Hi, I’m trying to replicate your training method from the paper using the settings from Appendix A and the subsets defined in the Experiments section. While it works very well on ATTGAN, S3GAN, RelGAN, and STGAN, I’ve noticed that on BEGAN, CramerGAN, and others it tends to overfit to fake samples.
If possible, could you share your training scripts or any additional implementation details?
Additionally, in line 17 of networks/LFMSN.py, you have:
ranks = torch.linspace(self.pmin, self.pmin, N, device=x.device)
I believe this should be:
ranks = torch.linspace(self.pmin, self.pmax, N, device=x.device)
Thanks!
Hi, I’m trying to replicate your training method from the paper using the settings from Appendix A and the subsets defined in the Experiments section. While it works very well on ATTGAN, S3GAN, RelGAN, and STGAN, I’ve noticed that on BEGAN, CramerGAN, and others it tends to overfit to fake samples.
If possible, could you share your training scripts or any additional implementation details?
Additionally, in line 17 of
networks/LFMSN.py, you have:I believe this should be:
Thanks!