phoenix104104 / lapsrn Goto Github PK

Some thing wrong when I use 3 channel bilinear kernel. Following is the 2 upsampling results and original images.


My init code is followings:

function net = add_USN_noise(net, layername, varname, parname, filter_size, scale)
% --------------------------------------------------------------------
crop_default = 1;
stride_default = scale;
net.addLayer(layername, ...
dagnn.ConvTranspose('upsample', stride_default,'crop', crop_default,'hasBias',false), ...
varname{1}, varname{2}, parname);

f = net.getParamIndex(parname) ;
filters = single(bilinear_kernel(4,3,3));
net.params(f).value = filters ;
net.params(f).learningRate = 1;
net.params(f).weightDecay  = 1;
net.params(f).trainMethod = 'adam';

end

Do you have any idea of this problem? Thank you.

Thanks for code publicly!
I encountered a version problem that as follows:
"No property ignoreAverage for a layer of type dagnn.vllab_dag_loss."
After searched in the internet, maybe caused by the different version of MatConvNet. So could u tell me what is the version of MatConvNet used in this code?

Hi

Thanks for sharing. Unfortunately I don't have MATLAB. Wondering if it is possible to have Python version ? Thanks

Error using vl_compilenn>search_cuda_devkit (line 658)
Could not find a valid NVCC executable\n

Error in vl_compilenn (line 270)
if isempty(opts.cudaRoot), opts.cudaRoot = search_cuda_devkit(opts) ; end

Error in install (line 17)
vl_compilenn('enableGPu', true, 'enableCudnn', true);

i want to apply your model to some gray scale images, i replicate the gray scale image 3 times to a new matrix, but this doesn't work out? any idea how to do it ?

Excuse me, thank you very much for your code. I would like to ask how to calculate the training loss? How did the HR picture come from?
Looking forward to your busy reply, thank you!

Hi,
System configuration:

• Windows 10
• Matlab 2017a
• Visual Studio 2013

>> demo_LapSRN
Warning: Name is nonexistent or not a directory:
D:\Users\user\sr\LapSNR\LapSRN-master\LapSRN-master\matconvnet\matlab
> In path (line 109)
In addpath (line 88)
In demo_LapSRN (line 25)
Load pretrained_models\LapSRN_x4.mat
Undefined function or variable 'dagnn.vllab_dag_loss'.

Error in dagnn.DagNN.loadobj (line 26)
block = constr() ;

Error in demo_LapSRN (line 33)
net = dagnn.DagNN.loadobj(net.net);

Any help please?
Thanks.

Thank you for your code. I have read your thesis, saying that you can perform real-time super-resolution on the video. How do you implement this part?

It is successful to run install.m in matlab, but error occurs as follow:
**_>> demo_LapSRN
Load pretrained_models\LapSRN_x4.mat
error dagnn.Layer/load (line 200)
No property ignoreAverage for a layer of type dagnn.vllab_dag_loss.
error dagnn.DagNN.loadobj (line 28)
block.load(struct(s.layers(l).block)) ;

error demo_LapSRN (line 33)
net = dagnn.DagNN.loadobj(net.net);_**

waiting your answer...

I followed your instructions but I get this error.
The error is "Undefined function or variable 'dagnn.vllab_dag_loss'".

Thank you for your help . @phoenix104104

hello, can I train this code in CPU not GPU?
if I train it in CPU,which part of code do I need to modify?
And　how many days do I need to train?

When I try to load the pre-trained_models, a error always occurs and I don't know how to fix it. I have checked that this is not caused by path but it seemed there are some problems in these .mat files or in the code.

Thanks for your codes!

However, I found that the link http://vllab1.ucmerced.edu/~wlai24/LapSRN/results/SR_testing_datasets.zip returns 403, so I cannot download it. Could someone fix this error or share their test data?

Thank you for your code. I have a question that how to calculate PSNR of Set5 and Set14? Could you help me?

The url of the data set seems to be invalid. Can you provide usable data set url? Or is there an alternative?

The model is trained on Y-channel of image, what should i do if i want to generate a RGB image according
to Y-channel of an image ?
Thank you!

Firstly. thank you for your code and paper.
As a beginner, I don't know 291 images in the paper are input LR images or the ground truth HR images. But whatever they are, they need the corresponding HR/LR images when being trained. I downloaded the Training Datasets (86.2 MB) in the link(http://vllab.ucmerced.edu/wlai24/LapSRN/). In the folder, there are 291 images in T91 and BSD200 not 291*2 images.
Looking forward to the recovery of you or others. Thanks.

Thanks for your code, I can't reproduce the result with my own model when training the T91 BSDS200 and General100 images. The PSNR of Set5(x4) only 29.7. but the papers PSNR of Set5(x4) is 31.54. Can you provide some details about the training parameters ?

How many epoch can be trained to achieve the effect of the paper？
Thank you for your answer.

Hi,
as the title mention, how can I get the dataset your provided?
thanks.

thanks for your code.As you mentioned in the paper,the training dataset is t91 and bsd 200.However,the dataset is t91+bsd200+g100... So ,what is the result for your paper? Is is consist of General 100?