liu-vis / dualresidualnetworks Goto Github PK

Thanks for your amazing work, my current work need to use your motion_blur.py, it can work when i use CarDataset,but how can i run my own image for test?

I want to test the results on the raindrop removal task, I execute [python raindrop.py], but the result is as following.
And the PSNR and SSIM is not correct:
psnr: 10.630165165245023
ssim: 0.25498810946974426

Thanks for good work.but I have a question,why you do not use norm layer in DuRN-US network in haze removal and use bacthnorm in rain steak removal?

Hi你好，在阅读论文的过程中，我对这句话不是很理解：We also choose the kernel size and dilation rate for each DuRB so that the receptive field of convolution in each DuRB grows its size with L.
我尝试自己计算每一层的感受野，并没有发现有倍数关系。能麻烦您发我一份补充材料吗，下面是我的邮箱：[email protected]
不胜感激。

Why are the image bit depths in the BSD500 gray dataset all 32, rather than 8 (single channel)? Aren't all grayscale denoising true single channel Grayscale?

Looking forward to your reply！

Hello, I try to run your code. But there is an error message.

And I want to know why.

I am wondering about the experimental environments

1. The number of gpus for training
2. The gpu you used for training

Thank you!

Hi, Dr Liu
Thanks for your great job!
I check the testing codes provided by this work， different tasks use different PSNR computation.
E.g.
(1) Denosing: the average channels of PSNR
(2) Deraining Drop: use the Y channel of images.
Could you please tell me that the reason about these implementations?
Thanks a lot!

Dear authors, hi.
There are some questions I want to ask you:
Your readme writes the environment is python3.7 + PyTorch 1.2, but when I clone your code, I find it has no different with the preceding version of code, which is python2.7 + PyTorch0.31;
Can you update your code version as you say?

Hi,

Thanks for you inspiring work. It would be great if you could share the pre-trained weights for each removals for us.

Thanks

I used the code (trained/test) and BSD_gray you provided. The results: 28.43 / 0.8130(noise level = 30), 26.29 / 0.7297(noise level = 50), 24.99 / 0.6685(noise level = 70). It has a bit gap than yours. Is there anything special about your training? Or is this result acceptable?

Thanks for your interesting work! I want to ask about the task of additive Gaussian noise removal. Are the different levels of noise dataset ( i.e., 30, 50, 70) trained in a model together or trained separately for each level?

Hi, thanks for your excellent jobs of restoration of multiple distortions .Have you finished the training code ? Thank you very much