Different results in your paper and the result url you provide

[yunniw2001]

Describe the bug

in the url result you provide https://drive.google.com/drive/folders/1dK3oIB2UvswlTtnIS1iHfx4s57MevWdZ?usp=sharing, the result is
.
but in your paper, the metric result is

there is a huge difference, especially for PSNR and LPIPS, why?

Reproduction

https://drive.google.com/drive/folders/1dK3oIB2UvswlTtnIS1iHfx4s57MevWdZ?usp=sharing

Logs

No response

System Info

CUDA 11.7

Who can help?

No response

[juxuan27]

Hi, @yunniw2001 ! Sorry for the confusing results. The results in the paper is a model trained with fewer steps compared to the released model. To release a stronger model for community, we train the model for more steps and release the better version. However, for fair comparison in ablation, we did not revise the result in the paper. The released model has a better result on most metrics compared to the reported ones, and also have a much better visual quality.