yossigandelsman/doubledip

Official implementation of the paper "Double-DIP: Unsupervised Image Decomposition via Coupled Deep-Image-Priors"

doubledip's Issues

how to use the code

Hi, author:
I don't know how to use the code. Can you write a demo to run this code? Thank you very much.

A question about exclusion loss

Thanks for the amazing work and code provided.
I found that the exclusion loss in this work seems not so similar to the provided reference paper, I can't understand how the exclusion loss works in this paper, may someone explain it?
Thanks to all who help me with this issue.

why airlight estimated is not uniform ?

This is the final airlight estimated

Cannot reproduce zebra segmentation example with supported codes.

Hi :) First, I'd like to appreciate authors to present sample codes!
As I'm a lot interested in "Double-DIP' structure for image segmentation,
I'm trying to reproduce the results reported in the original paper!
Especially, I'm trying to reproduce zebra segmentation, but I couldn't make it with provided sample codes ("segmentation.py"), same model configurations, network structures, etc.
Could you please check them if the sample codes properly generate reported layer images and the binary mask?
Thanks in advance :)

CUDA Runtime Error

Error when running watermarks_removal.py:

C:\.....\DoubleDIP\DoubleDIP-master>python watermarks_removal.py Traceback (most recent call last): File "watermarks_removal.py", line 567, in <module> remove_watermark_many_images(['f1'], [im1], "fotolia_many_images") File "watermarks_removal.py", line 544, in remove_watermark_many_images s.optimize() File "watermarks_removal.py", line 440, in optimize self._optimization_closure(j, step) File "watermarks_removal.py", line 502, in _optimization_closure self.watermark_net_output = self.watermark_net(watermark_net_input) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\module.py", line 550, in __call__ result = self.forward(*input, **kwargs) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\container.py", line 100, in forward input = module(input) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\module.py", line 550, in __call__ result = self.forward(*input, **kwargs) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\container.py", line 100, in forward input = module(input) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\module.py", line 550, in __call__ result = self.forward(*input, **kwargs) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\container.py", line 100, in forward input = module(input) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\module.py", line 550, in __call__ result = self.forward(*input, **kwargs) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\container.py", line 100, in forward input = module(input) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\module.py", line 550, in __call__ result = self.forward(*input, **kwargs) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\container.py", line 100, in forward input = module(input) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\module.py", line 550, in __call__ result = self.forward(*input, **kwargs) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\container.py", line 100, in forward input = module(input) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\module.py", line 550, in __call__ result = self.forward(*input, **kwargs) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\container.py", line 100, in forward input = module(input) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\module.py", line 550, in __call__ result = self.forward(*input, **kwargs) File "C:\Users\youre\Documents\machine_learning\DoubleDIP\DoubleDIP-master\net\layers.py", line 54, in forward inputs.append(module_(input_)) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\module.py", line 550, in __call__ result = self.forward(*input, **kwargs) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\container.py", line 100, in forward input = module(input) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\module.py", line 550, in __call__ result = self.forward(*input, **kwargs) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\container.py", line 100, in forward input = module(input) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\module.py", line 550, in __call__ result = self.forward(*input, **kwargs) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\conv.py", line 349, in forward return self._conv_forward(input, self.weight) File "C:\Users\youre\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\conv.py", line 346, in _conv_forward self.padding, self.dilation, self.groups) RuntimeError: CUDA out of memory. Tried to allocate 2.00 MiB (GPU 0; 4.00 GiB total capacity; 815.82 MiB already allocated; 0 bytes free; 2.42 GiB reserved in total by PyTorch)

On the web I found solutions like reducing the batch size, but in this case I can not find anything I could change to a lower value that might not take that much memory.

Any suggestion I could do?

The code and the example image for Image Segmentation

Hi authors. I am interested in using the code for image segmentation. Could you provide the example image and the whole code for segmentation. There seems to be that the main function for segmentation is lost. Thanks in advance.

Could you provide a self-contained example for segmentation?

Hi,

Interesting work! I would like to follow your work. But I did not find an example for the segmentation. Can you help me?

No driver script for segmentation

about dehazing

in dehazing.py, What is the meaning of gt_ambient? and why gt_ambient=np.array([0.5600084 , 0.64564645, 0.72515032]), please...

Running watermark removal script

Hi, @yossigandelsman !

Thanks for sharing the code!

I am trying to run the watermark removal script with the following pip packages installed:
torch==1.4.0
scikit-image==0.16.2
torchvision==0.5.0
sk-video==1.1.10

However, when running it on a server with a GPU, it outputs 3 original images to output/ folder, hangs and never stops. What could be the issue?

(Running on a laptop I get "CUDA driver insufficient", which I do expect and which I hope means that my server is actually using CUDA).

Thanks in advance

Worse results in segmentation task compared to Figure 4 in paper.

Hi,
I try to use the provided setup of segmentation for the bird segmentation task.
Saliency maps are processed by the given "bg_fg_prep.py" before being fed to segmentation.py.

