rimchang / realblur Goto Github PK

Hi,
It looks like links to RealBlur dataset README.md is broken.
I tried ones in the project homepage (http://cg.postech.ac.kr/research/realblur/), but they are not accessible either.

Would you be able to fix broken links?

Hi,
I can't download RealBlur Dataset from your homepage. Has it been published yet?

Hi,
Could you describe how to estimate kernel size in section. 2 in the supplementary material of your paper?

RuntimeError: invalid hash value (expected "520b38e4", got "65ebaeb82ed6afa16a78f544f387c26cab42773f89a7b0f2084e8fdbf5f4b69a")

I'd like to know how to fix this problem . I have tried pip install --upgrade pretrainedmodels, but still error.

It is not well organized and i have to pay more time rewirte the evaluation code. Would you please fix it or even give us a little suggestion

Hi All,

I am trying to download the RealBlur dataset from the dataset webpage. However, it hangs and fails. I wonder does there exist any other download source?

Thanks and Regards,
Mekhak

Hi, hope this finds you well. I find your work really awesome! I am recently trying to reproduce the results using the SRN-Deblur 's pre-trained model provided by your GitHub. I follow the instruction from the "Test" suggested in your GitHub page:
python run_model.py --phase=test --model=color --checkpoint_path=RealBlurJ_pre_trained+GOPRO+BSD500 --datalist=datalist/RealBlur_J_test_list.txt --height=784 --width=688

The following is what I get

To my understanding, the psnr result I get is 29.60189

Unfortunately, the result in the paper is 31.38

I have verified my code environment: Python 2.7, Cuda 8.0, open-cv 3.4.3.18, tf 1.4 & tf tensorboard 0.4, and I use the original random seed "0" from your code provided, but still, it seems that I cannot reproduce the same results of the paper.

I really appreciate it if you could point out something wrong regarding my experiments. Thank you very much!