 28)\n",
            "The number of train image: 60000\n",
            "The number of test image: 10000\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vEuAuJhkfIS4",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 110
        },
        "outputId": "08067ed7-759f-4821-d819-0251a1a6ee82"
      },
      "source": [
        "# Train Data\n",
        "plt.figure(figsize=(16, 10))\n",
        "for i in range(num_of_class):\n",
        "    ax = plt.subplot(1, num_of_class, i+1)\n",
        "    plt.imshow(x_train[i], cmap='gray')\n",
        "    ax.get_xaxis().set_visible(False)\n",
        "    ax.get_yaxis().set_visible(False)\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x720 with 10 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8EZzJywn3sbH",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 143
        },
        "outputId": "f21f6cef-8999-48d2-8256-3739a39dfe39"
      },
      "source": [
        "print('== Before normalization ==')\n",
        "print('Image Shape:', x_train[0].shape)\n",
        "print('Value MIN: %d, MEAN:%.2f, MAX:%d'%(np.min(x_train[0]), np.mean(x_train[0]), np.max(x_train[0])))\n",
        "\n",
        "## Dataset Pre-processing ##\n",
        "# Normalization\n",
        "x_train = x_train/255.  \n",
        "x_test = x_test/255. \n",
        "\n",
        "# Reshape(Adding channel)\n",
        "x_train = x_train.reshape((size_of_train, 28, 28, 1))\n",
        "x_test = x_test.reshape((size_of_test, 28, 28, 1))\n",
        "\n",
        "print('\\n== After normalization ==')\n",
        "print('Image Shape:', x_train[0].shape)\n",
        "print('Value MIN: %d, MEAN:%.2f, MAX:%d'%(np.min(x_train[0]), np.mean(x_train[0]), np.max(x_train[0])))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "== Before normalization ==\n",
            "Image Shape: (28, 28)\n",
            "Value MIN: 0, MEAN:97.25, MAX:255\n",
            "\n",
            "== After normalization ==\n",
            "Image Shape: (28, 28, 1)\n",
            "Value MIN: 0, MEAN:0.38, MAX:1\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "expEYG3toOS9",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def channel_attention(x, channels, name):\n",
        "    ####Fill your code####\n",
        "    F_in = x\n",
        "    r = 4\n",
        "    F_in = tf.keras.layers.GlobalAveragePooling2D()(F_in)               # 64 vec\n",
        "    F_in = tf.keras.layers.Dense(channels//r)(F_in)                     # 16 vec => (64/4)\n",
        "    F_in = tf.keras.layers.BatchNormalization()(F_in)\n",
        "    F_in = tf.keras.layers.ReLU()(F_in)                                 # relu\n",
        "    F_in = tf.keras.layers.Dense(channels, activation='sigmoid')(F_in)  # 64 vec and sigmoid\n",
        "    F_out = x * F_in[:, None, None]\n",
        "\n",
        "    return  F_out\n",
        "\n",
        "def spatial_attention(x, channels, name):\n",
        "    ####Fill your code####\n",
        "    F_in = x\n",
        "    r = 4\n",
        "    F_in = tf.keras.layers.Conv2D(filters=channels//r, kernel_size=1, strides=(1,1), padding='same')(F_in)\n",
        "    F_in = tf.keras.layers.BatchNormalization()(F_in)\n",
        "    F_in = tf.keras.layers.ReLU()(F_in)\n",
        "    F_in = tf.keras.layers.Conv2D(filters=channels//r, kernel_size=3, strides=(1,1), padding='same')(F_in)\n",
        "    F_in = tf.keras.layers.BatchNormalization()(F_in)\n",
        "    F_in = tf.keras.layers.ReLU()(F_in)\n",
        "    F_in = tf.keras.layers.Conv2D(filters=channels//r, kernel_size=3, strides=(1,1), padding='same')(F_in)\n",
        "    F_in = tf.keras.layers.BatchNormalization()(F_in)\n",
        "    F_in = tf.keras.layers.ReLU()(F_in)\n",
        "    F_in = tf.keras.layers.Conv2D(1, kernel_size=1, strides=(1,1), padding='same', activation='sigmoid')(F_in)\n",
        "    F_out = x * F_in\n",
        "\n",
        "    return  F_out\n",
        " \n",
        "def autoencoder_block(x, filter, name): \n",
        "    x = tf.keras.layers.Conv2D(filters=filter, kernel_size=3, strides=(1, 1), padding='same', name=name+'_conv')(x)\n",
        "    x = tf.keras.layers.BatchNormalization(name=name+'_bn')(x)\n",
        "    x = tf.keras.layers.ReLU(name=name+'_relu')(x)\n",
        "    return x"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0tQdIwM3fShI",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "##############################################\n",
        "# Add Attention Block in AutoEncoder network #\n",
        "##############################################\n",
        "\n",
        "def autoencoder(input):\n",
        "    # Encoder\n",
        "    x = autoencoder_block(input, 16, 'encoder1')\n",
        "\n",
        "    x = tf.keras.layers.MaxPooling2D((2, 2), name='encoder_pool1')(x)\n",
        "    x = autoencoder_block(x, 32, 'encoder2')\n",
        "\n",
        "    x = tf.keras.layers.MaxPooling2D((2, 2), name='encoder_pool2')(x)\n",
        "    \n",
        "    encoded_img = autoencoder_block(x, 64, 'encoder3')\n",
        "\n",
        "    # Decoder\n",
        "    x = autoencoder_block(encoded_img, 64, 'decoder1')\n",
        "\n",
        "    x = tf.keras.layers.UpSampling2D((2, 2), name='decoder_up1')(x)\n",
        "    x = autoencoder_block(x, 32, 'decoder2')\n",
        " \n",
        "    x = tf.keras.layers.UpSampling2D((2, 2), name='decoder_up2')(x)\n",
        "    x = autoencoder_block(x, 16, 'decoder3')\n",
        "\n",
        "    decoded_img = tf.keras.layers.Conv2D(filters=1, kernel_size=3, strides=(1, 1), \n",
        "                                         activation='sigmoid', padding='same', \n",
        "                                         name='last_conv')(x)\n",
        "    return decoded_img\n",
        "\n",
        "def autoencoder_channel(input):\n",
        "    # Encoder\n",
        "    x = autoencoder_block(input, 16, 'encoder1')\n",
        "    \n",
        "    x = channel_attention(x, 16, 'channel_attention1')\n",
        "    x = tf.keras.layers.MaxPooling2D((2, 2), name='encoder_pool1')(x)\n",
        "    x = autoencoder_block(x, 32, 'encoder2')\n",
        "\n",
        "    x = channel_attention(x, 32, 'channel_attention2')\n",
        "    x = tf.keras.layers.MaxPooling2D((2, 2), name='encoder_pool2')(x)\n",
        "    \n",
        "    encoded_img = autoencoder_block(x, 64, 'encoder3')\n",
        "\n",
        "    # Decoder\n",
        "    x = autoencoder_block(encoded_img, 64, 'decoder1')\n",
        "\n",
        "    x = channel_attention(x, 64, 'channel_attention3')\n",
        "    x = tf.keras.layers.UpSampling2D((2, 2), name='decoder_up1')(x)\n",
        "    x = autoencoder_block(x, 32, 'decoder2')\n",
        "\n",
        "    x = channel_attention(x, 32, 'channel_attention4')\n",
        "    x = tf.keras.layers.UpSampling2D((2, 2), name='decoder_up2')(x)\n",
        "    x = autoencoder_block(x, 16, 'decoder3')\n",
        "\n",
        "    decoded_img = tf.keras.layers.Conv2D(filters=1, kernel_size=3, strides=(1, 1), \n",
        "                                         activation='sigmoid', padding='same', \n",
        "                                         name='last_conv')(x)\n",
        "    return decoded_img    \n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-3THclvQP853",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 3113
        },
        "outputId": "e9f96119-7bdc-4367-a4f6-884a4de83379"
      },
      "source": [
        "# Normal AutoEncoder Model Create / Summary\n",
        "input_data = tf.keras.Input(shape=(28, 28, 1))\n",
        "\n",
        "output_data = autoencoder(input_data)\n",
        "output_channel_data = autoencoder_channel(input_data)\n",
        "\n",
        "model = tf.keras.Model(inputs=input_data, outputs=output_data)\n",
        "model_channel  = tf.keras.Model(inputs=input_data, outputs=output_channel_data)\n",
        "\n",
        "model.summary()\n",
        "model_channel.summary()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"model\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "input_1 (InputLayer)         [(None, 28, 28, 1)]       0         \n",
            "_________________________________________________________________\n",
            "encoder1_conv (Conv2D)       (None, 28, 28, 16)        160       \n",
            "_________________________________________________________________\n",
            "encoder1_bn (BatchNormalizat (None, 28, 28, 16)        64        \n",
            "_________________________________________________________________\n",
            "encoder1_relu (ReLU)         (None, 28, 28, 16)        0         \n",
            "_________________________________________________________________\n",
            "encoder_pool1 (MaxPooling2D) (None, 14, 14, 16)        0         \n",
            "_________________________________________________________________\n",
            "encoder2_conv (Conv2D)       (None, 14, 14, 32)        4640      \n",
            "_________________________________________________________________\n",
            "encoder2_bn (BatchNormalizat (None, 14, 14, 32)        128       \n",
            "_________________________________________________________________\n",
            "encoder2_relu (ReLU)         (None, 14, 14, 32)        0         \n",
            "_________________________________________________________________\n",
            "encoder_pool2 (MaxPooling2D) (None, 7, 7, 32)          0         \n",
            "_________________________________________________________________\n",
            "encoder3_conv (Conv2D)       (None, 7, 7, 64)          18496     \n",
            "_________________________________________________________________\n",
            "encoder3_bn (BatchNormalizat (None, 7, 7, 64)          256       \n",
            "_________________________________________________________________\n",
            "encoder3_relu (ReLU)         (None, 7, 7, 64)          0         \n",
            "_________________________________________________________________\n",
            "decoder1_conv (Conv2D)       (None, 7, 7, 64)          36928     \n",
            "_________________________________________________________________\n",
            "decoder1_bn (BatchNormalizat (None, 7, 7, 64)          256       \n",
            "_________________________________________________________________\n",
            "decoder1_relu (ReLU)         (None, 7, 7, 64)          0         \n",
            "_________________________________________________________________\n",
            "decoder_up1 (UpSampling2D)   (None, 14, 14, 64)        0         \n",
            "_________________________________________________________________\n",
            "decoder2_conv (Conv2D)       (None, 14, 14, 32)        18464     \n",
            "_________________________________________________________________\n",
            "decoder2_bn (BatchNormalizat (None, 14, 14, 32)        128       \n",
            "_________________________________________________________________\n",
            "decoder2_relu (ReLU)         (None, 14, 14, 32)        0         \n",
            "_________________________________________________________________\n",
            "decoder_up2 (UpSampling2D)   (None, 28, 28, 32)        0         \n",
            "_________________________________________________________________\n",
            "decoder3_conv (Conv2D)       (None, 28, 28, 16)        4624      \n",
            "_________________________________________________________________\n",
            "decoder3_bn (BatchNormalizat (None, 28, 28, 16)        64        \n",
            "_________________________________________________________________\n",
            "decoder3_relu (ReLU)         (None, 28, 28, 16)        0         \n",
            "_________________________________________________________________\n",
            "last_conv (Conv2D)           (None, 28, 28, 1)         145       \n",
            "=================================================================\n",
            "Total params: 84,353\n",
            "Trainable params: 83,905\n",
            "Non-trainable params: 448\n",
            "_________________________________________________________________\n",
            "Model: \"model_1\"\n",
            "__________________________________________________________________________________________________\n",
            "Layer (type)                    Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            "input_1 (InputLayer)            [(None, 28, 28, 1)]  0                                            \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_conv (Conv2D)          (None, 28, 28, 16)   160         input_1[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_bn (BatchNormalization (None, 28, 28, 16)   64          encoder1_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_relu (ReLU)            (None, 28, 28, 16)   0           encoder1_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d (Globa (None, 16)           0           encoder1_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "dense (Dense)                   (None, 4)            68          global_average_pooling2d[0][0]   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization (BatchNorma (None, 4)            16          dense[0][0]                      \n",
            "__________________________________________________________________________________________________\n",
            "re_lu (ReLU)                    (None, 4)            0           batch_normalization[0][0]        \n",
            "__________________________________________________________________________________________________\n",
            "dense_1 (Dense)                 (None, 16)           80          re_lu[0][0]                      \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice (Tens [(None, 1, 1, 16)]   0           dense_1[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul (TensorFlowOpLa [(None, 28, 28, 16)] 0           encoder1_relu[0][0]              \n",
            "                                                                 tf_op_layer_strided_slice[0][0]  \n",
            "__________________________________________________________________________________________________\n",
            "encoder_pool1 (MaxPooling2D)    (None, 14, 14, 16)   0           tf_op_layer_Mul[0][0]            \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_conv (Conv2D)          (None, 14, 14, 32)   4640        encoder_pool1[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_bn (BatchNormalization (None, 14, 14, 32)   128         encoder2_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_relu (ReLU)            (None, 14, 14, 32)   0           encoder2_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_1 (Glo (None, 32)           0           encoder2_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "dense_2 (Dense)                 (None, 8)            264         global_average_pooling2d_1[0][0] \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_1 (BatchNor (None, 8)            32          dense_2[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_1 (ReLU)                  (None, 8)            0           batch_normalization_1[0][0]      \n",
            "__________________________________________________________________________________________________\n",
            "dense_3 (Dense)                 (None, 32)           288         re_lu_1[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_1 (Te [(None, 1, 1, 32)]   0           dense_3[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_1 (TensorFlowOp [(None, 14, 14, 32)] 0           encoder2_relu[0][0]              \n",
            "                                                                 tf_op_layer_strided_slice_1[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "encoder_pool2 (MaxPooling2D)    (None, 7, 7, 32)     0           tf_op_layer_Mul_1[0][0]          \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_conv (Conv2D)          (None, 7, 7, 64)     18496       encoder_pool2[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_bn (BatchNormalization (None, 7, 7, 64)     256         encoder3_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_relu (ReLU)            (None, 7, 7, 64)     0           encoder3_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_conv (Conv2D)          (None, 7, 7, 64)     36928       encoder3_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_bn (BatchNormalization (None, 7, 7, 64)     256         decoder1_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_relu (ReLU)            (None, 7, 7, 64)     0           decoder1_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_2 (Glo (None, 64)           0           decoder1_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "dense_4 (Dense)                 (None, 16)           1040        global_average_pooling2d_2[0][0] \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_2 (BatchNor (None, 16)           64          dense_4[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_2 (ReLU)                  (None, 16)           0           batch_normalization_2[0][0]      \n",
            "__________________________________________________________________________________________________\n",
            "dense_5 (Dense)                 (None, 64)           1088        re_lu_2[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_2 (Te [(None, 1, 1, 64)]   0           dense_5[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_2 (TensorFlowOp [(None, 7, 7, 64)]   0           decoder1_relu[0][0]              \n",
            "                                                                 tf_op_layer_strided_slice_2[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "decoder_up1 (UpSampling2D)      (None, 14, 14, 64)   0           tf_op_layer_Mul_2[0][0]          \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_conv (Conv2D)          (None, 14, 14, 32)   18464       decoder_up1[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_bn (BatchNormalization (None, 14, 14, 32)   128         decoder2_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_relu (ReLU)            (None, 14, 14, 32)   0           decoder2_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_3 (Glo (None, 32)           0           decoder2_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "dense_6 (Dense)                 (None, 8)            264         global_average_pooling2d_3[0][0] \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_3 (BatchNor (None, 8)            32          dense_6[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_3 (ReLU)                  (None, 8)            0           batch_normalization_3[0][0]      \n",
