Artifact SWINIR (training Model as Generator GAN)

[jojupiter]

i'm trying to train the swinIR model separately in a GAN. but i'm getting light artifacts in my results. is this normal when training swinIR? if so, how can i remove them?

[jojupiter]

Hello, please which SWINIR parameter can be used to adjust the presence of light noises?

[TwilightArchon]

This is probably due to not clipping the result to 0-1 (or 0-255) in each of your output channels. If they are out of this range, say the blue channel is 256, it will become 0, and you will be left with red and green, which will result in funny looking patches.

[sajsxj]

Hi, I also meet this problem. Can you tell me how you solved it? Thanks!!!

[sajsxj]

I've solved the problem.

[jojupiter]

how you solved the problem ?

[sajsxj]

The pixel value of the output image may exceed 1 or 255. Just control the pixel value of each channel to 0-1 or 0-255. Like this: torch.clamp(netG(z), min=0, max=1). Also, how did you solve it?