search

zcemycl / Matlab-GAN

MATLAB implementations of Generative Adversarial Networks -- from GAN to Pixel2Pixel, CycleGAN
MIT License
184 stars 85 forks source link

With cycleGAN, the center is black even after huge epoch. #9

Closed ApplePomme closed 3 years ago

ApplePomme commented 3 years ago

The center is black even after huge epoch with cycleGAN.

Original code without any modification.

Hope someone can point out my running errors.

black_center

zcemycl commented 3 years ago

@ApplePomme If i remember correctly, GAN variants do not do well when you are overtraining it. Try training it under 60 epochs, and with less than batch size of 8.

Unlike image recognition network, GAN performance does not get improved with too much epoch and larger batch size.