            "__________________________________________________________________________________________________\n",
            "dense_7 (Dense)                 (None, 32)           288         re_lu_3[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_3 (Te [(None, 1, 1, 32)]   0           dense_7[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_3 (TensorFlowOp [(None, 14, 14, 32)] 0           decoder2_relu[0][0]              \n",
            "                                                                 tf_op_layer_strided_slice_3[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "decoder_up2 (UpSampling2D)      (None, 28, 28, 32)   0           tf_op_layer_Mul_3[0][0]          \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_conv (Conv2D)          (None, 28, 28, 16)   4624        decoder_up2[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_bn (BatchNormalization (None, 28, 28, 16)   64          decoder3_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_relu (ReLU)            (None, 28, 28, 16)   0           decoder3_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "last_conv (Conv2D)              (None, 28, 28, 1)    145         decoder3_relu[0][0]              \n",
            "==================================================================================================\n",
            "Total params: 87,877\n",
            "Trainable params: 87,357\n",
            "Non-trainable params: 520\n",
            "__________________________________________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "D_NeT6qxiEDK",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Model Compile\n",
        "model.compile(optimizer=tf.keras.optimizers.Adam(),\n",
        "              loss=tf.keras.losses.MeanSquaredError(),\n",
        "              metrics=['mse'])\n",
        "model_channel.compile(optimizer=tf.keras.optimizers.Adam(), loss=tf.keras.losses.MeanSquaredError(), metrics=['mse'])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "azM29ta9i25r",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1835
        },
        "outputId": "ec27e668-6d13-4ffa-9538-29afee70ccd2"
      },
      "source": [
        "# Training\n",
        "print('Training Model')\n",
        "epoch = 25\n",
        "# AutoEncoder는 input과 output이 같으므로 x_train을 넣어줌\n",
        "history = model.fit(x_train, x_train, \n",
        "                    batch_size=16, \n",
        "                    epochs=epoch,\n",
        "                    shuffle=True,\n",
        "                    validation_data=(x_test, x_test))\n",
        "\n",
        "history_channel = model_channel.fit(x_train, x_train, \n",
        "                    batch_size=16, \n",
        "                    epochs=epoch,\n",
        "                    shuffle=True,\n",
        "                    validation_data=(x_test, x_test))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Training Model\n",
            "Epoch 1/25\n",
            "3750/3750 [==============================] - 15s 4ms/step - loss: 0.0060 - mse: 0.0060 - val_loss: 0.0037 - val_mse: 0.0037\n",
            "Epoch 2/25\n",
            "3750/3750 [==============================] - 15s 4ms/step - loss: 0.0033 - mse: 0.0033 - val_loss: 0.0029 - val_mse: 0.0029\n",
            "Epoch 3/25\n",
            "3750/3750 [==============================] - 16s 4ms/step - loss: 0.0028 - mse: 0.0028 - val_loss: 0.0025 - val_mse: 0.0025\n",
            "Epoch 4/25\n",
            "3750/3750 [==============================] - 15s 4ms/step - loss: 0.0025 - mse: 0.0025 - val_loss: 0.0024 - val_mse: 0.0024\n",
            "Epoch 5/25\n",
            "3750/3750 [==============================] - 15s 4ms/step - loss: 0.0024 - mse: 0.0024 - val_loss: 0.0024 - val_mse: 0.0024\n",
            "Epoch 6/25\n",
            "3750/3750 [==============================] - 15s 4ms/step - loss: 0.0023 - mse: 0.0023 - val_loss: 0.0024 - val_mse: 0.0024\n",
            "Epoch 7/25\n",
            "3750/3750 [==============================] - 15s 4ms/step - loss: 0.0022 - mse: 0.0022 - val_loss: 0.0021 - val_mse: 0.0021\n",
            "Epoch 8/25\n",
            "3750/3750 [==============================] - 15s 4ms/step - loss: 0.0021 - mse: 0.0021 - val_loss: 0.0021 - val_mse: 0.0021\n",
            "Epoch 9/25\n",
            "3750/3750 [==============================] - 14s 4ms/step - loss: 0.0021 - mse: 0.0021 - val_loss: 0.0020 - val_mse: 0.0020\n",
            "Epoch 10/25\n",
            "3750/3750 [==============================] - 15s 4ms/step - loss: 0.0020 - mse: 0.0020 - val_loss: 0.0024 - val_mse: 0.0024\n",
            "Epoch 11/25\n",
            "3750/3750 [==============================] - 15s 4ms/step - loss: 0.0020 - mse: 0.0020 - val_loss: 0.0023 - val_mse: 0.0023\n",
            "Epoch 12/25\n",
            "3750/3750 [==============================] - 14s 4ms/step - loss: 0.0019 - mse: 0.0019 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 13/25\n",
            "3750/3750 [==============================] - 14s 4ms/step - loss: 0.0019 - mse: 0.0019 - val_loss: 0.0020 - val_mse: 0.0020\n",
            "Epoch 14/25\n",
            "3750/3750 [==============================] - 14s 4ms/step - loss: 0.0019 - mse: 0.0019 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 15/25\n",
            "3750/3750 [==============================] - 14s 4ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 16/25\n",
            "3750/3750 [==============================] - 14s 4ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 17/25\n",
            "3750/3750 [==============================] - 15s 4ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 18/25\n",
            "3750/3750 [==============================] - 15s 4ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 19/25\n",
            "3750/3750 [==============================] - 14s 4ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0020 - val_mse: 0.0020\n",
            "Epoch 20/25\n",
            "3750/3750 [==============================] - 14s 4ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 21/25\n",
            "3750/3750 [==============================] - 14s 4ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 22/25\n",
            "3750/3750 [==============================] - 15s 4ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 23/25\n",
            "3750/3750 [==============================] - 14s 4ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 24/25\n",
            "3750/3750 [==============================] - 15s 4ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 25/25\n",
            "3750/3750 [==============================] - 15s 4ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 1/25\n",
            "3750/3750 [==============================] - 23s 6ms/step - loss: 0.0063 - mse: 0.0063 - val_loss: 0.0037 - val_mse: 0.0037\n",
            "Epoch 2/25\n",
            "3750/3750 [==============================] - 23s 6ms/step - loss: 0.0034 - mse: 0.0034 - val_loss: 0.0030 - val_mse: 0.0030\n",
            "Epoch 3/25\n",
            "3750/3750 [==============================] - 23s 6ms/step - loss: 0.0029 - mse: 0.0029 - val_loss: 0.0030 - val_mse: 0.0030\n",
            "Epoch 4/25\n",
            "3750/3750 [==============================] - 23s 6ms/step - loss: 0.0026 - mse: 0.0026 - val_loss: 0.0025 - val_mse: 0.0025\n",
            "Epoch 5/25\n",
            "3750/3750 [==============================] - 23s 6ms/step - loss: 0.0024 - mse: 0.0024 - val_loss: 0.0022 - val_mse: 0.0022\n",
            "Epoch 6/25\n",
            "3750/3750 [==============================] - 23s 6ms/step - loss: 0.0023 - mse: 0.0023 - val_loss: 0.0022 - val_mse: 0.0022\n",
            "Epoch 7/25\n",
            "3750/3750 [==============================] - 23s 6ms/step - loss: 0.0022 - mse: 0.0022 - val_loss: 0.0029 - val_mse: 0.0029\n",
            "Epoch 8/25\n",
            "3750/3750 [==============================] - 23s 6ms/step - loss: 0.0021 - mse: 0.0021 - val_loss: 0.0027 - val_mse: 0.0027\n",
            "Epoch 9/25\n",
            "3750/3750 [==============================] - 24s 6ms/step - loss: 0.0020 - mse: 0.0020 - val_loss: 0.0020 - val_mse: 0.0020\n",
            "Epoch 10/25\n",
            "3750/3750 [==============================] - 23s 6ms/step - loss: 0.0020 - mse: 0.0020 - val_loss: 0.0021 - val_mse: 0.0021\n",
            "Epoch 11/25\n",
            "3750/3750 [==============================] - 23s 6ms/step - loss: 0.0019 - mse: 0.0019 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 12/25\n",
            "3750/3750 [==============================] - 22s 6ms/step - loss: 0.0019 - mse: 0.0019 - val_loss: 0.0021 - val_mse: 0.0021\n",
            "Epoch 13/25\n",
            "3750/3750 [==============================] - 23s 6ms/step - loss: 0.0019 - mse: 0.0019 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 14/25\n",
            "3750/3750 [==============================] - 23s 6ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 15/25\n",
            "3750/3750 [==============================] - 22s 6ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 16/25\n",
            "3750/3750 [==============================] - 22s 6ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 17/25\n",
            "3750/3750 [==============================] - 22s 6ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 18/25\n",
            "3750/3750 [==============================] - 22s 6ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 19/25\n",
            "3750/3750 [==============================] - 22s 6ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 20/25\n",
            "3750/3750 [==============================] - 22s 6ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 21/25\n",
            "3750/3750 [==============================] - 22s 6ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 22/25\n",
            "3750/3750 [==============================] - 22s 6ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0020 - val_mse: 0.0020\n",
            "Epoch 23/25\n",
            "3750/3750 [==============================] - 23s 6ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0024 - val_mse: 0.0024\n",
            "Epoch 24/25\n",
            "3750/3750 [==============================] - 22s 6ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 25/25\n",
            "3750/3750 [==============================] - 22s 6ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0020 - val_mse: 0.0020\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0Z5lwV09Y3V1",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def autoencoder_spatial(input):\n",
        "    # Encoder\n",
        "    x = autoencoder_block(input, 16, 'encoder1')\n",
        "    \n",
        "    x = spatial_attention(x, 16, 'spatial_attention1')\n",
        "    x = tf.keras.layers.MaxPooling2D((2, 2), name='encoder_pool1')(x)\n",
        "    x = autoencoder_block(x, 32, 'encoder2')\n",
        "\n",
        "    x = spatial_attention(x, 32, 'spatial_attention2')\n",
        "    x = tf.keras.layers.MaxPooling2D((2, 2), name='encoder_pool2')(x)\n",
        "    \n",
        "    encoded_img = autoencoder_block(x, 64, 'encoder3')\n",
        "\n",
        "    # Decoder\n",
        "    x = autoencoder_block(encoded_img, 64, 'decoder1')\n",
        "\n",
        "    x = spatial_attention(x, 64, 'spatial_attention3')\n",
        "    x = tf.keras.layers.UpSampling2D((2, 2), name='decoder_up1')(x)\n",
        "    x = autoencoder_block(x, 32, 'decoder2')\n",
        "\n",
        "    x = spatial_attention(x, 32, 'spatial_attention4')\n",
        "    x = tf.keras.layers.UpSampling2D((2, 2), name='decoder_up2')(x)\n",
        "    x = autoencoder_block(x, 16, 'decoder3')\n",
        "\n",
        "    decoded_img = tf.keras.layers.Conv2D(filters=1, kernel_size=3, strides=(1, 1), \n",
        "                                         activation='sigmoid', padding='same', \n",
        "                                         name='last_conv')(x)\n",
        "    return decoded_img   "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mwvH4F-JZaYN",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 2681
        },
        "outputId": "bde58b5d-15db-4fcc-ea4a-235a7737fad6"
      },
      "source": [
        "# Normal AutoEncoder Model Create / Summary\n",
        "input_autoencoder_spatial_data = tf.keras.Input(shape=(28, 28, 1))\n",
        "\n",
        "output_autoencoder_spatial_data = autoencoder_spatial(input_autoencoder_spatial_data)\n",
        "\n",
        "model_autoencoder_spatial  = tf.keras.Model(inputs=input_autoencoder_spatial_data, outputs=output_autoencoder_spatial_data)\n",
        "\n",
        "model_autoencoder_spatial.summary()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"model_2\"\n",
            "__________________________________________________________________________________________________\n",
            "Layer (type)                    Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            "input_2 (InputLayer)            [(None, 28, 28, 1)]  0                                            \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_conv (Conv2D)          (None, 28, 28, 16)   160         input_2[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_bn (BatchNormalization (None, 28, 28, 16)   64          encoder1_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_relu (ReLU)            (None, 28, 28, 16)   0           encoder1_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "conv2d (Conv2D)                 (None, 28, 28, 4)    68          encoder1_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_4 (BatchNor (None, 28, 28, 4)    16          conv2d[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_4 (ReLU)                  (None, 28, 28, 4)    0           batch_normalization_4[0][0]      \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_1 (Conv2D)               (None, 28, 28, 4)    148         re_lu_4[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_5 (BatchNor (None, 28, 28, 4)    16          conv2d_1[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_5 (ReLU)                  (None, 28, 28, 4)    0           batch_normalization_5[0][0]      \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_2 (Conv2D)               (None, 28, 28, 4)    148         re_lu_5[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_6 (BatchNor (None, 28, 28, 4)    16          conv2d_2[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_6 (ReLU)                  (None, 28, 28, 4)    0           batch_normalization_6[0][0]      \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_3 (Conv2D)               (None, 28, 28, 1)    5           re_lu_6[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_4 (TensorFlowOp [(None, 28, 28, 16)] 0           encoder1_relu[0][0]              \n",
            "                                                                 conv2d_3[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "encoder_pool1 (MaxPooling2D)    (None, 14, 14, 16)   0           tf_op_layer_Mul_4[0][0]          \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_conv (Conv2D)          (None, 14, 14, 32)   4640        encoder_pool1[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_bn (BatchNormalization (None, 14, 14, 32)   128         encoder2_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_relu (ReLU)            (None, 14, 14, 32)   0           encoder2_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_4 (Conv2D)               (None, 14, 14, 8)    264         encoder2_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_7 (BatchNor (None, 14, 14, 8)    32          conv2d_4[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_7 (ReLU)                  (None, 14, 14, 8)    0           batch_normalization_7[0][0]      \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_5 (Conv2D)               (None, 14, 14, 8)    584         re_lu_7[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_8 (BatchNor (None, 14, 14, 8)    32          conv2d_5[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_8 (ReLU)                  (None, 14, 14, 8)    0           batch_normalization_8[0][0]      \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_6 (Conv2D)               (None, 14, 14, 8)    584         re_lu_8[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_9 (BatchNor (None, 14, 14, 8)    32          conv2d_6[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_9 (ReLU)                  (None, 14, 14, 8)    0           batch_normalization_9[0][0]      \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_7 (Conv2D)               (None, 14, 14, 1)    9           re_lu_9[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_5 (TensorFlowOp [(None, 14, 14, 32)] 0           encoder2_relu[0][0]              \n",
