Stabilized GAN models training with kernel-histogram transformation and probability mass function distance

Jangwon Seo, Hyo-Seok Hwang, Minhyeok Lee, Junhee Seok

Abstract: Highlights•Introduction of a novel GAN model using diverse kernels and distances for distribution comparison.•A new histogram transformation method in the discriminator improves distribution differentiation.•Evaluations on MNIST, CIFAR, CelebA, LSUN, and AFHQ show PMF-GANs superior image generation.•PMF-GANs integrate with latest GAN architectures, offering flexibility for diverse applications.

External IDs:doi:10.1016/j.asoc.2024.112003