search

taki0112 / UGATIT

Official Tensorflow implementation of U-GAT-IT: Unsupervised Generative Attentional Networks with Adaptive Layer-Instance Normalization for Image-to-Image Translation (ICLR 2020)
MIT License
6.17k stars 1.04k forks source link

Crappy results #38

Closed 1mpossibleHacker closed 5 years ago

1mpossibleHacker commented 5 years ago

what are these results

taki0112 commented 5 years ago

Show me what you saw.

1mpossibleHacker commented 5 years ago

In the first try it just deepfried the photo and in the second try it blured it. It does nothing else than ruining photos

tafseerahmed commented 5 years ago

In the first try it just deepfried the photo and in the second try it blured it. It does nothing else than ruining photos

Well, can you show your results and give more details as to how you arrived at that instead of just lambasting at the results? Doesn't help anyone.

tafseerahmed commented 5 years ago

Show me what you saw.

image Hey, I am training the full model on 4 x 2080 Ti's and the dataset and config is same as yours with a minor difference 4k trainA, 4k train B, 100 testA and 100 testB. For how many epochs will I have to train? Also, the Generator loss is way higher than discriminator is that normal? @taki0112

taki0112 commented 5 years ago

@tafseerahmed

  1. For how many epochs will I have to train? I recommend a 50 or 100 epoch
  2. the Generator loss is way higher than discriminator is that normal? Yes, because cam_weight is high.