            "                                                                 conv2d_7[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "encoder_pool2 (MaxPooling2D)    (None, 7, 7, 32)     0           tf_op_layer_Mul_5[0][0]          \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_conv (Conv2D)          (None, 7, 7, 64)     18496       encoder_pool2[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_bn (BatchNormalization (None, 7, 7, 64)     256         encoder3_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_relu (ReLU)            (None, 7, 7, 64)     0           encoder3_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_conv (Conv2D)          (None, 7, 7, 64)     36928       encoder3_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_bn (BatchNormalization (None, 7, 7, 64)     256         decoder1_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_relu (ReLU)            (None, 7, 7, 64)     0           decoder1_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_8 (Conv2D)               (None, 7, 7, 16)     1040        decoder1_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_10 (BatchNo (None, 7, 7, 16)     64          conv2d_8[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_10 (ReLU)                 (None, 7, 7, 16)     0           batch_normalization_10[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_9 (Conv2D)               (None, 7, 7, 16)     2320        re_lu_10[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_11 (BatchNo (None, 7, 7, 16)     64          conv2d_9[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_11 (ReLU)                 (None, 7, 7, 16)     0           batch_normalization_11[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_10 (Conv2D)              (None, 7, 7, 16)     2320        re_lu_11[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_12 (BatchNo (None, 7, 7, 16)     64          conv2d_10[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_12 (ReLU)                 (None, 7, 7, 16)     0           batch_normalization_12[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_11 (Conv2D)              (None, 7, 7, 1)      17          re_lu_12[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_6 (TensorFlowOp [(None, 7, 7, 64)]   0           decoder1_relu[0][0]              \n",
            "                                                                 conv2d_11[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "decoder_up1 (UpSampling2D)      (None, 14, 14, 64)   0           tf_op_layer_Mul_6[0][0]          \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_conv (Conv2D)          (None, 14, 14, 32)   18464       decoder_up1[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_bn (BatchNormalization (None, 14, 14, 32)   128         decoder2_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_relu (ReLU)            (None, 14, 14, 32)   0           decoder2_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_12 (Conv2D)              (None, 14, 14, 8)    264         decoder2_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_13 (BatchNo (None, 14, 14, 8)    32          conv2d_12[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_13 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_13[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_13 (Conv2D)              (None, 14, 14, 8)    584         re_lu_13[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_14 (BatchNo (None, 14, 14, 8)    32          conv2d_13[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_14 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_14[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_14 (Conv2D)              (None, 14, 14, 8)    584         re_lu_14[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_15 (BatchNo (None, 14, 14, 8)    32          conv2d_14[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_15 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_15[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_15 (Conv2D)              (None, 14, 14, 1)    9           re_lu_15[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_7 (TensorFlowOp [(None, 14, 14, 32)] 0           decoder2_relu[0][0]              \n",
            "                                                                 conv2d_15[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "decoder_up2 (UpSampling2D)      (None, 28, 28, 32)   0           tf_op_layer_Mul_7[0][0]          \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_conv (Conv2D)          (None, 28, 28, 16)   4624        decoder_up2[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_bn (BatchNormalization (None, 28, 28, 16)   64          decoder3_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_relu (ReLU)            (None, 28, 28, 16)   0           decoder3_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "last_conv (Conv2D)              (None, 28, 28, 1)    145         decoder3_relu[0][0]              \n",
            "==================================================================================================\n",
            "Total params: 93,733\n",
            "Trainable params: 93,069\n",
            "Non-trainable params: 664\n",
            "__________________________________________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SWPleq-uaYii",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Model Compile\n",
        "model_autoencoder_spatial.compile(optimizer=tf.keras.optimizers.Adam(), loss=tf.keras.losses.MeanSquaredError(), metrics=['mse'])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YW6w6lBjacoQ",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 935
        },
        "outputId": "de1df318-b563-4a36-8aa4-9a9a223fe7e8"
      },
      "source": [
        "# Training\n",
        "print('Training Model - autoencoder-spatial')\n",
        "epoch = 25\n",
        "# AutoEncoder는 input과 output이 같으므로 x_train을 넣어줌\n",
        "history_autoencoder_spatial = model_autoencoder_spatial.fit(x_train, x_train, \n",
        "                    batch_size=16, \n",
        "                    epochs=epoch,\n",
        "                    shuffle=True,\n",
        "                    validation_data=(x_test, x_test))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Training Model - autoencoder-spatial\n",
            "Epoch 1/25\n",
            "3750/3750 [==============================] - 31s 8ms/step - loss: 0.0066 - mse: 0.0066 - val_loss: 0.0041 - val_mse: 0.0041\n",
            "Epoch 2/25\n",
            "3750/3750 [==============================] - 30s 8ms/step - loss: 0.0033 - mse: 0.0033 - val_loss: 0.0028 - val_mse: 0.0028\n",
            "Epoch 3/25\n",
            "3750/3750 [==============================] - 29s 8ms/step - loss: 0.0027 - mse: 0.0027 - val_loss: 0.0026 - val_mse: 0.0026\n",
            "Epoch 4/25\n",
            "3750/3750 [==============================] - 29s 8ms/step - loss: 0.0024 - mse: 0.0024 - val_loss: 0.0024 - val_mse: 0.0024\n",
            "Epoch 5/25\n",
            "3750/3750 [==============================] - 29s 8ms/step - loss: 0.0022 - mse: 0.0022 - val_loss: 0.0024 - val_mse: 0.0024\n",
            "Epoch 6/25\n",
            "3750/3750 [==============================] - 29s 8ms/step - loss: 0.0021 - mse: 0.0021 - val_loss: 0.0024 - val_mse: 0.0024\n",
            "Epoch 7/25\n",
            "3750/3750 [==============================] - 29s 8ms/step - loss: 0.0020 - mse: 0.0020 - val_loss: 0.0023 - val_mse: 0.0023\n",
            "Epoch 8/25\n",
            "3750/3750 [==============================] - 29s 8ms/step - loss: 0.0019 - mse: 0.0019 - val_loss: 0.0022 - val_mse: 0.0022\n",
            "Epoch 9/25\n",
            "3750/3750 [==============================] - 30s 8ms/step - loss: 0.0019 - mse: 0.0019 - val_loss: 0.0020 - val_mse: 0.0020\n",
            "Epoch 10/25\n",
            "3750/3750 [==============================] - 29s 8ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 11/25\n",
            "3750/3750 [==============================] - 29s 8ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 12/25\n",
            "3750/3750 [==============================] - 29s 8ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 13/25\n",
            "3750/3750 [==============================] - 29s 8ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 14/25\n",
            "3750/3750 [==============================] - 29s 8ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 15/25\n",
            "3750/3750 [==============================] - 29s 8ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 16/25\n",
            "3750/3750 [==============================] - 29s 8ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0015 - val_mse: 0.0015\n",
            "Epoch 17/25\n",
            "3750/3750 [==============================] - 29s 8ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 18/25\n",
            "3750/3750 [==============================] - 29s 8ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 19/25\n",
            "3750/3750 [==============================] - 29s 8ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 20/25\n",
            "3750/3750 [==============================] - 30s 8ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 21/25\n",
            "3750/3750 [==============================] - 30s 8ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 22/25\n",
            "3750/3750 [==============================] - 30s 8ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 23/25\n",
            "3750/3750 [==============================] - 30s 8ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 24/25\n",
            "3750/3750 [==============================] - 30s 8ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 25/25\n",
            "3750/3750 [==============================] - 29s 8ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0014 - val_mse: 0.0014\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eKa4czjSZX47",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def autoencoder_parallel_add(input):\n",
        "    # Encoder\n",
        "    x = autoencoder_block(input, 16, 'encoder1')\n",
        "    \n",
        "    x_channel = channel_attention(x, 16, 'channel_attention')\n",
        "    x_spatial = spatial_attention(x, 16, 'spatial_attention')\n",
        "    x = tf.keras.layers.add([x_channel, x_spatial])\n",
        "    \n",
        "    x = tf.keras.layers.MaxPooling2D((2, 2), name='encoder_pool1')(x)\n",
        "    x = autoencoder_block(x, 32, 'encoder2')\n",
        "\n",
        "    x_channel = channel_attention(x, 32, 'channel_attention2')\n",
        "    x_spatial = spatial_attention(x, 32, 'spatial_attention2')\n",
        "    x = tf.keras.layers.add([x_channel, x_spatial])\n",
        "\n",
        "    x = tf.keras.layers.MaxPooling2D((2, 2), name='encoder_pool2')(x)\n",
        "    \n",
        "    encoded_img = autoencoder_block(x, 64, 'encoder3')\n",
        "\n",
        "    # Decoder\n",
        "    x = autoencoder_block(encoded_img, 64, 'decoder1')\n",
        "\n",
        "    x_channel = channel_attention(x, 64, 'channel_attention3')\n",
        "    x_spatial = spatial_attention(x, 64, 'spatial_attention3')\n",
        "    x = tf.keras.layers.add([x_channel, x_spatial])\n",
        "\n",
        "    x = tf.keras.layers.UpSampling2D((2, 2), name='decoder_up1')(x)\n",
        "    x = autoencoder_block(x, 32, 'decoder2')\n",
        "\n",
        "    x_channel = channel_attention(x, 32, 'channel_attention4')\n",
        "    x_spatial = spatial_attention(x, 32, 'spatial_attention4')\n",
        "    x = tf.keras.layers.add([x_channel, x_spatial])\n",
        "\n",
        "    x = tf.keras.layers.UpSampling2D((2, 2), name='decoder_up2')(x)\n",
        "    x = autoencoder_block(x, 16, 'decoder3')\n",
        "\n",
        "    decoded_img = tf.keras.layers.Conv2D(filters=1, kernel_size=3, strides=(1, 1), \n",
        "                                         activation='sigmoid', padding='same', \n",
        "                                         name='last_conv')(x)\n",
        "    return decoded_img   "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Hje1afVfbvx0",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 3977
        },
        "outputId": "f35e98b2-5641-4e78-c9d0-2e72fa9055e6"
      },
      "source": [
        "# Normal AutoEncoder Model Create / Summary\n",
        "input_autoencoder_parallel_add_data = tf.keras.Input(shape=(28, 28, 1))\n",
        "\n",
        "output_autoencoder_parallel_add_data = autoencoder_parallel_add(input_autoencoder_parallel_add_data)\n",
        "\n",
        "model_autoencoder_parallel_add  = tf.keras.Model(inputs=input_autoencoder_parallel_add_data, outputs=output_autoencoder_parallel_add_data)\n",
        "\n",
        "model_autoencoder_parallel_add.summary()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"model_3\"\n",
            "__________________________________________________________________________________________________\n",
            "Layer (type)                    Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            "input_3 (InputLayer)            [(None, 28, 28, 1)]  0                                            \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_conv (Conv2D)          (None, 28, 28, 16)   160         input_3[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_bn (BatchNormalization (None, 28, 28, 16)   64          encoder1_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_relu (ReLU)            (None, 28, 28, 16)   0           encoder1_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_16 (Conv2D)              (None, 28, 28, 4)    68          encoder1_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_17 (BatchNo (None, 28, 28, 4)    16          conv2d_16[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_17 (ReLU)                 (None, 28, 28, 4)    0           batch_normalization_17[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_17 (Conv2D)              (None, 28, 28, 4)    148         re_lu_17[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_4 (Glo (None, 16)           0           encoder1_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_18 (BatchNo (None, 28, 28, 4)    16          conv2d_17[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "dense_8 (Dense)                 (None, 4)            68          global_average_pooling2d_4[0][0] \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_18 (ReLU)                 (None, 28, 28, 4)    0           batch_normalization_18[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_16 (BatchNo (None, 4)            16          dense_8[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_18 (Conv2D)              (None, 28, 28, 4)    148         re_lu_18[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_16 (ReLU)                 (None, 4)            0           batch_normalization_16[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_19 (BatchNo (None, 28, 28, 4)    16          conv2d_18[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "dense_9 (Dense)                 (None, 16)           80          re_lu_16[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_19 (ReLU)                 (None, 28, 28, 4)    0           batch_normalization_19[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_4 (Te [(None, 1, 1, 16)]   0           dense_9[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_19 (Conv2D)              (None, 28, 28, 1)    5           re_lu_19[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_8 (TensorFlowOp [(None, 28, 28, 16)] 0           encoder1_relu[0][0]              \n",
            "                                                                 tf_op_layer_strided_slice_4[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_9 (TensorFlowOp [(None, 28, 28, 16)] 0           encoder1_relu[0][0]              \n",
            "                                                                 conv2d_19[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "add (Add)                       (None, 28, 28, 16)   0           tf_op_layer_Mul_8[0][0]          \n",
            "                                                                 tf_op_layer_Mul_9[0][0]          \n",
            "__________________________________________________________________________________________________\n",
            "encoder_pool1 (MaxPooling2D)    (None, 14, 14, 16)   0           add[0][0]                        \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_conv (Conv2D)          (None, 14, 14, 32)   4640        encoder_pool1[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_bn (BatchNormalization (None, 14, 14, 32)   128         encoder2_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_relu (ReLU)            (None, 14, 14, 32)   0           encoder2_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_20 (Conv2D)              (None, 14, 14, 8)    264         encoder2_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_21 (BatchNo (None, 14, 14, 8)    32          conv2d_20[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_21 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_21[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_21 (Conv2D)              (None, 14, 14, 8)    584         re_lu_21[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_5 (Glo (None, 32)           0           encoder2_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_22 (BatchNo (None, 14, 14, 8)    32          conv2d_21[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "dense_10 (Dense)                (None, 8)            264         global_average_pooling2d_5[0][0] \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_22 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_22[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_20 (BatchNo (None, 8)            32          dense_10[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_22 (Conv2D)              (None, 14, 14, 8)    584         re_lu_22[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_20 (ReLU)                 (None, 8)            0           batch_normalization_20[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_23 (BatchNo (None, 14, 14, 8)    32          conv2d_22[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "dense_11 (Dense)                (None, 32)           288         re_lu_20[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_23 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_23[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_5 (Te [(None, 1, 1, 32)]   0           dense_11[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_23 (Conv2D)              (None, 14, 14, 1)    9           re_lu_23[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_10 (TensorFlowO [(None, 14, 14, 32)] 0           encoder2_relu[0][0]              \n",
            "                                                                 tf_op_layer_strided_slice_5[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_11 (TensorFlowO [(None, 14, 14, 32)] 0           encoder2_relu[0][0]              \n",
            "                                                                 conv2d_23[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "add_1 (Add)                     (None, 14, 14, 32)   0           tf_op_layer_Mul_10[0][0]         \n",
