an issue about the results of paper model.

cvlab-yonsei / PISR

An official implementation of "Learning with Privileged Information for Efficient Image Super-Resolution" (ECCV2020) in PyTorch.

243 stars 30 forks source link

an issue about the results of paper model. #7

Open Lininggggggg opened 3 years ago

Lininggggggg commented 3 years ago

hello, i have an issue is about the compared model of FSRCNN(origin), and your model of FSRCNN training by PISR. the origin model of fsrcnn is trained without div2k, but your model has. so, maybe your method is no good as your paper shows, because if the origin model of fsrcnn is trained with the same trainset like yours, it will better than before, right? if i am wrong, could you tell me why, please? thanks you very much!

Lininggggggg commented 3 years ago

emm，ok, i am sorry. i am wrong. you had deal with this problem in the paper. thanks you for your excellent work!

Lininggggggg commented 3 years ago

sorry, i still have another problem. the multiadds of fsrcnn is false, and the reason maybe is the package you used to count flops is error.(the deconv stride is the same as upscale, so the correct flops should be 0.738259GMac.) you can check it by the origin paper or use the package like "from thop import profile".

Lininggggggg commented 3 years ago

i hope you can reply me. thank you very much!!!