search

NVlabs / stylegan3

Official PyTorch implementation of StyleGAN3
Other
6.44k stars 1.13k forks source link

Generating CT images with corresponding labels #150

Open AndrisQBV opened 2 years ago

AndrisQBV commented 2 years ago

I am trying to generate synthetic CT images (grayscale) with corresponding pancreas and tumor labels (both are binary images). I use a 3 channel input image, where the CT image is on the first channel and the binary masks on the second and third channels:

reals_img_small reals_label_small reals_tumor_small

Is it possible to generate images and labels this way?

I have tried to disable augmentation and the RGB module of the network (set is_torgb false), although I got wrong results.

Thanks in advance! Every comment is appreciated!

whsu2s commented 2 years ago

Could you show the results you got? I guess the three channels should be more correlated in order for the StyleGAN to learn the features. One suggestion is to overlap the labels with the image, so the StyleGAN can tell that the pancreas and tumors are the regions of interest.

fonzen1 commented 1 year ago

Did you manage to work further upon this? I am interested to hear your approach and results.

xiaoerlaigeid commented 1 year ago

me too