search

Zhendong-Wang / Diffusion-GAN

Official PyTorch implementation for paper: Diffusion-GAN: Training GANs with Diffusion
MIT License
625 stars 67 forks source link

FID discrepancy #30

Open IFeelBloated opened 1 year ago

IFeelBloated commented 1 year ago

The stylegan2 FID on FFHQ reported in the paper (4.41) is significantly worse than the official evaluation (2.70), is there an explanation for this?

Zhendong-Wang commented 1 year ago

We trained the model on FFHQ(1024x1024, 70k). The performance matches with the values 5.28/4.30 reported in the Figure 7 of ADA paper, "Training Generative Adversarial Networks with Limited Data".

IFeelBloated commented 1 year ago

so all models didn't use horizontal flip?

Zhendong-Wang commented 1 year ago

Sorry that I missed the email notification. x-flip is applied. ADA contains flip and rotation in the augmentation.