Updated everything for Python3.6+ and TensorFlow1.13+

README.md

@@ -11,16 +11,15 @@ Tensorflow implementation for reproducing main results in the paper [StackGAN: T
 
 
 ### Dependencies
-python 2.7
+python 3.6+
 
-[TensorFlow 0.12](https://www.tensorflow.org/get_started/os_setup)
+[TensorFlow 1.13+](https://www.tensorflow.org/get_started/os_setup)
 
 [Optional] [Torch](http://torch.ch/docs/getting-started.html#_) is needed, if use the pre-trained char-CNN-RNN text encoder.
 
 [Optional] [skip-thought](https://github.com/ryankiros/skip-thoughts) is needed, if use the skip-thought text encoder.
 
 In addition, please add the project folder to PYTHONPATH and `pip install` the following packages:
-- `prettytensor`
 - `progressbar`
 - `python-dateutil`
 - `easydict`
@@ -32,7 +31,12 @@ In addition, please add the project folder to PYTHONPATH and `pip install` the f
 **Data**
 
 1. Download our preprocessed char-CNN-RNN text embeddings for [birds](https://drive.google.com/open?id=0B3y_msrWZaXLT1BZdVdycDY5TEE) and [flowers](https://drive.google.com/open?id=0B3y_msrWZaXLaUc0UXpmcnhaVmM) and save them to `Data/`.
+
   - [Optional] Follow the instructions [reedscot/icml2016](https://github.com/reedscot/icml2016) to download the pretrained char-CNN-RNN text encoders and extract text embeddings.
+
+  - [Optional] Download our preprocessed skip-thoughts text embeddings for [birds](https://drive.google.com/open?id=10jlSsU3g2ywDFXgUmn2Dh_UJCkQectzy) and save them to `Data/`.
+
+
 2. Download the [birds](http://www.vision.caltech.edu/visipedia/CUB-200-2011.html) and [flowers](http://www.robots.ox.ac.uk/~vgg/data/flowers/102/) image data. Extract them to `Data/birds/` and `Data/flowers/`, respectively.
 3. Preprocess images.
   - For birds: `python misc/preprocess_birds.py`
@@ -51,9 +55,9 @@ In addition, please add the project folder to PYTHONPATH and `pip install` the f
 
 
 **Pretrained Model**
-- [StackGAN for birds](https://drive.google.com/open?id=0B3y_msrWZaXLNUNKa3BaRjAyTzQ) trained from char-CNN-RNN text embeddings. Download and save it to `models/`.
-- [StackGAN for flowers](https://drive.google.com/open?id=0B3y_msrWZaXLX01FMC1JQW9vaFk) trained from char-CNN-RNN text embeddings. Download and save it to `models/`.
-- [StackGAN for birds](https://drive.google.com/open?id=0B3y_msrWZaXLZVNRNFg4d055Q1E) trained from skip-thought text embeddings. Download and save it to `models/` (Just used the same setting as the char-CNN-RNN. We assume better results can be achieved by playing with the hyper-parameters).
+- [StackGAN for birds](https://drive.google.com/open?id=1O1JHIoYO3h_qB5o27Td8KklvuLgTgpdV) trained from char-CNN-RNN text embeddings. Download and save it to `models/`.
+- [StackGAN for flowers]() trained from char-CNN-RNN text embeddings. Download and save it to `models/`.
+- [StackGAN for birds]() trained from skip-thought text embeddings. Download and save it to `models/` (Just used the same setting as the char-CNN-RNN. We assume better results can be achieved by playing with the hyper-parameters).
 
 
 
@@ -96,6 +100,12 @@ booktitle = {{ICCV}},
 - [StackGAN++: Realistic Image Synthesis with Stacked Generative Adversarial Networks](https://arxiv.org/abs/1710.10916)
 - [AttnGAN: Fine-Grained Text to Image Generation with Attentional Generative Adversarial Networks](https://arxiv.org/abs/1711.10485) [[supplementary]](https://1drv.ms/b/s!Aj4exx_cRA4ghK5-kUG-EqH7hgknUA) [[code]](https://github.com/taoxugit/AttnGAN)
 
+**Future**
+
+[Fashion Expansion](https://github.com/1o0ko/StackGAN-v1-TensorFlow)
+
+[Fashion Dataset](https://github.com/ayushidalmia/awesome-fashion-ai#datasets)
+
 **References**
 
 - Generative Adversarial Text-to-Image Synthesis [Paper](https://arxiv.org/abs/1605.05396) [Code](https://github.com/reedscot/icml2016)

demo/birds_demo.sh

@@ -1,3 +1,4 @@
+#!/usr/bin/env bash
 #
 # Extract text embeddings from the encoder
 #
@@ -15,7 +16,7 @@ th demo/get_embedding.lua
 #
 # Generate image from text embeddings
 #
-python demo/demo.py \
+python3 demo/demo.py \
 --cfg demo/cfg/birds-demo.yml \
 --gpu ${GPU} \
 --caption_path ${CAPTION_PATH}.t7

demo/birds_skip_thought_demo.py

@@ -1,30 +1,29 @@
 from __future__ import division
 from __future__ import print_function
 