            "                                                                 tf_op_layer_Mul_11[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "encoder_pool2 (MaxPooling2D)    (None, 7, 7, 32)     0           add_1[0][0]                      \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_conv (Conv2D)          (None, 7, 7, 64)     18496       encoder_pool2[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_bn (BatchNormalization (None, 7, 7, 64)     256         encoder3_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_relu (ReLU)            (None, 7, 7, 64)     0           encoder3_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_conv (Conv2D)          (None, 7, 7, 64)     36928       encoder3_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_bn (BatchNormalization (None, 7, 7, 64)     256         decoder1_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_relu (ReLU)            (None, 7, 7, 64)     0           decoder1_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_24 (Conv2D)              (None, 7, 7, 16)     1040        decoder1_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_25 (BatchNo (None, 7, 7, 16)     64          conv2d_24[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_25 (ReLU)                 (None, 7, 7, 16)     0           batch_normalization_25[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_25 (Conv2D)              (None, 7, 7, 16)     2320        re_lu_25[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_6 (Glo (None, 64)           0           decoder1_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_26 (BatchNo (None, 7, 7, 16)     64          conv2d_25[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "dense_12 (Dense)                (None, 16)           1040        global_average_pooling2d_6[0][0] \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_26 (ReLU)                 (None, 7, 7, 16)     0           batch_normalization_26[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_24 (BatchNo (None, 16)           64          dense_12[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_26 (Conv2D)              (None, 7, 7, 16)     2320        re_lu_26[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_24 (ReLU)                 (None, 16)           0           batch_normalization_24[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_27 (BatchNo (None, 7, 7, 16)     64          conv2d_26[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "dense_13 (Dense)                (None, 64)           1088        re_lu_24[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_27 (ReLU)                 (None, 7, 7, 16)     0           batch_normalization_27[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_6 (Te [(None, 1, 1, 64)]   0           dense_13[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_27 (Conv2D)              (None, 7, 7, 1)      17          re_lu_27[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_12 (TensorFlowO [(None, 7, 7, 64)]   0           decoder1_relu[0][0]              \n",
            "                                                                 tf_op_layer_strided_slice_6[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_13 (TensorFlowO [(None, 7, 7, 64)]   0           decoder1_relu[0][0]              \n",
            "                                                                 conv2d_27[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "add_2 (Add)                     (None, 7, 7, 64)     0           tf_op_layer_Mul_12[0][0]         \n",
            "                                                                 tf_op_layer_Mul_13[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "decoder_up1 (UpSampling2D)      (None, 14, 14, 64)   0           add_2[0][0]                      \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_conv (Conv2D)          (None, 14, 14, 32)   18464       decoder_up1[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_bn (BatchNormalization (None, 14, 14, 32)   128         decoder2_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_relu (ReLU)            (None, 14, 14, 32)   0           decoder2_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_28 (Conv2D)              (None, 14, 14, 8)    264         decoder2_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_29 (BatchNo (None, 14, 14, 8)    32          conv2d_28[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_29 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_29[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_29 (Conv2D)              (None, 14, 14, 8)    584         re_lu_29[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_7 (Glo (None, 32)           0           decoder2_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_30 (BatchNo (None, 14, 14, 8)    32          conv2d_29[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "dense_14 (Dense)                (None, 8)            264         global_average_pooling2d_7[0][0] \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_30 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_30[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_28 (BatchNo (None, 8)            32          dense_14[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_30 (Conv2D)              (None, 14, 14, 8)    584         re_lu_30[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_28 (ReLU)                 (None, 8)            0           batch_normalization_28[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_31 (BatchNo (None, 14, 14, 8)    32          conv2d_30[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "dense_15 (Dense)                (None, 32)           288         re_lu_28[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_31 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_31[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_7 (Te [(None, 1, 1, 32)]   0           dense_15[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_31 (Conv2D)              (None, 14, 14, 1)    9           re_lu_31[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_14 (TensorFlowO [(None, 14, 14, 32)] 0           decoder2_relu[0][0]              \n",
            "                                                                 tf_op_layer_strided_slice_7[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_15 (TensorFlowO [(None, 14, 14, 32)] 0           decoder2_relu[0][0]              \n",
            "                                                                 conv2d_31[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "add_3 (Add)                     (None, 14, 14, 32)   0           tf_op_layer_Mul_14[0][0]         \n",
            "                                                                 tf_op_layer_Mul_15[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "decoder_up2 (UpSampling2D)      (None, 28, 28, 32)   0           add_3[0][0]                      \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_conv (Conv2D)          (None, 28, 28, 16)   4624        decoder_up2[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_bn (BatchNormalization (None, 28, 28, 16)   64          decoder3_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_relu (ReLU)            (None, 28, 28, 16)   0           decoder3_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "last_conv (Conv2D)              (None, 28, 28, 1)    145         decoder3_relu[0][0]              \n",
            "==================================================================================================\n",
            "Total params: 97,257\n",
            "Trainable params: 96,521\n",
            "Non-trainable params: 736\n",
            "__________________________________________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "z0Tww0BscHW6",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Model Compile\n",
        "model_autoencoder_parallel_add.compile(optimizer=tf.keras.optimizers.Adam(), loss=tf.keras.losses.MeanSquaredError(), metrics=['mse'])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lV3kNbHScQ6E",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 935
        },
        "outputId": "9c581c52-909d-4a41-9968-55e24ac4c93d"
      },
      "source": [
        "# Training\n",
        "print('Training Model - autoencoder_parallel_add')\n",
        "epoch = 25\n",
        "# AutoEncoder는 input과 output이 같으므로 x_train을 넣어줌\n",
        "history_autoencoder_parallel_add = model_autoencoder_parallel_add.fit(x_train, x_train, \n",
        "                    batch_size=16, \n",
        "                    epochs=epoch,\n",
        "                    shuffle=True,\n",
        "                    validation_data=(x_test, x_test))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Training Model - autoencoder_parallel_add\n",
            "Epoch 1/25\n",
            "3750/3750 [==============================] - 39s 10ms/step - loss: 0.0061 - mse: 0.0061 - val_loss: 0.0036 - val_mse: 0.0036\n",
            "Epoch 2/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0032 - mse: 0.0032 - val_loss: 0.0030 - val_mse: 0.0030\n",
            "Epoch 3/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0026 - mse: 0.0026 - val_loss: 0.0025 - val_mse: 0.0025\n",
            "Epoch 4/25\n",
            "3750/3750 [==============================] - 39s 10ms/step - loss: 0.0024 - mse: 0.0024 - val_loss: 0.0022 - val_mse: 0.0022\n",
            "Epoch 5/25\n",
            "3750/3750 [==============================] - 39s 10ms/step - loss: 0.0022 - mse: 0.0022 - val_loss: 0.0021 - val_mse: 0.0021\n",
            "Epoch 6/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0021 - mse: 0.0021 - val_loss: 0.0021 - val_mse: 0.0021\n",
            "Epoch 7/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0020 - mse: 0.0020 - val_loss: 0.0020 - val_mse: 0.0020\n",
            "Epoch 8/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0019 - mse: 0.0019 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 9/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 10/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0021 - val_mse: 0.0021\n",
            "Epoch 11/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 12/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 13/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 14/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 15/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 16/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0015 - val_mse: 0.0015\n",
            "Epoch 17/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 18/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 19/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0015 - val_mse: 0.0015\n",
            "Epoch 20/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0015 - val_mse: 0.0015\n",
            "Epoch 21/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0015 - val_mse: 0.0015\n",
            "Epoch 22/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 23/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 24/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0015 - val_mse: 0.0015\n",
            "Epoch 25/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0014 - val_mse: 0.0014\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PvpzNeV0ccYI",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def autoencoder_parallel_mul(input):\n",
        "    # Encoder\n",
        "    x = autoencoder_block(input, 16, 'encoder1')\n",
        "    \n",
        "    x_channel = channel_attention(x, 16, 'channel_attention')\n",
        "    x_spatial = spatial_attention(x, 16, 'spatial_attention')\n",
        "    x = tf.keras.layers.Multiply()([x_channel, x_spatial])\n",
        "    \n",
        "    x = tf.keras.layers.MaxPooling2D((2, 2), name='encoder_pool1')(x)\n",
        "    x = autoencoder_block(x, 32, 'encoder2')\n",
        "\n",
        "    x_channel = channel_attention(x, 32, 'channel_attention2')\n",
        "    x_spatial = spatial_attention(x, 32, 'spatial_attention2')\n",
        "    x = tf.keras.layers.Multiply()([x_channel, x_spatial])\n",
        "\n",
        "    x = tf.keras.layers.MaxPooling2D((2, 2), name='encoder_pool2')(x)\n",
        "    \n",
        "    encoded_img = autoencoder_block(x, 64, 'encoder3')\n",
        "\n",
        "    # Decoder\n",
        "    x = autoencoder_block(encoded_img, 64, 'decoder1')\n",
        "\n",
        "    x_channel = channel_attention(x, 64, 'channel_attention3')\n",
        "    x_spatial = spatial_attention(x, 64, 'spatial_attention3')\n",
        "    x = tf.keras.layers.Multiply()([x_channel, x_spatial])\n",
        "\n",
        "    x = tf.keras.layers.UpSampling2D((2, 2), name='decoder_up1')(x)\n",
        "    x = autoencoder_block(x, 32, 'decoder2')\n",
        "\n",
        "    x_channel = channel_attention(x, 32, 'channel_attention4')\n",
        "    x_spatial = spatial_attention(x, 32, 'spatial_attention4')\n",
        "    x = tf.keras.layers.Multiply()([x_channel, x_spatial])\n",
        "\n",
        "    x = tf.keras.layers.UpSampling2D((2, 2), name='decoder_up2')(x)\n",
        "    x = autoencoder_block(x, 16, 'decoder3')\n",
        "\n",
        "    decoded_img = tf.keras.layers.Conv2D(filters=1, kernel_size=3, strides=(1, 1), \n",
        "                                         activation='sigmoid', padding='same', \n",
        "                                         name='last_conv')(x)\n",
        "    return decoded_img   "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1MzpCtC3cuPp",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 3977
        },
        "outputId": "e7806d92-7ab0-422e-d057-0cc162459ba8"
      },
      "source": [
        "# Normal AutoEncoder Model Create / Summary\n",
        "input_autoencoder_parallel_mul_data = tf.keras.Input(shape=(28, 28, 1))\n",
        "\n",
        "output_autoencoder_parallel_mul_data = autoencoder_parallel_mul(input_autoencoder_parallel_mul_data)\n",
        "\n",
        "model_autoencoder_parallel_mul  = tf.keras.Model(inputs=input_autoencoder_parallel_mul_data, outputs=output_autoencoder_parallel_mul_data)\n",
        "\n",
        "model_autoencoder_parallel_mul.summary()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"model_4\"\n",
            "__________________________________________________________________________________________________\n",
            "Layer (type)                    Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            "input_4 (InputLayer)            [(None, 28, 28, 1)]  0                                            \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_conv (Conv2D)          (None, 28, 28, 16)   160         input_4[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_bn (BatchNormalization (None, 28, 28, 16)   64          encoder1_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_relu (ReLU)            (None, 28, 28, 16)   0           encoder1_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_32 (Conv2D)              (None, 28, 28, 4)    68          encoder1_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_33 (BatchNo (None, 28, 28, 4)    16          conv2d_32[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_33 (ReLU)                 (None, 28, 28, 4)    0           batch_normalization_33[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_33 (Conv2D)              (None, 28, 28, 4)    148         re_lu_33[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_8 (Glo (None, 16)           0           encoder1_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_34 (BatchNo (None, 28, 28, 4)    16          conv2d_33[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "dense_16 (Dense)                (None, 4)            68          global_average_pooling2d_8[0][0] \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_34 (ReLU)                 (None, 28, 28, 4)    0           batch_normalization_34[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_32 (BatchNo (None, 4)            16          dense_16[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_34 (Conv2D)              (None, 28, 28, 4)    148         re_lu_34[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_32 (ReLU)                 (None, 4)            0           batch_normalization_32[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_35 (BatchNo (None, 28, 28, 4)    16          conv2d_34[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "dense_17 (Dense)                (None, 16)           80          re_lu_32[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_35 (ReLU)                 (None, 28, 28, 4)    0           batch_normalization_35[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_8 (Te [(None, 1, 1, 16)]   0           dense_17[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_35 (Conv2D)              (None, 28, 28, 1)    5           re_lu_35[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_16 (TensorFlowO [(None, 28, 28, 16)] 0           encoder1_relu[0][0]              \n",
            "                                                                 tf_op_layer_strided_slice_8[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_17 (TensorFlowO [(None, 28, 28, 16)] 0           encoder1_relu[0][0]              \n",
            "                                                                 conv2d_35[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "multiply (Multiply)             (None, 28, 28, 16)   0           tf_op_layer_Mul_16[0][0]         \n",
