Using blurred images only, how to train your code??
edwardcho opened this issue · 4 comments
Hello sir,
I have interesting image deblurring and i searched your code.
I had tried your sample data(gopro).
I checked your sample data that is paired datasets.(blurred image vs sharp image)
But i have blurred images only.
Can i try to train your code??
Thanks,
Edward Cho.
This repository is for supervised training method introduced in the CVPR 2017 publication,
If you want to train without paired sharp images, you might be interested in self-supervised or unsupervised methods.
ex)
N. T. Madam et al., "Unsupervised Class-Specific Deblurring," ECCV 2018
B. Lu et al., "Unsupervised Domain-Specific Deblurring via Disentangled Representations," CVPR 2019
Best,
Seungjun
Hello Sir,
Thank you for your response quickly.
According to your comments, I will search another papers.
Thanks.
Edward Cho.
Hello Sir,
I checked your comment paper.
'B. Lu et al., "Unsupervised Domain-Specific Deblurring via Disentangled Representations," CVPR 2019'
Unfortunately, the input-format of this paper is images_A and images_B, but not paired image-set.
As you know, I want to network that has blurred images only as inputs.
Thanks,
Edward Cho.
Dear @edwardcho,
Though this paper is not for deblurring, I recommend you this paper: Unsupervised Image Restoration Using Partially Linear Denoisers (TPAMI 2021) https://ieeexplore.ieee.org/document/9395217
As deblurring or restoration is an ill-posed problem, it is demanding to acheive the task objective in a pure unsupervised manner, I think.