Advice Loss function

What is a good loss function to calculate difference between a generated image and the ground truth. This is not a GAN. I’m trying to generate an image that is as close to the ground truth as possible. I’m using KLDivLoss in my model right now. Would like to hear more ideas though. Image size is (3,128,128)

Typically people use MSELoss for that.

1 Like

Thanks. Just tried that.

If it is classification (segmentation), use cross entropy. If it is regression (detection, super-resolution), use mse,l1, smooth L1.