            "                                                                 tf_op_layer_Mul_17[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "encoder_pool1 (MaxPooling2D)    (None, 14, 14, 16)   0           multiply[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_conv (Conv2D)          (None, 14, 14, 32)   4640        encoder_pool1[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_bn (BatchNormalization (None, 14, 14, 32)   128         encoder2_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_relu (ReLU)            (None, 14, 14, 32)   0           encoder2_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_36 (Conv2D)              (None, 14, 14, 8)    264         encoder2_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_37 (BatchNo (None, 14, 14, 8)    32          conv2d_36[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_37 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_37[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_37 (Conv2D)              (None, 14, 14, 8)    584         re_lu_37[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_9 (Glo (None, 32)           0           encoder2_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_38 (BatchNo (None, 14, 14, 8)    32          conv2d_37[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "dense_18 (Dense)                (None, 8)            264         global_average_pooling2d_9[0][0] \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_38 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_38[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_36 (BatchNo (None, 8)            32          dense_18[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_38 (Conv2D)              (None, 14, 14, 8)    584         re_lu_38[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_36 (ReLU)                 (None, 8)            0           batch_normalization_36[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_39 (BatchNo (None, 14, 14, 8)    32          conv2d_38[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "dense_19 (Dense)                (None, 32)           288         re_lu_36[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_39 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_39[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_9 (Te [(None, 1, 1, 32)]   0           dense_19[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_39 (Conv2D)              (None, 14, 14, 1)    9           re_lu_39[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_18 (TensorFlowO [(None, 14, 14, 32)] 0           encoder2_relu[0][0]              \n",
            "                                                                 tf_op_layer_strided_slice_9[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_19 (TensorFlowO [(None, 14, 14, 32)] 0           encoder2_relu[0][0]              \n",
            "                                                                 conv2d_39[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "multiply_1 (Multiply)           (None, 14, 14, 32)   0           tf_op_layer_Mul_18[0][0]         \n",
            "                                                                 tf_op_layer_Mul_19[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "encoder_pool2 (MaxPooling2D)    (None, 7, 7, 32)     0           multiply_1[0][0]                 \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_conv (Conv2D)          (None, 7, 7, 64)     18496       encoder_pool2[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_bn (BatchNormalization (None, 7, 7, 64)     256         encoder3_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_relu (ReLU)            (None, 7, 7, 64)     0           encoder3_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_conv (Conv2D)          (None, 7, 7, 64)     36928       encoder3_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_bn (BatchNormalization (None, 7, 7, 64)     256         decoder1_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_relu (ReLU)            (None, 7, 7, 64)     0           decoder1_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_40 (Conv2D)              (None, 7, 7, 16)     1040        decoder1_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_41 (BatchNo (None, 7, 7, 16)     64          conv2d_40[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_41 (ReLU)                 (None, 7, 7, 16)     0           batch_normalization_41[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_41 (Conv2D)              (None, 7, 7, 16)     2320        re_lu_41[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_10 (Gl (None, 64)           0           decoder1_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_42 (BatchNo (None, 7, 7, 16)     64          conv2d_41[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "dense_20 (Dense)                (None, 16)           1040        global_average_pooling2d_10[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "re_lu_42 (ReLU)                 (None, 7, 7, 16)     0           batch_normalization_42[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_40 (BatchNo (None, 16)           64          dense_20[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_42 (Conv2D)              (None, 7, 7, 16)     2320        re_lu_42[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_40 (ReLU)                 (None, 16)           0           batch_normalization_40[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_43 (BatchNo (None, 7, 7, 16)     64          conv2d_42[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "dense_21 (Dense)                (None, 64)           1088        re_lu_40[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_43 (ReLU)                 (None, 7, 7, 16)     0           batch_normalization_43[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_10 (T [(None, 1, 1, 64)]   0           dense_21[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_43 (Conv2D)              (None, 7, 7, 1)      17          re_lu_43[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_20 (TensorFlowO [(None, 7, 7, 64)]   0           decoder1_relu[0][0]              \n",
            "                                                                 tf_op_layer_strided_slice_10[0][0\n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_21 (TensorFlowO [(None, 7, 7, 64)]   0           decoder1_relu[0][0]              \n",
            "                                                                 conv2d_43[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "multiply_2 (Multiply)           (None, 7, 7, 64)     0           tf_op_layer_Mul_20[0][0]         \n",
            "                                                                 tf_op_layer_Mul_21[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "decoder_up1 (UpSampling2D)      (None, 14, 14, 64)   0           multiply_2[0][0]                 \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_conv (Conv2D)          (None, 14, 14, 32)   18464       decoder_up1[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_bn (BatchNormalization (None, 14, 14, 32)   128         decoder2_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_relu (ReLU)            (None, 14, 14, 32)   0           decoder2_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_44 (Conv2D)              (None, 14, 14, 8)    264         decoder2_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_45 (BatchNo (None, 14, 14, 8)    32          conv2d_44[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_45 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_45[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_45 (Conv2D)              (None, 14, 14, 8)    584         re_lu_45[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_11 (Gl (None, 32)           0           decoder2_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_46 (BatchNo (None, 14, 14, 8)    32          conv2d_45[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "dense_22 (Dense)                (None, 8)            264         global_average_pooling2d_11[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "re_lu_46 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_46[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_44 (BatchNo (None, 8)            32          dense_22[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_46 (Conv2D)              (None, 14, 14, 8)    584         re_lu_46[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_44 (ReLU)                 (None, 8)            0           batch_normalization_44[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_47 (BatchNo (None, 14, 14, 8)    32          conv2d_46[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "dense_23 (Dense)                (None, 32)           288         re_lu_44[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_47 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_47[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_11 (T [(None, 1, 1, 32)]   0           dense_23[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_47 (Conv2D)              (None, 14, 14, 1)    9           re_lu_47[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_22 (TensorFlowO [(None, 14, 14, 32)] 0           decoder2_relu[0][0]              \n",
            "                                                                 tf_op_layer_strided_slice_11[0][0\n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_23 (TensorFlowO [(None, 14, 14, 32)] 0           decoder2_relu[0][0]              \n",
            "                                                                 conv2d_47[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "multiply_3 (Multiply)           (None, 14, 14, 32)   0           tf_op_layer_Mul_22[0][0]         \n",
            "                                                                 tf_op_layer_Mul_23[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "decoder_up2 (UpSampling2D)      (None, 28, 28, 32)   0           multiply_3[0][0]                 \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_conv (Conv2D)          (None, 28, 28, 16)   4624        decoder_up2[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_bn (BatchNormalization (None, 28, 28, 16)   64          decoder3_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_relu (ReLU)            (None, 28, 28, 16)   0           decoder3_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "last_conv (Conv2D)              (None, 28, 28, 1)    145         decoder3_relu[0][0]              \n",
            "==================================================================================================\n",
            "Total params: 97,257\n",
            "Trainable params: 96,521\n",
            "Non-trainable params: 736\n",
            "__________________________________________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "i9eWimmKc0tP",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Model Compile\n",
        "model_autoencoder_parallel_mul.compile(optimizer=tf.keras.optimizers.Adam(), loss=tf.keras.losses.MeanSquaredError(), metrics=['mse'])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4hma3_zsc6Nt",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 935
        },
        "outputId": "dc741346-5796-4ea4-9542-0467a9ce3229"
      },
      "source": [
        "# Training\n",
        "print('Training Model - autoencoder_parallel_mul')\n",
        "epoch = 25\n",
        "# AutoEncoder는 input과 output이 같으므로 x_train을 넣어줌\n",
        "history_autoencoder_parallel_mul = model_autoencoder_parallel_mul.fit(x_train, x_train, \n",
        "                    batch_size=16, \n",
        "                    epochs=epoch,\n",
        "                    shuffle=True,\n",
        "                    validation_data=(x_test, x_test))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Training Model - autoencoder_parallel_mul\n",
            "Epoch 1/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0088 - mse: 0.0088 - val_loss: 0.0042 - val_mse: 0.0042\n",
            "Epoch 2/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0037 - mse: 0.0037 - val_loss: 0.0032 - val_mse: 0.0032\n",
            "Epoch 3/25\n",
            "3750/3750 [==============================] - 39s 10ms/step - loss: 0.0029 - mse: 0.0029 - val_loss: 0.0028 - val_mse: 0.0028\n",
            "Epoch 4/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0025 - mse: 0.0025 - val_loss: 0.0023 - val_mse: 0.0023\n",
            "Epoch 5/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0023 - mse: 0.0023 - val_loss: 0.0026 - val_mse: 0.0026\n",
            "Epoch 6/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0022 - mse: 0.0022 - val_loss: 0.0020 - val_mse: 0.0020\n",
            "Epoch 7/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0021 - mse: 0.0021 - val_loss: 0.0022 - val_mse: 0.0022\n",
            "Epoch 8/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0020 - mse: 0.0020 - val_loss: 0.0023 - val_mse: 0.0023\n",
            "Epoch 9/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0020 - mse: 0.0020 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 10/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0019 - mse: 0.0019 - val_loss: 0.0023 - val_mse: 0.0023\n",
            "Epoch 11/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0024 - val_mse: 0.0024\n",
            "Epoch 12/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 13/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0020 - val_mse: 0.0020\n",
            "Epoch 14/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 15/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 16/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 17/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 18/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 19/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 20/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0022 - val_mse: 0.0022\n",
            "Epoch 21/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 22/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 23/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 24/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 25/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0015 - val_mse: 0.0015\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "L_61IEPainch",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def autoencoder_serial_channel_first(input):\n",
        "    # Encoder\n",
        "    x = autoencoder_block(input, 16, 'encoder1')\n",
        "    \n",
        "    x = channel_attention(x, 16, 'channel_attention')\n",
        "    x = spatial_attention(x, 16, 'spatial_attention')\n",
        "    \n",
        "    x = tf.keras.layers.MaxPooling2D((2, 2), name='encoder_pool1')(x)\n",
        "    x = autoencoder_block(x, 32, 'encoder2')\n",
        "\n",
        "    x = channel_attention(x, 32, 'channel_attention2')\n",
        "    x = spatial_attention(x, 32, 'spatial_attention2')\n",
        "\n",
        "    x = tf.keras.layers.MaxPooling2D((2, 2), name='encoder_pool2')(x)\n",
        "    \n",
        "    encoded_img = autoencoder_block(x, 64, 'encoder3')\n",
        "\n",
        "    # Decoder\n",
        "    x = autoencoder_block(encoded_img, 64, 'decoder1')\n",
        "\n",
        "    x = channel_attention(x, 64, 'channel_attention3')\n",
        "    x = spatial_attention(x, 64, 'spatial_attention3')\n",
        "\n",
        "    x = tf.keras.layers.UpSampling2D((2, 2), name='decoder_up1')(x)\n",
        "    x = autoencoder_block(x, 32, 'decoder2')\n",
        "\n",
        "    x = channel_attention(x, 32, 'channel_attention4')\n",
        "    x = spatial_attention(x, 32, 'spatial_attention4')\n",
        "\n",
        "    x = tf.keras.layers.UpSampling2D((2, 2), name='decoder_up2')(x)\n",
        "    x = autoencoder_block(x, 16, 'decoder3')\n",
        "\n",
        "    decoded_img = tf.keras.layers.Conv2D(filters=1, kernel_size=3, strides=(1, 1), \n",
        "                                         activation='sigmoid', padding='same', \n",
        "                                         name='last_conv')(x)\n",
        "    return decoded_img   "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wHgw1SJgjKK4",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 3761
        },
        "outputId": "f9676646-31bf-4e2c-b6ac-8ebcb546dd89"
      },
      "source": [
        "# Normal AutoEncoder Model Create / Summary\n",
        "input_autoencoder_serial_channel_first_data = tf.keras.Input(shape=(28, 28, 1))\n",
        "\n",
        "output_autoencoder_serial_channel_first_data = autoencoder_serial_channel_first(input_autoencoder_serial_channel_first_data)\n",
        "\n",
        "model_autoencoder_serial_channel_first  = tf.keras.Model(inputs=input_autoencoder_serial_channel_first_data, outputs=output_autoencoder_serial_channel_first_data)\n",
        "\n",
        "model_autoencoder_serial_channel_first.summary()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"model_5\"\n",
            "__________________________________________________________________________________________________\n",
            "Layer (type)                    Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            "input_5 (InputLayer)            [(None, 28, 28, 1)]  0                                            \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_conv (Conv2D)          (None, 28, 28, 16)   160         input_5[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_bn (BatchNormalization (None, 28, 28, 16)   64          encoder1_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_relu (ReLU)            (None, 28, 28, 16)   0           encoder1_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_12 (Gl (None, 16)           0           encoder1_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "dense_24 (Dense)                (None, 4)            68          global_average_pooling2d_12[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_48 (BatchNo (None, 4)            16          dense_24[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_48 (ReLU)                 (None, 4)            0           batch_normalization_48[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "dense_25 (Dense)                (None, 16)           80          re_lu_48[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_12 (T [(None, 1, 1, 16)]   0           dense_25[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_24 (TensorFlowO [(None, 28, 28, 16)] 0           encoder1_relu[0][0]              \n",
            "                                                                 tf_op_layer_strided_slice_12[0][0\n",
            "__________________________________________________________________________________________________\n",
