c-yn / adair Goto Github PK

Hello,

Thank you for sharing this interesting study.

I have one question about the denoise dataset. In the "denoise.txt" file, the BSD400 data file names appear to differ from the acutal image names (e.g. 'a117025.jpg' in denoise.txt / '117025'jpg' in the actual image name). Could you please clarify if the BSD400 image are used for training in the All-in-one IR??

Thank you for your help.

Sorry to bother you in your busy schedule! I wanna ask do you have a "demo.py" file similar to the one provided by "PromptIR" that I can use to test other images? Thanks a lot!

Hi @c-yn,

Hope this message finds you well!

I am writing to inquire about the performance of the PromptIR under the 5-Degradations setting, not sure if you have run this experiments?

Bests and many thanks,

I found a problem, since I was training with DDP, that would indicate the presence of parameters that were not involved in the training. Through my investigation, self.score_gen and self.conv are unnecessary, and these problems are not serious. But the most important thing is that self.rate_conv will not participate in the gradient calculation, because the operation of generating mask with threshold is not differentiable.

Hello,

terrific work, is available the code for running inference? are the model weights available? Could be possible to run it without the need for training?

Thank you for your help

您好，在3090可以运行，但切换到4090上就出现RuntimeError: cuFFT error: CUFFT_INTERNAL_ERROR，请问这个该如何解决？
期待您的回答，谢谢您！

In the top super-row of the five-task table mentioned in the paper, was the accuracy of the single-task models trained with your training settings? I did not find the accuracies provided in your paper in these papers.

Hi @c-yn ,

Thanks for your interesting explorations in IR from the frequency perspective.

May I know how many GPUs (And which type you used) are needed for training the 3-task setting and the 5-task setting ?

Especially, how long it costs to train a full model with your GPU setting.

Thanks in advance, this would help me a lot as a reference to follow this nice work.

Best and have a nice day,

where is FMiM

Excellent work! I wanna ask if you will release all-in-one models? I want to use it to test the effect. Thank you so much!!!