search

sian-wusidi / StegoGAN

Official code for StegoGAN: Leveraging Steganography for Non-bijective Image-to-Image Translation
17 stars 0 forks source link

unpaired image-to-image translation #2

Open kutieling opened 3 months ago

kutieling commented 3 months ago

Thank you for your sharing, your work is very wonderful. I would like to ask if your work can do unpaired image to image translation? That is, the number of images in trainA and trainB is different, thank you.

sian-wusidi commented 3 months ago

Our current model is designed for unpaired data as long as the numbers of images from trainA and trainB are the same. In your case, just get equal numbers for trainA and trainB.

guyleaf commented 3 weeks ago

Hi, I think the different number of images in A and B should be fine. Because in the unaligned_dataset.py, the __getitem__ implementation is totally the same as in CycleGAN one unless there are some changes that have to be equal outside of the dataset.

The author of CycleGAN answer the similar question before, https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/issues/1365#issuecomment-1012608572.