            "conv2d_48 (Conv2D)              (None, 28, 28, 4)    68          tf_op_layer_Mul_24[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_49 (BatchNo (None, 28, 28, 4)    16          conv2d_48[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_49 (ReLU)                 (None, 28, 28, 4)    0           batch_normalization_49[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_49 (Conv2D)              (None, 28, 28, 4)    148         re_lu_49[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_50 (BatchNo (None, 28, 28, 4)    16          conv2d_49[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_50 (ReLU)                 (None, 28, 28, 4)    0           batch_normalization_50[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_50 (Conv2D)              (None, 28, 28, 4)    148         re_lu_50[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_51 (BatchNo (None, 28, 28, 4)    16          conv2d_50[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_51 (ReLU)                 (None, 28, 28, 4)    0           batch_normalization_51[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_51 (Conv2D)              (None, 28, 28, 1)    5           re_lu_51[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_25 (TensorFlowO [(None, 28, 28, 16)] 0           tf_op_layer_Mul_24[0][0]         \n",
            "                                                                 conv2d_51[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "encoder_pool1 (MaxPooling2D)    (None, 14, 14, 16)   0           tf_op_layer_Mul_25[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_conv (Conv2D)          (None, 14, 14, 32)   4640        encoder_pool1[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_bn (BatchNormalization (None, 14, 14, 32)   128         encoder2_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_relu (ReLU)            (None, 14, 14, 32)   0           encoder2_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_13 (Gl (None, 32)           0           encoder2_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "dense_26 (Dense)                (None, 8)            264         global_average_pooling2d_13[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_52 (BatchNo (None, 8)            32          dense_26[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_52 (ReLU)                 (None, 8)            0           batch_normalization_52[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "dense_27 (Dense)                (None, 32)           288         re_lu_52[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_13 (T [(None, 1, 1, 32)]   0           dense_27[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_26 (TensorFlowO [(None, 14, 14, 32)] 0           encoder2_relu[0][0]              \n",
            "                                                                 tf_op_layer_strided_slice_13[0][0\n",
            "__________________________________________________________________________________________________\n",
            "conv2d_52 (Conv2D)              (None, 14, 14, 8)    264         tf_op_layer_Mul_26[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_53 (BatchNo (None, 14, 14, 8)    32          conv2d_52[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_53 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_53[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_53 (Conv2D)              (None, 14, 14, 8)    584         re_lu_53[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_54 (BatchNo (None, 14, 14, 8)    32          conv2d_53[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_54 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_54[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_54 (Conv2D)              (None, 14, 14, 8)    584         re_lu_54[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_55 (BatchNo (None, 14, 14, 8)    32          conv2d_54[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_55 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_55[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_55 (Conv2D)              (None, 14, 14, 1)    9           re_lu_55[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_27 (TensorFlowO [(None, 14, 14, 32)] 0           tf_op_layer_Mul_26[0][0]         \n",
            "                                                                 conv2d_55[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "encoder_pool2 (MaxPooling2D)    (None, 7, 7, 32)     0           tf_op_layer_Mul_27[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_conv (Conv2D)          (None, 7, 7, 64)     18496       encoder_pool2[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_bn (BatchNormalization (None, 7, 7, 64)     256         encoder3_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_relu (ReLU)            (None, 7, 7, 64)     0           encoder3_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_conv (Conv2D)          (None, 7, 7, 64)     36928       encoder3_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_bn (BatchNormalization (None, 7, 7, 64)     256         decoder1_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_relu (ReLU)            (None, 7, 7, 64)     0           decoder1_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_14 (Gl (None, 64)           0           decoder1_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "dense_28 (Dense)                (None, 16)           1040        global_average_pooling2d_14[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_56 (BatchNo (None, 16)           64          dense_28[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_56 (ReLU)                 (None, 16)           0           batch_normalization_56[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "dense_29 (Dense)                (None, 64)           1088        re_lu_56[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_14 (T [(None, 1, 1, 64)]   0           dense_29[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_28 (TensorFlowO [(None, 7, 7, 64)]   0           decoder1_relu[0][0]              \n",
            "                                                                 tf_op_layer_strided_slice_14[0][0\n",
            "__________________________________________________________________________________________________\n",
            "conv2d_56 (Conv2D)              (None, 7, 7, 16)     1040        tf_op_layer_Mul_28[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_57 (BatchNo (None, 7, 7, 16)     64          conv2d_56[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_57 (ReLU)                 (None, 7, 7, 16)     0           batch_normalization_57[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_57 (Conv2D)              (None, 7, 7, 16)     2320        re_lu_57[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_58 (BatchNo (None, 7, 7, 16)     64          conv2d_57[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_58 (ReLU)                 (None, 7, 7, 16)     0           batch_normalization_58[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_58 (Conv2D)              (None, 7, 7, 16)     2320        re_lu_58[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_59 (BatchNo (None, 7, 7, 16)     64          conv2d_58[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_59 (ReLU)                 (None, 7, 7, 16)     0           batch_normalization_59[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_59 (Conv2D)              (None, 7, 7, 1)      17          re_lu_59[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_29 (TensorFlowO [(None, 7, 7, 64)]   0           tf_op_layer_Mul_28[0][0]         \n",
            "                                                                 conv2d_59[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "decoder_up1 (UpSampling2D)      (None, 14, 14, 64)   0           tf_op_layer_Mul_29[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_conv (Conv2D)          (None, 14, 14, 32)   18464       decoder_up1[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_bn (BatchNormalization (None, 14, 14, 32)   128         decoder2_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_relu (ReLU)            (None, 14, 14, 32)   0           decoder2_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_15 (Gl (None, 32)           0           decoder2_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "dense_30 (Dense)                (None, 8)            264         global_average_pooling2d_15[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_60 (BatchNo (None, 8)            32          dense_30[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_60 (ReLU)                 (None, 8)            0           batch_normalization_60[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "dense_31 (Dense)                (None, 32)           288         re_lu_60[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_15 (T [(None, 1, 1, 32)]   0           dense_31[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_30 (TensorFlowO [(None, 14, 14, 32)] 0           decoder2_relu[0][0]              \n",
            "                                                                 tf_op_layer_strided_slice_15[0][0\n",
            "__________________________________________________________________________________________________\n",
            "conv2d_60 (Conv2D)              (None, 14, 14, 8)    264         tf_op_layer_Mul_30[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_61 (BatchNo (None, 14, 14, 8)    32          conv2d_60[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_61 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_61[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_61 (Conv2D)              (None, 14, 14, 8)    584         re_lu_61[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_62 (BatchNo (None, 14, 14, 8)    32          conv2d_61[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_62 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_62[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_62 (Conv2D)              (None, 14, 14, 8)    584         re_lu_62[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_63 (BatchNo (None, 14, 14, 8)    32          conv2d_62[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_63 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_63[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_63 (Conv2D)              (None, 14, 14, 1)    9           re_lu_63[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_31 (TensorFlowO [(None, 14, 14, 32)] 0           tf_op_layer_Mul_30[0][0]         \n",
            "                                                                 conv2d_63[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "decoder_up2 (UpSampling2D)      (None, 28, 28, 32)   0           tf_op_layer_Mul_31[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_conv (Conv2D)          (None, 28, 28, 16)   4624        decoder_up2[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_bn (BatchNormalization (None, 28, 28, 16)   64          decoder3_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_relu (ReLU)            (None, 28, 28, 16)   0           decoder3_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "last_conv (Conv2D)              (None, 28, 28, 1)    145         decoder3_relu[0][0]              \n",
            "==================================================================================================\n",
            "Total params: 97,257\n",
            "Trainable params: 96,521\n",
            "Non-trainable params: 736\n",
            "__________________________________________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uaneafhQjPMF",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Model Compile\n",
        "model_autoencoder_serial_channel_first.compile(optimizer=tf.keras.optimizers.Adam(), loss=tf.keras.losses.MeanSquaredError(), metrics=['mse'])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sqkd0RBfjWrk",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 935
        },
        "outputId": "fd694eba-f612-4deb-eeb2-afab37c62b3b"
      },
      "source": [
        "# Training\n",
        "print('Training Model - autoencoder_serial_channel_first')\n",
        "epoch = 25\n",
        "# AutoEncoder는 input과 output이 같으므로 x_train을 넣어줌\n",
        "history_autoencoder_serial_channel_first = model_autoencoder_serial_channel_first.fit(x_train, x_train, \n",
        "                    batch_size=16, \n",
        "                    epochs=epoch,\n",
        "                    shuffle=True,\n",
        "                    validation_data=(x_test, x_test))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Training Model - autoencoder_serial_channel_first\n",
            "Epoch 1/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0065 - mse: 0.0065 - val_loss: 0.0037 - val_mse: 0.0037\n",
            "Epoch 2/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0033 - mse: 0.0033 - val_loss: 0.0029 - val_mse: 0.0029\n",
            "Epoch 3/25\n",
            "3750/3750 [==============================] - 39s 10ms/step - loss: 0.0027 - mse: 0.0027 - val_loss: 0.0024 - val_mse: 0.0024\n",
            "Epoch 4/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0024 - mse: 0.0024 - val_loss: 0.0023 - val_mse: 0.0023\n",
            "Epoch 5/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0022 - mse: 0.0022 - val_loss: 0.0021 - val_mse: 0.0021\n",
            "Epoch 6/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0021 - mse: 0.0021 - val_loss: 0.0022 - val_mse: 0.0022\n",
            "Epoch 7/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0020 - mse: 0.0020 - val_loss: 0.0020 - val_mse: 0.0020\n",
            "Epoch 8/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0020 - mse: 0.0020 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 9/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0019 - mse: 0.0019 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 10/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0019 - mse: 0.0019 - val_loss: 0.0023 - val_mse: 0.0023\n",
            "Epoch 11/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 12/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 13/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 14/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 15/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 16/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0015 - val_mse: 0.0015\n",
            "Epoch 17/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 18/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 19/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 20/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0015 - val_mse: 0.0015\n",
            "Epoch 21/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 22/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0015 - val_mse: 0.0015\n",
            "Epoch 23/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0015 - val_mse: 0.0015\n",
            "Epoch 24/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 25/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0015 - val_mse: 0.0015\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kZ-W4LxljhLA",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def autoencoder_serial_spatial_first(input):\n",
        "    # Encoder\n",
        "    x = autoencoder_block(input, 16, 'encoder1')\n",
        "    \n",
        "    x = spatial_attention(x, 16, 'spatial_attention')\n",
        "    x = channel_attention(x, 16, 'channel_attention')\n",
        "    \n",
        "    x = tf.keras.layers.MaxPooling2D((2, 2), name='encoder_pool1')(x)\n",
        "    x = autoencoder_block(x, 32, 'encoder2')\n",
        "\n",
        "    x = spatial_attention(x, 32, 'spatial_attention2')\n",
        "    x = channel_attention(x, 32, 'channel_attention2')\n",
        "\n",
        "    x = tf.keras.layers.MaxPooling2D((2, 2), name='encoder_pool2')(x)\n",
        "    \n",
        "    encoded_img = autoencoder_block(x, 64, 'encoder3')\n",
        "\n",
        "    # Decoder\n",
        "    x = autoencoder_block(encoded_img, 64, 'decoder1')\n",
        "\n",
        "    x = spatial_attention(x, 64, 'spatial_attention3')\n",
        "    x = channel_attention(x, 64, 'channel_attention3')\n",
        "\n",
        "    x = tf.keras.layers.UpSampling2D((2, 2), name='decoder_up1')(x)\n",
        "    x = autoencoder_block(x, 32, 'decoder2')\n",
        "\n",
        "    x = spatial_attention(x, 32, 'spatial_attention4')\n",
        "    x = channel_attention(x, 32, 'channel_attention4')\n",
        "\n",
        "    x = tf.keras.layers.UpSampling2D((2, 2), name='decoder_up2')(x)\n",
        "    x = autoencoder_block(x, 16, 'decoder3')\n",
        "\n",
        "    decoded_img = tf.keras.layers.Conv2D(filters=1, kernel_size=3, strides=(1, 1), \n",
        "                                         activation='sigmoid', padding='same', \n",
        "                                         name='last_conv')(x)\n",
        "    return decoded_img   "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "buUvXfzFjf0t",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 3761
        },
        "outputId": "4a91b1b2-840e-4a77-9d4b-4648acc9de99"
      },
      "source": [
        "# Normal AutoEncoder Model Create / Summary\n",
        "input_autoencoder_serial_spatial_first_data = tf.keras.Input(shape=(28, 28, 1))\n",
        "\n",
        "output_autoencoder_serial_spatial_first_data = autoencoder_serial_spatial_first(input_autoencoder_serial_spatial_first_data)\n",
        "\n",
        "model_autoencoder_serial_spatial_first  = tf.keras.Model(inputs=input_autoencoder_serial_spatial_first_data, outputs=output_autoencoder_serial_spatial_first_data)\n",
        "\n",
        "model_autoencoder_serial_spatial_first.summary()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"model_6\"\n",
            "__________________________________________________________________________________________________\n",
