this paper presents a network and training strategy that relies on the strong use of data augmentation to use the available annotated samples more efficiently. The architecture consists of a contracting path to capture context and a symmetric expanding path that enables precise localization.
Membrane contains 90 images for training and 30 for testing. The corresponding binary labels are provided in an in-out fashion, i.e. white for the pixels of segmented objects and black for the rest of pixels (which correspond mostly to membranes)
you can download Membrane dataset on the BaiduCloud Drive and put it in ./, then you can train it and then evaluate your model
$ python train.py
Epoch 1/5
Found 90 images belonging to 1 classes.
Found 90 images belonging to 1 classes.
5000/5000 [==============================] - 1443s 289ms/step - loss: 0.1926 - accuracy: 0.9456
Epoch 2/5
5000/5000 [==============================] - 1438s 288ms/step - loss: 0.1026 - accuracy: 0.9874
...
=> accuracy: 0.7934, saving ./results/pred_0.png
=> accuracy: 0.8132, saving ./results/pred_1.png
...
@Github_Project{TensorFlow2.0-Examples,
author = YunYang1994,
email = [email protected],
title = "U-Net: Convolutional Networks for Biomedical Image Segmentation",
url = https://github.com/YunYang1994/TensorFlow2.0-Examples,
year = 2019,
}