search

sanghyun-son / EDSR-PyTorch

PyTorch version of the paper 'Enhanced Deep Residual Networks for Single Image Super-Resolution' (CVPRW 2017)
MIT License
2.43k stars 669 forks source link

How to train MDSR? #229

Open Atumyugi opened 4 years ago

Atumyugi commented 4 years ago

I use demo.sh ->python main.py --template MDSR --model MDSR --scale 2+3+4 --n_resblocks 80 --save MDSR --reset --save_models

but this model only train X2 model, I want to know why. The PSNR in X2model become better and better, but X3 X4 model cant change thank you!

Ccbov commented 4 years ago

@Atumyugi Hello, I meet the same problem,could you please give some advise?

Atumyugi commented 4 years ago

I think this problem is from pre train model, if you don't have the pre train model for users, you won't get the better performance.

Ccbov commented 4 years ago

@Atumyugi 谢谢，我会试一下！

harshraj22 commented 3 years ago

Though I am too late, commenting in case others face the same issue: The reason for this is this code: https://github.com/sanghyun-son/EDSR-PyTorch/blob/9d3bb0ec620ea2ac1b5e5e7a32b0133fbba66fd2/src/model/edsr.py#L24 Only first scale model is trained at a time. xD

pzq-xjtu commented 3 years ago

@Ccbov Hey brother, have you resolved this problem yet? It seems that it's no helpful to use the pre_train. Looking forward to your reply!

pzq-xjtu commented 3 years ago

@Atumyugi It seems that it's no helpful to use the pre_train. The MDSR model only train X2 model too. How did you deal with the problem? Looking forward to your reply!