High-quality face image generation using particle swarm optimization-based generative adversarial networks

Abstract

Face image generation based on generative adversarial networks (GAN) is a hot research topic in computer vision. Existing GAN-based algorithms are constrained by training instability and mode collapse. Considering that particle swarm optimization (PSO) algorithm has good global optimization ability, we propose a generation antagonism network based on PSO algorithm to improve the training stability. More specifically, the inertia weight of particle swarm is improved by using the parameters of particle representative generator network in particle swarm optimization, and the aggregation degree of particles is judged to ensure the optimization ability of particle swarm optimization and the diversity of population. In addition, we evaluate the performance of the generator by generating quality and diversity evaluation functions to better guide the iterative updating of particle swarm optimization. Our face image generation experiment is conducted on CelebA dataset and experimental result shows the effectiveness and robustness of our proposed method.