search

zhaoyuzhi / Self-Guided-Network-for-Fast-Image-Denoising

The PyTorch implementation of ICCV 2019 paper SGN
44 stars 6 forks source link

Something about FullResDenoisingDataset #7

Open YDDDDG opened 4 years ago

YDDDDG commented 4 years ago

Hi, I ran _validationfolder.py with model provided by you(SGN_iter100000_bs8_mu0_sigma30.pth) and my own testing set(DIV2K_valid_HR). I found that the output images are randomly cropped from the full-resolution. I want to store the full-resolution-denoised images，so I reviewed your code and modify this:

  `  testset = dataset.DenoisingDataset(opt)` in _validation_folder.py_

to this: testset = dataset.FullResDenoisingDataset(opt)

(I thought it was the key to get full-res images)

however, something wrong happened:

Traceback (most recent call last): File "validation_folder.py", line 67, in <module> for batch_idx, (noisy_img, img) in enumerate(dataloader): File "E:\Anaconda3\Anaconda\envs\Experiment\lib\site-packages\torch\utils\data\dataloader.py", line 363, in __next__ data = self._next_data() File "E:\Anaconda3\Anaconda\envs\Experiment\lib\site-packages\torch\utils\data\dataloader.py", line 403, in _next_data data = self._dataset_fetcher.fetch(index) # may raise StopIteration File "E:\Anaconda3\Anaconda\envs\Experiment\lib\site-packages\torch\utils\data\_utils\fetch.py", line 44, in fetch data = [self.dataset[idx] for idx in possibly_batched_index] File "E:\Anaconda3\Anaconda\envs\Experiment\lib\site-packages\torch\utils\data\_utils\fetch.py", line 44, in <listcomp> data = [self.dataset[idx] for idx in possibly_batched_index] File "D:\paper_experiments\Self-Guided-Network-for-Fast-Image-Denoising\SGN\dataset.py", line 100, in __getitem__ img = cv2.resize(img, (W_out, H_out)) cv2.error: OpenCV(4.4.0) C:\Users\appveyor\AppData\Local\Temp\1\pip-req-build-52oirelq\opencv\modules\imgproc\src\resize.cpp:3932: error: (-215:Assertion failed) inv_scale_x > 0 in function 'cv::resize'

by the way, i met similar cv2 error when i ran validation.py

So, Is it wrong to modify code from DenoisingDataset to FullResDenoisingDataset in order to get full-res outputs?

If i am wrong ,what should i do to get full-resolution outputs?

Hope for your reply !

zhaoyuzhi commented 4 years ago

I am not sure for your provided wrong log. It seems that there are some errors at the place calling ''cv2.resize''.

YDDDDG commented 4 years ago

ok...I knew something wrong with "cv2.resize". Leave my bug aside, could you tell me which part of the code should I modify to produce full resolution denosing results like pictures in Gu's paper? Thanks a lot !