            "Layer (type)                    Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            "input_6 (InputLayer)            [(None, 28, 28, 1)]  0                                            \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_conv (Conv2D)          (None, 28, 28, 16)   160         input_6[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_bn (BatchNormalization (None, 28, 28, 16)   64          encoder1_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder1_relu (ReLU)            (None, 28, 28, 16)   0           encoder1_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_64 (Conv2D)              (None, 28, 28, 4)    68          encoder1_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_64 (BatchNo (None, 28, 28, 4)    16          conv2d_64[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_64 (ReLU)                 (None, 28, 28, 4)    0           batch_normalization_64[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_65 (Conv2D)              (None, 28, 28, 4)    148         re_lu_64[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_65 (BatchNo (None, 28, 28, 4)    16          conv2d_65[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_65 (ReLU)                 (None, 28, 28, 4)    0           batch_normalization_65[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_66 (Conv2D)              (None, 28, 28, 4)    148         re_lu_65[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_66 (BatchNo (None, 28, 28, 4)    16          conv2d_66[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_66 (ReLU)                 (None, 28, 28, 4)    0           batch_normalization_66[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_67 (Conv2D)              (None, 28, 28, 1)    5           re_lu_66[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_32 (TensorFlowO [(None, 28, 28, 16)] 0           encoder1_relu[0][0]              \n",
            "                                                                 conv2d_67[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_16 (Gl (None, 16)           0           tf_op_layer_Mul_32[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "dense_32 (Dense)                (None, 4)            68          global_average_pooling2d_16[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_67 (BatchNo (None, 4)            16          dense_32[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_67 (ReLU)                 (None, 4)            0           batch_normalization_67[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "dense_33 (Dense)                (None, 16)           80          re_lu_67[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_16 (T [(None, 1, 1, 16)]   0           dense_33[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_33 (TensorFlowO [(None, 28, 28, 16)] 0           tf_op_layer_Mul_32[0][0]         \n",
            "                                                                 tf_op_layer_strided_slice_16[0][0\n",
            "__________________________________________________________________________________________________\n",
            "encoder_pool1 (MaxPooling2D)    (None, 14, 14, 16)   0           tf_op_layer_Mul_33[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_conv (Conv2D)          (None, 14, 14, 32)   4640        encoder_pool1[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_bn (BatchNormalization (None, 14, 14, 32)   128         encoder2_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder2_relu (ReLU)            (None, 14, 14, 32)   0           encoder2_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_68 (Conv2D)              (None, 14, 14, 8)    264         encoder2_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_68 (BatchNo (None, 14, 14, 8)    32          conv2d_68[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_68 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_68[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_69 (Conv2D)              (None, 14, 14, 8)    584         re_lu_68[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_69 (BatchNo (None, 14, 14, 8)    32          conv2d_69[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_69 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_69[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_70 (Conv2D)              (None, 14, 14, 8)    584         re_lu_69[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_70 (BatchNo (None, 14, 14, 8)    32          conv2d_70[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_70 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_70[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_71 (Conv2D)              (None, 14, 14, 1)    9           re_lu_70[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_34 (TensorFlowO [(None, 14, 14, 32)] 0           encoder2_relu[0][0]              \n",
            "                                                                 conv2d_71[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_17 (Gl (None, 32)           0           tf_op_layer_Mul_34[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "dense_34 (Dense)                (None, 8)            264         global_average_pooling2d_17[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_71 (BatchNo (None, 8)            32          dense_34[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_71 (ReLU)                 (None, 8)            0           batch_normalization_71[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "dense_35 (Dense)                (None, 32)           288         re_lu_71[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_17 (T [(None, 1, 1, 32)]   0           dense_35[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_35 (TensorFlowO [(None, 14, 14, 32)] 0           tf_op_layer_Mul_34[0][0]         \n",
            "                                                                 tf_op_layer_strided_slice_17[0][0\n",
            "__________________________________________________________________________________________________\n",
            "encoder_pool2 (MaxPooling2D)    (None, 7, 7, 32)     0           tf_op_layer_Mul_35[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_conv (Conv2D)          (None, 7, 7, 64)     18496       encoder_pool2[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_bn (BatchNormalization (None, 7, 7, 64)     256         encoder3_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "encoder3_relu (ReLU)            (None, 7, 7, 64)     0           encoder3_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_conv (Conv2D)          (None, 7, 7, 64)     36928       encoder3_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_bn (BatchNormalization (None, 7, 7, 64)     256         decoder1_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder1_relu (ReLU)            (None, 7, 7, 64)     0           decoder1_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_72 (Conv2D)              (None, 7, 7, 16)     1040        decoder1_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_72 (BatchNo (None, 7, 7, 16)     64          conv2d_72[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_72 (ReLU)                 (None, 7, 7, 16)     0           batch_normalization_72[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_73 (Conv2D)              (None, 7, 7, 16)     2320        re_lu_72[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_73 (BatchNo (None, 7, 7, 16)     64          conv2d_73[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_73 (ReLU)                 (None, 7, 7, 16)     0           batch_normalization_73[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_74 (Conv2D)              (None, 7, 7, 16)     2320        re_lu_73[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_74 (BatchNo (None, 7, 7, 16)     64          conv2d_74[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_74 (ReLU)                 (None, 7, 7, 16)     0           batch_normalization_74[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_75 (Conv2D)              (None, 7, 7, 1)      17          re_lu_74[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_36 (TensorFlowO [(None, 7, 7, 64)]   0           decoder1_relu[0][0]              \n",
            "                                                                 conv2d_75[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_18 (Gl (None, 64)           0           tf_op_layer_Mul_36[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "dense_36 (Dense)                (None, 16)           1040        global_average_pooling2d_18[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_75 (BatchNo (None, 16)           64          dense_36[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_75 (ReLU)                 (None, 16)           0           batch_normalization_75[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "dense_37 (Dense)                (None, 64)           1088        re_lu_75[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_18 (T [(None, 1, 1, 64)]   0           dense_37[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_37 (TensorFlowO [(None, 7, 7, 64)]   0           tf_op_layer_Mul_36[0][0]         \n",
            "                                                                 tf_op_layer_strided_slice_18[0][0\n",
            "__________________________________________________________________________________________________\n",
            "decoder_up1 (UpSampling2D)      (None, 14, 14, 64)   0           tf_op_layer_Mul_37[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_conv (Conv2D)          (None, 14, 14, 32)   18464       decoder_up1[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_bn (BatchNormalization (None, 14, 14, 32)   128         decoder2_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder2_relu (ReLU)            (None, 14, 14, 32)   0           decoder2_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_76 (Conv2D)              (None, 14, 14, 8)    264         decoder2_relu[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_76 (BatchNo (None, 14, 14, 8)    32          conv2d_76[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_76 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_76[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_77 (Conv2D)              (None, 14, 14, 8)    584         re_lu_76[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_77 (BatchNo (None, 14, 14, 8)    32          conv2d_77[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_77 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_77[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_78 (Conv2D)              (None, 14, 14, 8)    584         re_lu_77[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_78 (BatchNo (None, 14, 14, 8)    32          conv2d_78[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_78 (ReLU)                 (None, 14, 14, 8)    0           batch_normalization_78[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_79 (Conv2D)              (None, 14, 14, 1)    9           re_lu_78[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_38 (TensorFlowO [(None, 14, 14, 32)] 0           decoder2_relu[0][0]              \n",
            "                                                                 conv2d_79[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "global_average_pooling2d_19 (Gl (None, 32)           0           tf_op_layer_Mul_38[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "dense_38 (Dense)                (None, 8)            264         global_average_pooling2d_19[0][0]\n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_79 (BatchNo (None, 8)            32          dense_38[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "re_lu_79 (ReLU)                 (None, 8)            0           batch_normalization_79[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "dense_39 (Dense)                (None, 32)           288         re_lu_79[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_strided_slice_19 (T [(None, 1, 1, 32)]   0           dense_39[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "tf_op_layer_Mul_39 (TensorFlowO [(None, 14, 14, 32)] 0           tf_op_layer_Mul_38[0][0]         \n",
            "                                                                 tf_op_layer_strided_slice_19[0][0\n",
            "__________________________________________________________________________________________________\n",
            "decoder_up2 (UpSampling2D)      (None, 28, 28, 32)   0           tf_op_layer_Mul_39[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_conv (Conv2D)          (None, 28, 28, 16)   4624        decoder_up2[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_bn (BatchNormalization (None, 28, 28, 16)   64          decoder3_conv[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "decoder3_relu (ReLU)            (None, 28, 28, 16)   0           decoder3_bn[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "last_conv (Conv2D)              (None, 28, 28, 1)    145         decoder3_relu[0][0]              \n",
            "==================================================================================================\n",
            "Total params: 97,257\n",
            "Trainable params: 96,521\n",
            "Non-trainable params: 736\n",
            "__________________________________________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sBBl69fgjfoq",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Model Compile\n",
        "model_autoencoder_serial_spatial_first.compile(optimizer=tf.keras.optimizers.Adam(), loss=tf.keras.losses.MeanSquaredError(), metrics=['mse'])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_SRZj9jDjfVV",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 935
        },
        "outputId": "9c707617-4e92-4e0a-8319-a9bb43cbe6f9"
      },
      "source": [
        "# Training\n",
        "print('Training Model - autoencoder_serial_spatial_first')\n",
        "epoch = 25\n",
        "# AutoEncoder는 input과 output이 같으므로 x_train을 넣어줌\n",
        "history_autoencoder_serial_spatial_first = model_autoencoder_serial_spatial_first.fit(x_train, x_train, \n",
        "                    batch_size=16, \n",
        "                    epochs=epoch,\n",
        "                    shuffle=True,\n",
        "                    validation_data=(x_test, x_test))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Training Model - autoencoder_serial_spatial_first\n",
            "Epoch 1/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0060 - mse: 0.0060 - val_loss: 0.0040 - val_mse: 0.0040\n",
            "Epoch 2/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0032 - mse: 0.0032 - val_loss: 0.0027 - val_mse: 0.0027\n",
            "Epoch 3/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0026 - mse: 0.0026 - val_loss: 0.0024 - val_mse: 0.0024\n",
            "Epoch 4/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0024 - mse: 0.0024 - val_loss: 0.0022 - val_mse: 0.0022\n",
            "Epoch 5/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0022 - mse: 0.0022 - val_loss: 0.0022 - val_mse: 0.0022\n",
            "Epoch 6/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0021 - mse: 0.0021 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 7/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0020 - mse: 0.0020 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 8/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0019 - mse: 0.0019 - val_loss: 0.0022 - val_mse: 0.0022\n",
            "Epoch 9/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0019 - mse: 0.0019 - val_loss: 0.0023 - val_mse: 0.0023\n",
            "Epoch 10/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 11/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0018 - mse: 0.0018 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 12/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 13/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0019 - val_mse: 0.0019\n",
            "Epoch 14/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0017 - mse: 0.0017 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 15/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0018 - val_mse: 0.0018\n",
            "Epoch 16/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 17/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 18/25\n",
            "3750/3750 [==============================] - 37s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0015 - val_mse: 0.0015\n",
            "Epoch 19/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0016 - mse: 0.0016 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 20/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0015 - val_mse: 0.0015\n",
            "Epoch 21/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0016 - val_mse: 0.0016\n",
            "Epoch 22/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0015 - val_mse: 0.0015\n",
            "Epoch 23/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0017 - val_mse: 0.0017\n",
            "Epoch 24/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0015 - val_mse: 0.0015\n",
            "Epoch 25/25\n",
            "3750/3750 [==============================] - 38s 10ms/step - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0015 - val_mse: 0.0015\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_Q7loQ-xjFGU",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Model Predict\n",
        "recons = model.predict(x_test)\n",
        "recons_channel = model_channel.predict(x_test)\n",
        "recons_autoencoder_spatial = model_autoencoder_spatial.predict(x_test)\n",
        "recons_autoencoder_parallel_add = model_autoencoder_parallel_add.predict(x_test)\n",
        "recons_autoencoder_parallel_mul = model_autoencoder_parallel_mul.predict(x_test)\n",
        "recons_autoencoder_serial_channel_first = model_autoencoder_serial_channel_first.predict(x_test)\n",
        "recons_autoencoder_serial_spatial_first = model_autoencoder_serial_spatial_first.predict(x_test)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8kTUHDA0mcdD",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "recons = recons.reshape(-1, 28, 28)\n",
        "recons_channel = recons_channel.reshape(-1, 28, 28)\n",
        "recons_autoencoder_spatial = recons_autoencoder_spatial.reshape(-1, 28, 28)\n",
        "recons_autoencoder_parallel_add = recons_autoencoder_parallel_add.reshape(-1, 28, 28)\n",
        "recons_autoencoder_parallel_mul = recons_autoencoder_parallel_mul.reshape(-1, 28, 28)\n",
        "recons_autoencoder_serial_channel_first = recons_autoencoder_serial_channel_first.reshape(-1, 28, 28)\n",
        "recons_autoencoder_serial_spatial_first = recons_autoencoder_serial_spatial_first.reshape(-1, 28, 28)\n",
        "\n",
        "test_imgs = (x_test * 255).reshape(-1, 28, 28)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1s17pZQOkWaD",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 188
        },
        "outputId": "79acca43-09f0-4a2e-aff1-fa590b6a2af1"
      },
      "source": [
        "# Test Data Visualization\n",
