waifu2x
waifu2x copied to clipboard
How do you solve the noise/artefact problem ?
Hi,
I am very interested about how you train the model to solve the image artifact problem caused by image compression (jpeg images) ?
I try some other SR models before, most of them can only handle png images and performs very bad in jpeg format image.
Best Regard, Even.
simply train jpeg denoising.
denoise: input image -> add jpeg noise -> CNN -> output image loss = MSE(input image, output image) SR+denoise: input image -> downsampling -> add jpeg noise -> CNN(upsampling) -> output image loss = MSE(input image, output image)