Can not get reasonable results.

universome / epigraf

[NeurIPS 2022] Official pytorch implementation of EpiGRAF

https://universome.github.io/epigraf

150 stars 6 forks source link

Can not get reasonable results. #6

Open Talegqz opened 2 years ago

Talegqz commented 2 years ago

Thanks for your excellent work! I trained the model following python src/infra/launch.py hydra.run.dir=. exp_suffix=<EXPERIMENT_NAME> dataset=<DATASET_NAME> dataset.resolution=<DATASET_RESOLUTION> model.training.gamma=0.1 in FFHQ dataset, but I can not get reasonable results. Could you please provide configs for your training? Thanks for your help!

universome commented 2 years ago

Hi, thank you! Could you please tell whether you use the FFHQ dataset with pre-computed camera poses or without them? The FFHQ dataset which we use is located in configs/dataset/ffhq_posed.yaml

universome commented 2 years ago

Could you please tell in what way does the model fail for you? Does it fail to learn the texture or structure or both? If that's possible, can you please share your FID@2k plot?

bomcon123456 commented 1 year ago

I'm having the same issue, using the FFHQ dataset you provided and the commands you showed in this Issue, FID_2k consistently stays at 433 for 5-7 first evaluation

universome commented 1 year ago

It's hard to tell what is going wrong without digging deeper into your setup, for which I need more details. The Carver repo reproduced several 3D GAN papers, including ours, and they seem to have even better FID on FFHQ. I have not tried running their code, but maybe it's worth taking a look.