Some questions about data processing

[acho-118]

Which conditions do you choose to excluded image patches that were mostly air? How do you randomly extract the image pairs? And in line 35 of train_wgan.py, your papar has given lambda=10 about the gradient penality, but this coefficient does not exist in your code, why?

[yyqqss09]

The condition for detecting air image patches is the 50% image pixels are zero. Random extracting is to generate a random number for each image pairs and select the image pairs if the residue between the number and 8 greater than 5 (you can decide these two numbers as needed for random sampling).

[acho-118]

The padding=valid in your code cnn_model, why not padding=same? Is the size of your input image and target image different during training? Is the input image 8080 and the target image 6464? I think the final purpose of the trained generator network is to denoise 512512 low-dose images and then generate 512512 high-dose images. If padding=valid will definitely change the image size, so I think it should be changed to padding=same.