neoamos/3d-pix2pix-cyclegan

3D pix2pix/CycleGAN

Added 3D convolutional support to pix2pix/CycleGAN (https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix) in order to use 3D CT scans as a data source. Used in experiments in segmenting lung nodules.

3d-pix2pix-cyclegan's Issues

GPU issues

Hi, I have two GPUs from the cluster. But the GPU can not be used. If i change use_gpu = False, I can run the script. Do you have any idea what is going on there? Thanks a lot. i really appreciate all your help.

How should the dataset look like?

Sorry, maybe I'm thick but I can't figure out what format of volumetric data should be in the dataset?
In the image_folder.py the specified formats are .jpg, .jpeg, .png, .mbp and .ppm, which are from my knowledge all 2d formats, but you said you used this to segment lung CT images? Should I make subfolders inside of trainA, trainB, testA and testB for every image with slices in one of the above mentioned formats?
Thank you for your time and code!

Does a GAN improve your segmentation result?

What's the main idea of using a GAN in segmentation? Data Augmentation? And does it work?