Code repository for Frontiers article 'Generative Adversarial Networks for Image-to-Image Translation on Multi-Contrast MR Images - A Comparison of CycleGAN and UNIT'
Dear Simon, I switch our own dataset to the project but we got complete dark images. I mean, We just test the cycleGAN.py and our synthetic images are not what we want. Could you please tell me why this happen? I'll appreciate it if you can give me a direction? Like dataset problem or anything else?please.
Dear Simon, I switch our own dataset to the project but we got complete dark images. I mean, We just test the cycleGAN.py and our synthetic images are not what we want. Could you please tell me why this happen? I'll appreciate it if you can give me a direction? Like dataset problem or anything else?please.