        "fig = plt.figure(figsize=(16, 10))\n",
        "\n",
        "for i in range(5):\n",
        "    axes = fig.add_subplot(1, 5, i+1)\n",
        "    axes.imshow(test_imgs[i], cmap='gray')\n",
        "    axes.set_xticks([])\n",
        "    axes.set_yticks([])\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x720 with 5 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7VeP_uXykXYj",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 188
        },
        "outputId": "59e7d020-a5e5-453f-d2c6-7550a686f804"
      },
      "source": [
        "# Predictions Visualization\n",
        "fig = plt.figure(figsize=(16, 10))\n",
        "\n",
        "for i in range(5):\n",
        "    axes = fig.add_subplot(1, 5, i+1)\n",
        "    axes.imshow(recons[i], cmap='gray')\n",
        "    axes.set_xticks([])\n",
        "    axes.set_yticks([])\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAA44AAACrCAYAAAApfsKMAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAfeklEQVR4nO3dbWzWZ/n/8YMxWmhLSym0UMrNUO4EBhkgYBQcukmMbpIY75AgcXtg4n1IjGaJyx7NGM0So0an0SxjTpOZTHSLc2gGYWpRpgEdrA4KLS0tvb9v5eb/5O8/f3/H5zhzfX+9u2jfr4dHDq7r2+s6v+f3PL08P5tx69YtAwAAAAAgcsdkXwAAAAAAIL+xcQQAAAAAJLFxBAAAAAAksXEEAAAAACSxcQQAAAAAJN2ZpXnGjBlEsGIitN26dWvhRL/pVB7fM2bMcLU77tD/u9GNGzfG5Rqi91Nu3rw5LteQJyZlfJtN7TFeUlLialVVVbJ3eHjY1aIxp+pZxnLUW1BQ4GoXLlzI+XXzHGN8FGbOnCnrFRUVrjYyMiJ71Zw/Z84c2ZvlflDjOUrnHxoayqmWer88xhifILNnz865Hs23Wf4LEureGRgYkL3ReJ4i5BjPtHEEJsilyb6AqUYtUouLi2VvT0+Pq2XZTKpJ10wv7KPFQn9/v6tNof90EON7HNxzzz2u9sUvflH2Xrrkv4K+vj7Zq8ZitABXY3/u3Lmyt7q62tU+/vGPy97bcFHNGB8FNVeame3fv9/V6uvrZa/afG7cuFH2Xrx40dWi+0GN52jxXFdX52rnzp2TvdH75THG+DhQ4/Ytb3mL7F29erWrFRUVyd5///vfrhatKe6802+Nzpw5I3vVeL5+/brsvQ3JMc7/VRUAAAAAkMTGEQAAAACQxMYRAAAAAJDEGUcgzxUWFsr6wYMHXe3w4cOyt7y83NVKS0tzvoYogCELdW4gOgugDqK/9tprsveJJ55wtd/+9reydwqdk8T/8PTTT7taTU2N7FVndqOzuep8YdSrXledrTHT5yS3bt0qe2tra2UdU9PDDz8s69/85jddrbGxUfZ2dXW5WhQWdfXqVVcbHByUveq5EQWHqDn7ueeek70vvviirOP2p87F7tixQ/Z+5CMfcbXdu3fL3ra2NleLnvFqHo9yHtRrRGsVNY//5Cc/kb1/+MMfXC06OzleIYVjgV8cAQAAAABJbBwBAAAAAElsHAEAAAAASWwcAQAAAABJbBwBAAAAAEmkqgJ55Nvf/rarRQl7d9zh/3ef/v5+2asSujo7O2WvSj+Nkl2VoaEhWVfpksPDw7JXvd/OnTtlr6rX19fL3m3btrnaWCTGYuLMnDlT1lVCXm9vr+zt7u52tSxJqVHCnrono5S/yspKV7v77rtlL6mq04tKoTQza2lpcbVojKv5NupVY7SgoED2zpo1y9VUsqSZnseLiopkL24vav46dOiQ7F28eLGrRfO4SvP9/e9/L3tVwm+UlKrWJdGzX6UERynDK1eudLU9e/bI3n379rlatGZT674oQXmi8YsjAAAAACCJjSMAAAAAIImNIwAAAAAgiY0jAAAAACCJcBxgEnzve9+T9QMHDrhaFCCjwjyigA8VXpAlFCbqzRJ4E12bosJHosPpKsRh7dq1svf48eOutmPHjpyvC5MvClWYN2+eq/X09MheFdqhQqHM9NiPrkEFh0ThOOoaNmzYIHsxvUQBH2puVWPOTAc1Rb1q7N+8eVP2qmdJ9Loq5Of8+fOyF/kpem4fPHjQ1aLgo7a2NleLQvTUfBldw/Lly12turpa9qr7Ibre9vZ2V4vm/CtXrrhatF5S17Bq1SrZ+/3vf9/V9u/fL3uj59x44RdHAAAAAEASG0cAAAAAQBIbRwAAAABAEhtHAAAAAEASG0cAAAAAQBKpqsAkePDBB2VdpXHduHFD9qrEut7eXtmr0vhUGqlZtvSxKHkv116VyhqJ0im7u7tdLfrMVNpqSUmJ7O3r68v52jBxou9Ljdso4VelQEbJfSrhNxpfuV5X9BoVFRU5vy6mrmiuU+mJ0Tw+e/ZsV4vm62iMKuoZpd7LTD+Pmpubc34vTL5FixbJek1NjauphFEzPbdGY06ll0aJpg0NDa62YMEC2aue/a2trbL32rVrrlZZWSl71fMhSypydA1LlixxtShVVSWwjid+cQQAAAAAJLFxBAAAAAAksXEEAAAAACSxcQQAAAAAJBGOA0yC0tJSWVdhMVFohwrX+PKXvyx71WHvn/3sZ7K3trbW1RYvXix7VchIdJBdHRjPEkAzODgo6+Xl5a4WHbxX17t9+3bZe+zYsZyvDRNn06ZNOfdG40sF7ERjRgWVRPekeo2oV71uYWGh7MX00tHRIev19fWu1t/fL3vVMya6d7IEkpw/f97VVFCbmb7/outFftqxY4esq7kuGjPKwMCArEfzpaLWBGr9Yma2e/duV2tqapK96m+LAquy3DsqRCoKRFPX8M53vlP2Eo4DAAAAAMgrbBwBAAAAAElsHAEAAAAASWwcAQAAAABJbBwBAAAAAEmkqgKT4Pr167Ku0rhU0qqZ2Zw5c1wtSpF8+eWXXS1KdlVJY6+88orsVUlljzzyiOz92Mc+5mpRyt/jjz/uasuWLZO93/jGN1wtS+rlhg0bZC+pqvlp69atsn7z5k1Xu/NO/YhT6boq9ddMp+lFCawq6Vhdl5keo93d3bIX00uUqnr27FlXe/PNN2Vvc3Ozqz366KOyV6U9rlu3Tva+8MILrhalSDY2NrrayMiI7EV+WrlypaxnmRejOVBR653o36t0bJUgb6bXMNF8W1ZWlrrE/6LWctH6Q9Wj546yatWqnF83SoEdC/ziCAAAAABIYuMIAAAAAEhi4wgAAAAASGLjCAAAAABIIhwnz0WHbMfz4OtoLFq0SNZbWlpcLV//hrGmgjiiMIEsh5zVazz44IOy9/nnn8/5GtRB9L/+9a+yd9++fa524cIF2fvUU0+52nvf+17Ze+LECVerqqqSveqA/ODgoOxVh/ejgB7kp/vvv1/W1biNxri6p6IAhitXrrja2rVrZe/Vq1ddTd3/Zjq4Z2hoSPZieokCZFQQRxSepsbdL3/5S9mrxviPfvQj2avCR6IwHxXWNl2e+7cjNV9GoXRq/lIhS2Z6XssSIJNlDi0vL5e9tbW1rlZYWCh7h4eHXW3+/Pmyt6KiwtWikKBc/72Zvs+iIB0VlDgwMJDzNWTFL44AAAAAgCQ2jgAAAACAJDaOAAAAAIAkNo4AAAAAgCQ2jgAAAACAJFJV84hKbXrPe94je3/zm9+42nimKCkqyelDH/qQ7P3pT3/qatMlQbCsrMzVVBqYmU7Ni6jkvfXr18te9X43btyQvb29va52/Phx2XvkyBFXixLF1Os+++yzslelmqmUMTOdKhilpSnbt2/PuReTr7q6WtbVvRPdTyqJt6GhQfbu3bvX1S5fvix71ZwYpbWqdMmuri7Zi+klyziIxteaNWtc7a1vfavsfeGFF1wtSiRW6ww1t0f12y0pfjpRqfjRs1R9jwsWLJC9jY2NOV+DGnfRmkLN71Gya67vlZVKcY3Gsvoso7TW1tZWV4vu9aVLl7ra+fPnZe9Y4BdHAAAAAEASG0cAAAAAQBIbRwAAAABAEhtHAAAAAEAS4Th5ZM+ePa62ZcsW2asOIT/55JOyV4WgZDmIrgIfzMx27drlatHh+yyhL1ONOvwcHfZW30sUpKMOSquD2mY6oKepqUn2Hjt2zNWiQJJ3vetdrva2t71N9q5du9bVTp48KXvr6+tdLQqxUeOzr69P9qqD85WVlbIX+SkKYFD3Q3TvFBQUuFpdXZ3sbWtrG9XrRiFgKlyCgBCYjU1wXJbgkP7+fleLxmJRUZGrzZo1S/b29PSkLhF5RoXrqfA5Mz2WKioqZG9LS4urRXOoGndRKEwUtJRrb/Tv1TweXYMKWos+M7VWiQJ6VECgqpmZveMd73A1wnEAAAAAAJOGjSMAAAAAIImNIwAAAAAgiY0jAAAAACCJjSMAAAAAIIlU1UmwevVqWd+0aZOrRclm9913n6uphEszndqkklbNzK5du+ZqUQKoSlfr7u6WvVEi1XSgUlUj6nMqLCyUvep7jZLKVO/WrVtlr0oEi5IsP//5z7va+9//ftn70Y9+1NWeeOIJ2Xv06FFXU8lsZmaf/OQnXU2lBJrpJLji4mLZi/yk5h0zs5GREVeL5jmVpved73xH9qp7MkrjU+mS0VhUc3t0vZheohRyNRZVAqSZnvNVQrCZWUNDg6tFyZDq/ot6VTrsdF4L5LuVK1e6WvTdqvlr2bJlslc9Yzs7O2VvtIbJVTQ3q3qUBqzWXNE6WN1/0bpd1aPPV9W7urpk7/Lly2V9vPCLIwAAAAAgiY0jAAAAACCJjSMAAAAAIImNIwAAAAAgiXCccVZaWupqX/jCF2Rvb2+vq0UHctXB4iiERR2Sjw4Qz5kzx9WGh4dlr9LX1yfr0ftNBxs2bHC16PC0+qxVoIuZPjxdVVUlew8fPuxq6iC8mQ4ZiQ5fq2soKyuTverveOyxx2RvR0eHq82dO1f2qrEV3TfqeqPPF5MvS6CBCuKIetX999JLL+V8XVGwgwpViMJA1N9GOA7M9FrATM9rWQKgoueOCuOJektKSlwtWiNEwVDIT4sWLXK16DtUYzF6Rqu5OctzV61JzLIFjKl1cDTGVUBP1Kv+tugaVD36fNVaJZoXtmzZIuvjhV8cAQAAAABJbBwBAAAAAElsHAEAAAAASWwcAQAAAABJbBwBAAAAAEmkqo4RldhkZvaZz3zG1f71r3/JXpVstmTJEtmrEq2iRFOVBhVdr0oAVKlRZjoVa+HChbJXJbF1d3fL3qlGpZdmSfPKkqwbpaqqcajey8xscHDQ1aKUsCwpfyolrKamRvb+6le/crVozKr7Ro23yHRO/M13WRJzsySwqhTILImmp0+flvXdu3e7WkFBgewlVRWRtrY2WVcp7Wq+NtPrgaKiItmrnvtqvjbTyesRlY6N/KW+2yjRVI27KFH973//u6utX79e9qqU0Szrj4iab6Nnv1obqXVGVI+eO2fOnHG1jRs3yt6enh5ZV+6+++6ce8cCvzgCAAAAAJLYOAIAAAAAktg4AgAAAACS2DgCAAAAAJIIx/lfUAdyH3jgAdm7bt06V3v11VdlrwqbqayszPm6ovAQVY+CUdQhZBUkYaYP1EeH75cuXepq0yUcJ0uYQGFhoatF4UQvvviiqx08eFD2RsFJuVLftZm+3ujAujpErg7CR68bhTUcPXrU1Q4dOiR71YH+6IB8RUWFq7W3t8tejA8VIhV9X6qu5lQzs66urlFd19NPPy3re/bsGdXrRmMc00tra6usqzEezc2qHgV8qNeNnvtqPRGF+UR15Kf58+e7WnNzs+zt7e11NRXeZKYDIaNAF7UGjcIE1VojWn+oenTvqPeL1szqNaJ759SpU6728MMPy14lun8XLFjgatH1Rq+RBb84AgAAAACS2DgCAAAAAJLYOAIAAAAAktg4AgAAAACS2DgCAAAAAJJIVU2IUok+8YlPuNrmzZtl7+uvv+5q8+bNk71z5851NZWWZGZ28eJFVysvL5e9KqUqotKkouQplQA4a9Ys2bt3715X+8c//pHzNdzOVNJY9DeqFMhLly7J3h/+8Ieu9qlPfUr2qrEcfa8qfSzLdxKlXipR4qwaW1FCpkopfuihh2SvSqiNXlfde6Sq3l6isfjaa6+N6nWjZOws16BEKcOYXgYGBmRdpTWOjIzI3oKCAleLxpea86NrUAnd165dk73IT1HyqFonRM9+Ne6iVH313I3GrRLNoSp9Pbre6L86MNprUJ9ldJ+98cYbrqb+BjP9d0Tfm3oNlQpvZtbS0iLrWfCLIwAAAAAgiY0jAAAAACCJjSMAAAAAIImNIwAAAAAgiXCc/6usrMzVDhw4IHtVuE1DQ4PsVQfUVQiOmT5Y3NHRIXvVwdnr16/LXhWCEgWjqLo6kG+mg3Ci692yZYurRcEo0aH825X6XqNDzuoAd3Nzs+xV33f0vaqD6FGvOmgd9argpShUSvVGB9nV3xbdN5cvX875dXO9LrM4xAoTR90nUUhBlhAaFVKQhRpzkSyhDITjwCyek9R6InqWqHCvrq6unK/hypUrsq7m9yhgLHrGY3JFIYZZghTV87GpqUn2Dg4Oulo0X6tnf7SmUH9H9Deo98sSEKgC+6LXiP623t5eV1PBQWbZgorUZ7Z48WLZSzgOAAAAAGDcsXEEAAAAACSxcQQAAAAAJLFxBAAAAAAksXEEAAAAACTlRaqqSiCK0oNUb5TctXr1aldT6Z5mZosWLXI1lURpppPviouLZa9KRooSrVRvlGg62qTOiPp8S0pKZG+W6125cmVONTOzs2fPpi7xtqPS7aIEXPWZtrW1yd5ly5a5WpTmpb6XaBxmSadU92mWZLYsaa3RZ6Z6s6SlRXNNlBSIiaO+W5XQZ6a/W5VCaTb65Obo2aCS96KxqO4zUlVhFs9JKulUpWCbmVVXV7vayy+/LHvV3Pq3v/0t59eNEq9LS0tlHZOroqJC1rMkqi9fvtzVorWbeo3y8nLZq9Y70RpfzflZnv3Rmlm9RjTnq/V1tOZW91lra6vsVc+oaE2i5ovxTIXnF0cAAAAAQBIbRwAAAABAEhtHAAAAAEASG0cAAAAAQNKow3GiQ6Dq4GtNTY3sVYdsIyoUZsmSJbJXHcyOghXUYdjoIGqWwBtVj4J01AFg9V6R6LtQIQzRAeIsr6vq0aF+9b0tXbpU9k61cBz1HUYBMuq7unjxouxVIQVRsI36rqJQhSxhSkp0DVlCd9S9F12XOkwffb7qfoxCd6qqqlKXiEly+fJlWV+xYoWrRd/t9u3bx/KS/h81/2W5H6IwH8BMz4HRGkGtMy5cuJDze507d07W1b2jnu/RNWDyVVZWyrqav6LxpV6jvr5e9o52TRHJEsSn/rboulSvCj4z02uKKCRSvW5jY6PsVZ979Dzr7u52tShQaCzwiyMAAAAAIImNIwAAAAAgiY0jAAAAACCJjSMAAAAAIImNIwAAAAAgKVOq6owZM1wS486dO2WvSiTMkl6YJaX0jjv0/re/v9/VoiRJ9RpR4pJKOxqL1KiBgQFXi5KcSkpKXC1KmFL1kZER2aveL0pgVdcQpQKqv62zs1P2TjUq3SpLmtfrr78ue7dt2+ZqWZJLxyItV4l61bVFY1Z9ZlFvc3Nzzr1q/ojuMZXKjMl38uRJWb/rrrtcLXrmRInOo6XGXZRKmCXlD9NLtKY5dOiQq9XV1cnee+65x9U+97nP5XwNf/zjH2X98OHDrrZ//37Ze+bMGVerra3N+RowPioqKmRdzUlROufChQtd7cSJE7K3rKzM1To6OlKX+F+i5/lo1yVDQ0OyV83Daj9hphNU582bJ3vV5xAl0arU2ugzU8nyKmV8rPCLIwAAAAAgiY0jAAAAACCJjSMAAAAAIImNIwAAAAAgKVM4TllZmb373e/+r9qnP/1p2asOQHd1dcleFZrR3d0te9Uh2ShIR4USROEtRUVFrhYdClbBMsXFxbI3y8HZ8vJyV6uurpa96vCuCqCJeqPPTIVJROEh6rBxdA19fX2upg70ThdR4JAShWusXbvW1aIwEHXfRAfL1bWNRW+W0B31utE9pg7pt7e3y97Fixe7WhQUpcKfMPl+97vfyfqBAwdcLQpVmD9/vqtFwTTqNaIQKvUaWQJvsoQ9YOqKng9qfKxZs0b2qrWOeg5HoiCO2bNnu9rmzZtlrwr4eOSRR2QvY398qPXDhg0bZK8aH9H6Q82hp06dkr1qbRtR82UUFqXm4ahXydKr9ghmeq2xZMkS2bty5UpXi/Y669atc7Vozay+t2heGAv84ggAAAAASGLjCAAAAABIYuMIAAAAAEhi4wgAAAAASGLjCAAAAABIypSq2tPT4xLt3v72t8venTt3utrcuXNlr0oritI5W1paXK25uVn29vb2ulqUOqmSQ6MUR5W2GKVGrVq1ytWipFT1N0fpTFevXnW1KDFWJUlGVLJZlHam0gaHhoZkb5bPd6pRqZ3ROFSiFF41NlSKbyTL9xrJkkqm0tKi5GKVwBylnO7evdvV/vnPf8reRYsWuVqUkFlaWirrmFwXL17MuTdK8lai9MAsqaqqN0rIVPVoDgfMzPr7+10tWlepBMYoIV2J5na1XouSg+vq6lyN9NTJF60/VFLqihUrZK9alzQ1NcneTZs2uVo0vtRYiuZmNQ9Hc7MSzc3q81FpwmZmra2trhbtXz74wQ+62okTJ2TvV77yFVeL0ouPHTvmapcuXZK9Y4FfHAEAAAAASWwcAQAAAABJbBwBAAAAAElsHAEAAAAASZnCcW7evOkOZ3/ta1+Tverga0VFhezdvn27q61fv172btmyJaeaWbZDtuqQrDoEbmY2a9YsV2tvb5e9r776qqu98sorsvf555/P6brM9AHzwsJC2fvMM8+4Wk1NjexV4TYq3MVMHwru7OyUvap+7tw52TvVZBmHqnfZsmWyt6yszNWicCJ1sDsKNFBjKxqHWcJx1KH16HNQf0d0DR/4wAdc7fTp07L33nvvzem6zPR9jsl39uxZWVfzVDQnqkCRaE588803XS0K+FDhVNF9psZzNNcCZnoeX7hwoezt6OhwtWgOVaKADzWeo7kyS5AVxocKennyySdl77PPPutq0fy1YMECV1NrQjOzrVu3ulr03FWhltFaXK0/ojWJGvvR36ZeIxrj6j6JQoLU+uPRRx+VvSrIL1pfq9Cr8Qyh4hdHAAAAAEASG0cAAAAAQBIbRwAAAABAEhtHAAAAAEASG0cAAAAAQFKmVNUsVILRtWvXZO+vf/3rnGpjIUpyUsYzlWg8DA4Oyvq+ffsm+Erw/1P3Qpa03Hnz5slelfLV3d2d8+veuHFD9qpEsSz3TZYUySypZtFntm7dOlc7cuSI7FV/R/Q5lJSUyDomV5Qc3NXV5WpRaq8SpVNmSVVta2tztaqqKtmrXkMlCgL/ocZMNDerNM0sVIJr9H7R3Dzaa8D4iNKbVWpnNNf19PS4WnFxsexVY0YlgZrpOTvqVWug0tJS2ZtFX1+fq82ZM0f2qjHe3Nwse9X17tq1S/a+9NJLrpYvqdv84ggAAAAASGLjCAAAAABIYuMIAAAAAEhi4wgAAAAASBq3cJx8dbsF3uD2Nzw87GpRmIAK/vjud78re/fu3etq5eXlsndgYCDna1DBNFHISJb7Sb1fFJ5QWFjoalEIhDpwXldXJ3uzhEtEnyXy05/+9CdXU/eImVlBQYGr3XvvvTm/bkQFRixevFj2qsCHKKgJMNPzYjQ3q/k2S8hZFNSk5mz1zDDTzz7kr2hNoKiglvvvv1/2rlixwtVOnTole6OAHUVdrwq2MdP3SXQ/qHWYemZE9Wjcd3R0uNrOnTtl7/Hjx11NPV8mA784AgAAAACS2DgCAAAAAJLYOAIAAAAAktg4AgAAAACS2DgCAAAAAJKmXaoqMNFU+liUQqdSwq5evSp7m5qaXG3ZsmU5X0OUxqcSxaK0R5UMGaWPqUTAKJVVJfdFSWXPPfecq7W0tMjeLKlxJSUlOfdi8n3rW99ytQceeED2qvF13333yd7HH3/c1aJxqxL9stzrKiEY+I/m5mZXq6mpkb2jTZCP0imV6FnS3t4+qmtA/lLz15o1a2SvWn9EqeVR0rqiUlGjOVQly2e5R6I10MyZM12tv79f9nZ1dbnahg0bZG9nZ2fO1zbR+MURAAAAAJDExhEAAAAAkMTGEQAAAACQxMYRAAAAAJBEOA4wzk6dOuVqO3fuzPnfX7lyRdY3btzoalnCX9TB8lRdUYfLo3+vAhSiUAV1EH327NmyV4U4ZHnd6IB8Y2OjrCM//eUvf3G1trY22VtaWppTzUyHH0QBDg0NDa62efNm2atCc1SAFPAff/7zn11t27ZtslfNi1meDyokLXrdsrIy2VtfX5/z++H2N3fuXFlfuHChq0VzswqF6e7ulr0qwC5LuE4UpKNC/6JwPhUAFa0/1Bpm+fLlsjfLOmyi8YsjAAAAACCJjSMAAAAAIImNIwAAAAAgiY0jAAAAACCJjSMAAAAAIIlUVWCc1dXVuVpBQUHO/37+/PmyHqWt5ipKE43qo33dkZGRnGqRKNVMqa6ulnWVdhall3V0dOT8fph8KjGyv79f9qqUv7vuukv2qntVpfOamS1ZssTVojRgJboGwMyspaXF1VTqr5lZZWWlq6kkX7NsadMq+TdKpySZenr5wQ9+IOuPPfaYq73vfe+TvWp8RWnAPT09rqYSUc2ypa2q+6SoqEj2FhcXu1qUAnv58mVX+/rXvy57s6yNJhq/OAIAAAAAktg4AgAAAACS2DgCAAAAAJLYOAIAAAAAkgjHAcZZbW2tq50+fVr2qsCb9vb2Mb+mqa61tVXWf/7zn7uaCpwwM/vFL34xpteEifelL31J1j/84Q+72lNPPSV7BwYGcn6/z372s6721a9+Vfaq8JEf//jHOb8Xpp9nnnnG1Xbt2iV7jx496mpRqFMWR44ccbUovOSNN94Y9fvh9hEF9j300EOuFgUEVlVVudqaNWtkb01NjavNmzdP9qpwHBXEY6bvkygsTwXenDx5Uvbmc+BNFvziCAAAAABIYuMIAAAAAEhi4wgAAAAASGLjCAAAAABIYuMIAAAAAEiaoZLdwuYZM66Z2aXxuxzAzMyW37p1a+FEvynjGxNkUsa3GWMcE4YxjqmOMY6pTo7xTBtHAAAAAMD0w/9VFQAAAACQxMYRAAAAAJDExhEAAAAAkMTGEQAAAACQxMYRAAAAAJDExhEAAAAAkMTGEQAAAACQxMYRAAAAAJDExhEAAAAAkPR/ALAnPnJP/UfNAAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<Figure size 1152x720 with 5 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ydRuypdq9RSx",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 188
        },
        "outputId": "6b137428-9b0a-4617-e06a-5e40651f4358"
      },
      "source": [
        "# Predictions Visualization - channel\n",
        "fig = plt.figure(figsize=(16, 10))\n",
        "\n",
        "for i in range(5):\n",
        "    axes = fig.add_subplot(1, 5, i+1)\n",
        "    axes.imshow(recons_channel[i], cmap='gray')\n",
        "    axes.set_xticks([])\n",
        "    axes.set_yticks([])\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x720 with 5 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Tq3Pl_TP9RCL",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 188
        },
        "outputId": "56e9f205-4b8f-40b9-9f6e-3cd0d2204fed"
      },
      "source": [
        "# Predictions Visualization\n",
        "fig = plt.figure(figsize=(16, 10))\n",
        "\n",
        "for i in range(5):\n",
        "    axes = fig.add_subplot(1, 5, i+1)\n",
        "    axes.imshow(recons_autoencoder_spatial[i], cmap='gray')\n",
        "    axes.set_xticks([])\n",
        "    axes.set_yticks([])\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x720 with 5 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Cvqgf1Ff9Q6j",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 188
        },
        "outputId": "af0ff17d-32d2-41f8-b665-587bd419a1bd"
      },
      "source": [
        "# Predictions Visualization\n",
        "fig = plt.figure(figsize=(16, 10))\n",
        "\n",
        "for i in range(5):\n",
        "    axes = fig.add_subplot(1, 5, i+1)\n",
        "    axes.imshow(recons_autoencoder_parallel_add[i], cmap='gray')\n",
        "    axes.set_xticks([])\n",
        "    axes.set_yticks([])\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x720 with 5 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HJwbFYIh9Qyj",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 188
        },
        "outputId": "eb46cd2c-a1f2-4d55-c9fa-1c364c15263b"
      },
      "source": [
        "# Predictions Visualization\n",
        "fig = plt.figure(figsize=(16, 10))\n",
        "\n",
        "for i in range(5):\n",
        "    axes = fig.add_subplot(1, 5, i+1)\n",
        "    axes.imshow(recons_autoencoder_parallel_mul[i], cmap='gray')\n",
        "    axes.set_xticks([])\n",
        "    axes.set_yticks([])\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x720 with 5 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ui0APoVy9Qkh",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 188
        },
        "outputId": "901ba2b3-8ef8-45b3-89de-f7c0f16019f4"
      },
      "source": [
        "# Predictions Visualization\n",
        "fig = plt.figure(figsize=(16, 10))\n",
        "\n",
        "for i in range(5):\n",
        "    axes = fig.add_subplot(1, 5, i+1)\n",
        "    axes.imshow(recons_autoencoder_serial_channel_first[i], cmap='gray')\n",
        "    axes.set_xticks([])\n",
        "    axes.set_yticks([])\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x720 with 5 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AchtUBdD9QG2",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 188
        },
        "outputId": "da8b0383-f838-43fa-df30-a9b8f560b367"
      },
      "source": [
        "# Predictions Visualization\n",
        "fig = plt.figure(figsize=(16, 10))\n",
        "\n",
        "for i in range(5):\n",
        "    axes = fig.add_subplot(1, 5, i+1)\n",
        "    axes.imshow(recons_autoencoder_serial_spatial_first[i], cmap='gray')\n",
        "    axes.set_xticks([])\n",
        "    axes.set_yticks([])\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x720 with 5 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NjLjbHGrS2dp",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 598
        },
        "outputId": "6f241ee7-fef1-4e6e-988e-a48bbf606d77"
      },
      "source": [
        "plt.figure(figsize=(24, 12))\n",
        "\n",
        "plt.subplot(2,4,1)\n",
        "plt.plot(history.history['loss'])\n",
        "plt.plot(history.history['val_loss'])\n",
        "plt.title('Model Loss')\n",
        "plt.ylabel('Loss')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Training Loss', 'Validation Loss'], loc=0)\n",
        "\n",
        "plt.subplot(2,4,2)\n",
        "plt.plot(history_channel.history['loss'])\n",
        "plt.plot(history_channel.history['val_loss'])\n",
        "plt.title('Model Loss')\n",
        "plt.ylabel('Loss')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Training Loss', 'Validation Loss'], loc=0)\n",
        "\n",
        "plt.subplot(2,4,3)\n",
        "plt.plot(history_autoencoder_spatial.history['loss'])\n",
        "plt.plot(history_autoencoder_spatial.history['val_loss'])\n",
        "plt.title('Model Loss')\n",
        "plt.ylabel('Loss')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Training Loss', 'Validation Loss'], loc=0)\n",
        "\n",
        "plt.subplot(2,4,4)\n",
        "plt.plot(history_autoencoder_parallel_add.history['loss'])\n",
        "plt.plot(history_autoencoder_parallel_add.history['val_loss'])\n",
        "plt.title('Model Loss')\n",
        "plt.ylabel('Loss')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Training Loss', 'Validation Loss'], loc=0)\n",
        "\n",
        "plt.subplot(2,4,5)\n",
        "plt.plot(history_autoencoder_parallel_mul.history['loss'])\n",
        "plt.plot(history_autoencoder_parallel_mul.history['val_loss'])\n",
        "plt.title('Model Loss')\n",
        "plt.ylabel('Loss')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Training Loss', 'Validation Loss'], loc=0)\n",
        "\n",
        "plt.subplot(2,4,6)\n",
        "plt.plot(history_autoencoder_serial_channel_first.history['loss'])\n",
        "plt.plot(history_autoencoder_serial_channel_first.history['val_loss'])\n",
        "plt.title('Model Loss')\n",
        "plt.ylabel('Loss')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Training Loss', 'Validation Loss'], loc=0)\n",
        "\n",
        "plt.subplot(2,4,7)\n",
        "plt.plot(history_autoencoder_serial_spatial_first.history['loss'])\n",
        "plt.plot(history_autoencoder_serial_spatial_first.history['val_loss'])\n",
        "plt.title('Model Loss')\n",
        "plt.ylabel('Loss')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Training Loss', 'Validation Loss'], loc=0)\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1728x864 with 7 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DNC8Uq2nvUHc",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}