Closed hhcs9527 closed 3 years ago
Hi
I am able to train the models on pytorch 1.8.1. However, the pretrained weights were provided for pytorch 1.1 and you might not be able to test using another version.
Thanks
Hi
Thanks for the useful information! I have another question for the evaluation of PSNR in image_utils.py in utils file.
Why do you calculate the difference between prediction and target by imdff = torch.clamp(prd_img,0,1) - torch.clamp(tar_img,0,1)?
Thanks
It is needed to compute the PSNR
Thanks for the formula for calculating PSNR. I am wondering why do you perform torch.clamp(prd_img,0,1) before you perform imdff = torch.clamp(prd_img,0,1) - torch.clamp(tar_img,0,1). In the formula that you provided, there is no such operation.
Could you tell me the reason for performing this torch.clamp(prd_img,0,1)?
Thanks
hello, how can i test the real rain12 dataset?
Hi @hhcs9527
The outputs from the model might be a little outside the range of [0,1]. Hence, in order to make sure that the correct PSNR is calculated, we clamp the outputs between [0,1].
Thanks
Hi @hellogry
As per Table 1 in our paper, Rain12 is included in the train set. Hence we do not test on it.
If you still want to test on Rain12 dataset, you can put the images at this path https://github.com/swz30/MPRNet/blob/435f483b6433a6cf8d43f4df9c20eebba47587bb/Deraining/test.py#L25 and add Rain12 at https://github.com/swz30/MPRNet/blob/435f483b6433a6cf8d43f4df9c20eebba47587bb/Deraining/test.py#L44
Thanks
Thanks! The last question, I am wondering how do you measure the PSNR/SSIM(PSNR for 32.66 in your paper)? By running test.py, then run evaluate_GOPRO_HIDE.m?
Thanks
Hi @hhcs9527
Running test.py
would generate the predictions. After that, running evaluate_GOPRO_HIDE.m
would evaluate the PSNR/SSIM.
Thanks
ok, I think I've done my questions, thanks!
Hi! Thank you so much for your work! I am wondering if anyone can run the model on PyTorch 1.8.1(stable version)?
Thanks