-import prettytensor as pt
 import tensorflow as tf
 import numpy as np
-import scipy.misc
+import imageio
 import os
 import argparse
 from PIL import Image, ImageDraw, ImageFont
 
-from misc.config import cfg, cfg_from_file
-from misc.utils import mkdir_p
-from misc import skipthoughts
-from stageII.model import CondGAN
+import sys
+sys.path.append('misc')
+sys.path.append('stageII')
+
+import skipthoughts
+from config import cfg, cfg_from_file
+from utils import mkdir_p
+from model import CondGAN
+from skimage.transform import resize
 
 
 def parse_args():
     parser = argparse.ArgumentParser(description='Train a GAN network')
-    parser.add_argument('--cfg', dest='cfg_file',
-                        help='optional config file',
-                        default=None, type=str)
-    parser.add_argument('--gpu', dest='gpu_id',
-                        help='GPU device id to use [0]',
-                        default=-1, type=int)
-    parser.add_argument('--caption_path', type=str, default=None,
-                        help='Path to the file with text sentences')
+    parser.add_argument('--cfg', dest='cfg_file', help='optional config file', default=None, type=str)
+    parser.add_argument('--gpu', dest='gpu_id', help='GPU device id to use [0]', default=-1, type=int)
+    parser.add_argument('--caption_path', type=str, default=None, help='Path to the file with text sentences')
     # if len(sys.argv) == 1:
     #    parser.print_help()
     #    sys.exit(1)
@@ -49,21 +48,17 @@ def sample_encoded_context(embeddings, model, bAugmentation=True):
 
 
 def build_model(sess, embedding_dim, batch_size):
-    model = CondGAN(
-        lr_imsize=cfg.TEST.LR_IMSIZE,
-        hr_lr_ratio=int(cfg.TEST.HR_IMSIZE/cfg.TEST.LR_IMSIZE))
-
-    embeddings = tf.placeholder(
-        tf.float32, [batch_size, embedding_dim],
-        name='conditional_embeddings')
-    with pt.defaults_scope(phase=pt.Phase.test):
-        with tf.variable_scope("g_net"):
-            c = sample_encoded_context(embeddings, model)
-            z = tf.random_normal([batch_size, cfg.Z_DIM])
-            fake_images = model.get_generator(tf.concat(1, [c, z]))
-        with tf.variable_scope("hr_g_net"):
-            hr_c = sample_encoded_context(embeddings, model)
-            hr_fake_images = model.hr_get_generator(fake_images, hr_c)
+    model = CondGAN(lr_imsize=cfg.TEST.LR_IMSIZE, hr_lr_ratio=int(cfg.TEST.HR_IMSIZE/cfg.TEST.LR_IMSIZE))
+
+    embeddings = tf.placeholder(tf.float32, [batch_size, embedding_dim], name='conditional_embeddings')
+
+    with tf.variable_scope("g_net"):
+        c = sample_encoded_context(embeddings, model)
+        z = tf.random_normal([batch_size, cfg.Z_DIM])
+        fake_images = model.get_generator(tf.concat([c, z], 1,), False)
+    with tf.variable_scope("hr_g_net"):
+        hr_c = sample_encoded_context(embeddings, model)
+        hr_fake_images = model.hr_get_generator(fake_images, hr_c, False)
 
     ckt_path = cfg.TEST.PRETRAINED_MODEL
     if ckt_path.find('.ckpt') != -1:
@@ -101,9 +96,7 @@ def drawCaption(img, caption):
     return img_txt
 
