yuanzhi-zhu / DiffPIR

"Denoising Diffusion Models for Plug-and-Play Image Restoration", Yuanzhi Zhu, Kai Zhang, Jingyun Liang, Jiezhang Cao, Bihan Wen, Radu Timofte, Luc Van Gool.
https://yuanzhi-zhu.github.io/DiffPIR/
MIT License
383 stars 31 forks source link

what is the best parameter for different tasks according to the paper data? #43

Open AKA-hhh opened 4 days ago

AKA-hhh commented 4 days ago

where is the Appendix B,how to see it

yuanzhi-zhu commented 4 days ago

https://arxiv.org/pdf/[2305.08995](https://arxiv.org/pdf/2305.08995)

hey, pleas check out the paper

AKA-hhh commented 3 days ago

Thank you,I get it

AKA-hhh commented 3 days ago

If I set the sigma=9. , zeta = 0.5 and save the estimated image in main_ddpir_sisr.py, Can I get the best result?

AKA-hhh commented 3 days ago

On the ImageNet dataset

yuanzhi-zhu commented 3 days ago

That's the best result I got, you are welcome to tune these hyperparameters yourself.