JingyunLiang / SwinIR

SwinIR: Image Restoration Using Swin Transformer (official repository)

Home Page:https://arxiv.org/abs/2108.10257

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool


vztu opened this issue · comments


Thanks for this great work! Could you provide the FLOPs/MACs of your SwinIR model?

We have already provided the FLOPs in README.md.


Method Training Set Training time
(8 GeForceRTX2080Ti
batch=32, iter=500k)
on Manga109
Run time
(1 GeForceRTX2080Ti
, on 256x256 LR image)*
#Params #FLOPs Testing memory
RCAN DIV2K 1.6 days 31.22/0.9173 0.180s 15.6M 850.6G 593.1M
SwinIR DIV2K 1.8 days 31.67/0.9226 0.539s 11.9M 788.6G 986.8M

Hi, @JingyunLiang ,
The FLops in the above table are calculated with the input resolution=256x256?

Runtime, #FLOPs and testing memory are all calculated on 256x256 LR image for X4 SR.

Hi, @JingyunLiang, how about the FLOPs for the denoising model on 256x256 inputs?

It is similar to the SR model because the main difference is the upsample module. The exact FLOPS for denoising on 256x256 models is 787.993G.

Feel free to open it if you have more questions.

ezoic increase your site revenue