Generative adversarial networks via a composite annealing of noise and diffusion

Kensuke Nakamura, Simon Korman, Byung-Woo Hong

Abstract: Highlights•Optimization method of generative adversarial networks using a balanced composition of noise with diffusion process.•Generator is trained using simplified data first while keeping its learning capability for high frequency information.•Synthetic dataset demonstrates the side-effect of the conventional diffusion.•Experimental results using the state-of-arts with major benchmarks support the effectiveness of the presented method.•Code is presented in Supplemental File.

External IDs:dblp:journals/pr/NakamuraKH24