cszn / dpsr

Deep Plug-and-Play Super-Resolution for Arbitrary Blur Kernels (CVPR, 2019) (PyTorch)

Kai Zhang

[Homepage] [Google Scholar] [ResearchGate] [知乎]

I am currently a postdoctoral researcher at Computer Vision Lab, ETH Zurich, Switzerland.

⚡ News:

Our new work SCUNet for practical image denoising.

⚡ News

We released the testing code of BSRGAN!

🌱 My Repositories

SCUNet	BSRGAN	KAIR	USRNet	DPIR	IRCNN	SRMD	DPSR	DnCNN	FFDNet

The pytorch training and testing codes of SRMD, DnCNN, FFDNet can be found in KAIR.

dpsr's Issues

Estimated Blur Kernel

Hello，which method did you apply to estimate blur kernel?

About dataset

How to get the low-resolution picture under your code file by using real pictures, as you show the test_01_m_19 picture under x4_m folder and how to get the corresponding .mat file.I want to simulate the entire CBSD68 data set.
Thank you for your answer.

Release training codes?

the demo codes can generate amazing SR image, is there a plan to release training codes?

main_test_dpsr problems

when I use 'python main_test_dpsr.py --model_zoo dpsr_x2 --testset set5' to run the testing, image resolution hasn't changed and also after changing sr to dn, nothing changed. Did I set something wrong?
thank you

|--model_zoo
|--dpsr_x2
|--dpsr_x3
|--dpsr_x4
|--dpsr_x4_gan
|--testset
|--set5
|--srbsd68
|--results
|--set5_dpsr_x2

|--main_test_dpsr.py

The kernel appears to have died. It will restart automatically.

Hi, when I run demo_test_dpsr.py, I get an LR image, then the following error message appears: "The kernel appears to have died. It will restart automatically."

When I run demo_test_dpsr.py from the terminal, the same thing basically happens. I see an LR image, then it appears to be running forever.

serious artifacts with the results

I generate the blur kernel with matlab code

mu=[0,0];
[X,Y]=meshgrid(-10:1:10,-10:1:10);
j = 9;
out_path = 'kernels21/G/';
if isdir(out_path)==0
mkdir(out_path)
end
for i = 0:0
for k = 2.5:0.5:6
sigma = [k i ; i k];
kernal = mvnpdf([X(:) Y(:)],mu,sigma);
k = reshape(kernal, size(X));
save([out_path, int2str(j), '.mat'], 'k')
j = j+1;
end
end

and blur the image with opencv function filter2D and input the generated image and kernel lable to test. But I get results with serious artifacts, examples are as follows


I don't know why it happens.

Test model problem

Dear Mr. Author@cszn, the work of DPSR is of great significance and gives me great inspiration. when I fully followed the requirements of your training environment and used the "main_train_dpsr.py" provided by you to train the model, I put the generated model into the "main_test_dpsr.py", but reported the following Error:
RuntimeError: Error(s) in loading state_dict for SRResNet:
Missing key(s) in state_dict: "model.2. Weight ", "model.2. Bias ", "model.5. Weight ", "model.5.
Unexpected key(s) in state_dict: "model.3. Weight ", "model.3. Bias ", "model.6. Weight ", "model

When I use your trained "dpsr_x2.pth","dpsr_x3.pth" and "dpsr_x4.pth" to check, but there is no problem, I have tried many ways in the inquiry and failed to solve, Do you have any good suggestion or solutions to my problem

this is the error picture：phttps://imgchr.com/i/JjAZVK

请问该如何加载tensorflow代码训练出的模型呢

如题。

Training code

Hi, is there a plan to release training codes?

How do you derive formula 9?

can you tell us how do you derive formula 9 with FFT ? thanks very much

Use

How can I test with my own images? It's not clear to me how it's useful without the matlab files

C++ version

Dear, Everybody.
This is great project. Can you provide c++ version of this project?
Because of i used win10 and Visual Studio 2015 or 2017 or 2019.
Kind regards.

Can you tell the source code of the training?

The test result show artfacts

@cszn
I test the testsets/BSD68 dataset, but my test results show artfacts. Can you give me some advice.Here are my test results.

利用自己的数据去测试，效果很差，这是什么原因呢

利用自己的数据去测试，效果很差，这是什么原因呢？这个算法的适合什么样的图片

How do I generate blur image using your kernels?

Hi, thanks for the ready-to-go python script. It makes my life so much easier. I was able to successfully replicate your study using your existing photos.

However, I was not able to deblur my own images. To blur my photo, I used your motion kernel (below) on a third-party script.

Is there a way to generate blurred images using your code in python?