 
-def save_super_images(sample_batchs, hr_sample_batchs,
-                      captions_batch, batch_size,
-                      startID, save_dir):
+def save_super_images(sample_batchs, hr_sample_batchs, captions_batch, batch_size, startID, save_dir):
     if not os.path.isdir(save_dir):
         print('Make a new folder: ', save_dir)
         mkdir_p(save_dir)
@@ -119,7 +112,7 @@ def save_super_images(sample_batchs, hr_sample_batchs,
             lr_img = sample_batchs[i][j]
             hr_img = hr_sample_batchs[i][j]
             hr_img = (hr_img + 1.0) * 127.5
-            re_sample = scipy.misc.imresize(lr_img, hr_img.shape[:2])
+            re_sample = resize(lr_img, hr_img.shape[:2])
             row1.append(re_sample)
             row2.append(hr_img)
         row1 = np.concatenate(row1, axis=1)
@@ -134,27 +127,23 @@ def save_super_images(sample_batchs, hr_sample_batchs,
                 lr_img = sample_batchs[i][j]
                 hr_img = hr_sample_batchs[i][j]
                 hr_img = (hr_img + 1.0) * 127.5
-                re_sample = scipy.misc.imresize(lr_img, hr_img.shape[:2])
+                re_sample = resize(lr_img, hr_img.shape[:2])
                 row1.append(re_sample)
                 row2.append(hr_img)
             row1 = np.concatenate(row1, axis=1)
             row2 = np.concatenate(row2, axis=1)
             super_row = np.concatenate([row1, row2], axis=0)
             superimage2 = np.zeros_like(superimage)
-            superimage2[:super_row.shape[0],
-                        :super_row.shape[1],
-                        :super_row.shape[2]] = super_row
+            superimage2[:super_row.shape[0], :super_row.shape[1], :super_row.shape[2]] = super_row
             mid_padding = np.zeros((64, superimage.shape[1], 3))
-            superimage =\
-                np.concatenate([superimage, mid_padding, superimage2], axis=0)
+            superimage = np.concatenate([superimage, mid_padding, superimage2], axis=0)
 
         top_padding = np.zeros((128, superimage.shape[1], 3))
-        superimage =\
-            np.concatenate([top_padding, superimage], axis=0)
+        superimage = np.concatenate([top_padding, superimage], axis=0)
 
         fullpath = '%s/sentence%d.jpg' % (save_dir, startID + j)
         superimage = drawCaption(np.uint8(superimage), captions_batch[j])
-        scipy.misc.imsave(fullpath, superimage)
+        imageio.imsave(fullpath, superimage)
 
 
 if __name__ == "__main__":
@@ -188,8 +177,8 @@ def save_super_images(sample_batchs, hr_sample_batchs,
         config = tf.ConfigProto(allow_soft_placement=True)
         with tf.Session(config=config) as sess:
             with tf.device("/gpu:%d" % cfg.GPU_ID):
-                embeddings_holder, fake_images_opt, hr_fake_images_opt =\
-                    build_model(sess, embeddings.shape[-1], batch_size)
+                embeddings_holder, fake_images_opt, hr_fake_images_opt = build_model(sess, embeddings.shape[-1],
+                                                                                     batch_size)
 
                 count = 0
                 while count < num_embeddings:
@@ -205,19 +194,14 @@ def save_super_images(sample_batchs, hr_sample_batchs,
                     # Generate up to 16 images for each sentence with
                     # randomness from noise z and conditioning augmentation.
                     for i in range(np.minimum(16, cfg.TEST.NUM_COPY)):
-                        hr_samples, samples =\
-                            sess.run([hr_fake_images_opt, fake_images_opt],
-                                     {embeddings_holder: embeddings_batch})
+                        hr_samples, samples = sess.run([hr_fake_images_opt, fake_images_opt],
+                                                       {embeddings_holder: embeddings_batch})
                         samples_batchs.append(samples)
                         hr_samples_batchs.append(hr_samples)
-                    save_super_images(samples_batchs,
-                                      hr_samples_batchs,
-                                      captions_batch,
-                                      batch_size,
-                                      count, save_dir)
+                    save_super_images(samples_batchs, hr_samples_batchs, captions_batch, batch_size, count, save_dir)
                     count += batch_size
 
         print('Finish generating samples for %d sentences:' % num_embeddings)
         print('Example sentences:')
-        for i in xrange(np.minimum(10, num_embeddings)):
+        for i in range(np.minimum(10, num_embeddings)):
             print('Sentence %d: %s' % (i, captions_list[i]))

demo/cfg/birds-demo.yml

@@ -5,7 +5,7 @@ GPU_ID: 0
 Z_DIM: 100
 
 TEST:
-    PRETRAINED_MODEL: './models/birds_model_164000.ckpt'
+    PRETRAINED_MODEL: './models/stageII/model_330000.ckpt'
     BATCH_SIZE: 64
     NUM_COPY: 8
 

demo/cfg/birds-eval.yml

@@ -7,7 +7,7 @@ Z_DIM: 100
 
 TRAIN:
     FLAG: False
-    PRETRAINED_MODEL: './models/birds_model_164000.ckpt'
+    PRETRAINED_MODEL: './models/stageII/model_330000.ckpt'
     BATCH_SIZE: 64
     NUM_COPY: 8
 

demo/cfg/birds-skip-thought-demo.yml

@@ -6,7 +6,7 @@ Z_DIM: 100
 
 TEST:
     CAPTION_PATH: './Data/birds/example_captions.txt'
-    PRETRAINED_MODEL: './models/birds_skip_thought_model_164000.ckpt'
+    PRETRAINED_MODEL: './models/stageII/model_330000.ckpt'
     BATCH_SIZE: 64
     NUM_COPY: 8