PeterouZh / CIPS-3D

3D-aware GANs based on NeRF (arXiv).
MIT License
610 stars 60 forks source link

The quality of generated images for FFHQ #31

Closed enisimsar closed 2 years ago

enisimsar commented 2 years ago

Hello,

Thanks for sharing your source code and pre-trained weights. I am trying to generate high-quality images from FFHQ pre-trained model. However, the quality of generated images is not as good enough as stated in the paper. I could not reproduce the results.

I am using the pre-trained weights from here https://github.com/PeterouZh/CIPS-3D/releases/tag/v0.0.2

The command I tried: python exp/cips3d/scripts/sample_images.py --tl_config_file exp/cips3d/configs/ffhq_exp.yaml --tl_command sample_images

Generated images: 0048220334 0038712131 0002215104

Do you have any idea regarding the problem?

xuyanwu commented 2 years ago

I got the same question

PeterouZh commented 2 years ago

Hi,

Thank you for your attention. You can tune the truncation hyperparameter of GANs.

By the way, one way to improve generation quality is to regularize 3D shapes. I am currently training new models and will release code and models when the effect is stable.

PeterouZh commented 2 years ago

I got the same question

I think the reason for the poor generation effect is the poor quality of the underlying 3D shape. I am working to solve this problem.

enisimsar commented 2 years ago

Does truncation correspond to psi? @PeterouZh

PeterouZh commented 2 years ago

Does truncation correspond to psi? @PeterouZh

Yes.

enisimsar commented 2 years ago

It's better now, psi=0.45

0023813720 0015039847 0002215104

Thanks!