can I use this network for compression artifact reduction? and how to create training data?

[maliho0803]

can I use this network for compression artifact reduction? and how to create training data?

[avinabsaha]

I guess yeah. Just pass scale =1 and create datasets accordingly.

[Wangxin25]

I guess yeah. Just pass scale =1 and create datasets accordingly.

Have you tried it? Why did I loss(L1) a lot in training?

[avinabsaha]

Yeah I have used it with scale = 1! Keep the other folder structure and other image name convention similar to scale 2/4 and it should work!

[avinabsaha]

Check out this paper for loss functions : https://arxiv.org/pdf/1511.08861.pdf

[Wangxin25]

Check out this paper for loss functions : https://arxiv.org/pdf/1511.08861.pdf

Many thanks