Generated images from GAN look blurry

vision
1

Hey everyone, I trained my GAN on both a 50 images dataset and a 500 images dataset, each for 400 epochs, and they both look blurry. Does anyone know why this happens please?

50 images training run.
Losses

gen_disc_losses
gen_disc_losses1440×1080 49.3 KB

Input
input_0
Generated
gen_400
Actual output
target_0

500 images training run
Losses

gen_disc_losses
gen_disc_losses1440×1080 51.1 KB

Input
input_0
Generated
gen_400
Actual output
target_0

I’m not sure why they look blurry, and am not sure if it’s linked to the GAN losses not being correct or something like that. Would appreciate any input on this please.

2

I don’t think your question is PyTorch-like.
Since only we have is a loss curve, the only comment I can tell you is to focus on generator rather than discriminator.