About the time to end training

[ZhenyuLiu-SYSU]

Hello! Thank you so much for doing such a good job. I would like to know that when do I end the training while training for the best results? And is there any basis for judgment?

Thank you！

[JunlinHan]

Hello Zhenyu, Many thanks for your kind words! For best results, you may record the FID score every epoch. But better FID score does not always suggests better translation. Hence I would suggest referring to FID, KID, LPIPS for a comprehensive evaluation.

Usually I would suggest training for 1M iterations, for instance , to train on a dataset contraining 5000 images, 200 epochs should be fine.

[ZhenyuLiu-SYSU]

Thanks for your reply! And I try to train the code on the winter2summer dataset and found that DCL's FID >SimDCL's FID. This doesn't seem consistent with the fact that DCL mentioned in your paper works better than SimDCL Do you think this is normal？

Thanks！

[JunlinHan]

Thanks for your reply! And I try to train the code on the winter2summer dataset and found that DCL's FID >SimDCL's FID. This doesn't seem consistent with the fact that DCL mentioned in your paper works better than SimDCL Do you think this is normal？

Thanks！

Oh I guess that's OK. Datasets may have some bias, I won't be surprised if the difference is not very significant (let's say, 100 for SimDCL but 130+ for DCL).