Noises in Deblurred Images During Training
Closed this issue · 3 comments
I noticed that some deblurred images have red and green noises when I trained it myself. Continue training doesn't remove all the unpleasant noises. Here is the link to the picture ,
noisy deblurred images
Also, I re-train other deblurring models based on your provided codes and had the same issue.
I am wondering what might cause this issue. Any hints will be appreciated. Thank you.
Hello,
What kind of datasets do you use for training?
During training, I did not find noisy artifacts when using GoPro or RealBlur datasets.
You can also check if you loaded the correct blurry and sharp pairs during training.
I used GoPro for training. I am not sure if the issue comes from different versions of libraries. I will keep looking for the reason why colorful artifacts were introduced after training on my machine. The deblurring qualitative results were good if ignoring the annoying noises. So weird.
The tensorboard showed me noisy deblurred images. I tested the model not on tensorboard and results are of good quality. I guess it is due to bad color correction or something like that by tensorboard. Just ignore tensorboard and everything else is good. Thanks for your work again.