It seems like the generated mask has a very rough estimation of the object boundaries, compared to the sharp and crisp mask in the paper (Figure 4).

Solider image has worse result and the network cannot separate soldiers and background.

Could you please tell me what are the potential problems and some advice to improve it?
Thanks.

posted by mistake

posted by mistake

readme

Can you provide a detailed readme? thanks

requirements.txt file would be helpful.

It is quite painful to install the packages one by one :(

The code for Video Transparency Separation

Interesting and amazing project!

Would you mind providing any information about the code for video transparency separation, please?

I am new with python and unable to execute this program

Hello, I installed python on my PC, set up environmental variables and try to run dehazing.py, but get:

Traceback (most recent call last):
File "c:\Users\Matas Liutikas\source\repos\DoubleDIP\dehazing.py", line 3, in
from cv2.ximgproc import guidedFilter
ModuleNotFoundError: No module named 'cv2.ximgproc'

I cannot find anything online and I assume there are more dependencies I have to install. Is there a way to do it easily?

RuntimeError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 2.00 GiB total capacity; 1.30 GiB already allocated; 13.37 MiB free; 29.76 MiB cached)

Traceback (most recent call last):

File "", line 1, in
runfile('E:/Projects/AI/DoubleDIP-master/watermarks_removal.py', wdir='E:/Projects/AI/DoubleDIP-master')

File "C:\Users\Kinjal\Anaconda3\lib\site-packages\spyder_kernels\customize\spydercustomize.py", line 827, in runfile
execfile(filename, namespace)

File "C:\Users\Kinjal\Anaconda3\lib\site-packages\spyder_kernels\customize\spydercustomize.py", line 110, in execfile
exec(compile(f.read(), filename, 'exec'), namespace)

File "E:/Projects/AI/DoubleDIP-master/watermarks_removal.py", line 567, in
remove_watermark_many_images(['f1', 'f2', 'f3'], [im1, im2, im3], "fotolia_many_images")

File "E:/Projects/AI/DoubleDIP-master/watermarks_removal.py", line 544, in remove_watermark_many_images
s.optimize()

File "E:/Projects/AI/DoubleDIP-master/watermarks_removal.py", line 440, in optimize
self._optimization_closure(j, step)

File "E:/Projects/AI/DoubleDIP-master/watermarks_removal.py", line 503, in _optimization_closure
self.mask_net_output = self.mask_net(mask_net_input)

File "C:\Users\Kinjal\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 493, in call
result = self.forward(*input, **kwargs)

File "C:\Users\Kinjal\Anaconda3\lib\site-packages\torch\nn\modules\container.py", line 92, in forward
input = module(input)

File "C:\Users\Kinjal\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 493, in call
result = self.forward(*input, **kwargs)

File "C:\Users\Kinjal\Anaconda3\lib\site-packages\torch\nn\modules\container.py", line 92, in forward
input = module(input)

File "C:\Users\Kinjal\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 493, in call
result = self.forward(*input, **kwargs)

File "C:\Users\Kinjal\Anaconda3\lib\site-packages\torch\nn\modules\container.py", line 92, in forward
input = module(input)

File "C:\Users\Kinjal\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 493, in call
result = self.forward(*input, **kwargs)

File "C:\Users\Kinjal\Anaconda3\lib\site-packages\torch\nn\modules\container.py", line 92, in forward
input = module(input)

File "C:\Users\Kinjal\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 493, in call
result = self.forward(*input, **kwargs)

File "C:\Users\Kinjal\Anaconda3\lib\site-packages\torch\nn\modules\container.py", line 92, in forward
input = module(input)

File "C:\Users\Kinjal\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 493, in call
result = self.forward(*input, **kwargs)

File "C:\Users\Kinjal\Anaconda3\lib\site-packages\torch\nn\modules\container.py", line 92, in forward
input = module(input)

File "C:\Users\Kinjal\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 493, in call
result = self.forward(*input, **kwargs)

File "C:\Users\Kinjal\Anaconda3\lib\site-packages\torch\nn\modules\padding.py", line 171, in forward
return F.pad(input, self.padding, 'reflect')

File "C:\Users\Kinjal\Anaconda3\lib\site-packages\torch\nn\functional.py", line 2817, in pad
ret = torch._C._nn.reflection_pad2d(input, pad)

RuntimeError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 2.00 GiB total capacity; 1.30 GiB already allocated; 13.37 MiB free; 29.76 MiB cached)

Early stopping rule

Greetings!

I'm highly interested in your work, especially the image dehazing part.

I tried to use the results you posted on the project page to compare the results of mine,
but I found that those show different resolutions compared to the original hazy inputs.

I wonder if I could know the early stopping rule to reproduce the in-paper results.
Or if that's impossible could I get the original results of your paper?

Thanks.

Bad performance about the segmentation part of this work!

完全复现不出论文里的结果，这作者纯纯乱写的，大家别信，别引用对比这篇工作了
Yossi Gandelsman, shame on u!!!!!!!!

Image segmentation

Are there plans to release an image segmentation example ?