FSRDiff: A fast diffusion-based super-resolution method using GAN

Ni Tang, Dongxiao Zhang, Juhao Gao, Yanyun Qu

Abstract: Highlights•FSRDiff combines the advantages of GAN and diffusion model for single image super-resolution.•FSRDiff uses conditional GAN to approximate the multimodal distribution in the inverse diffusion process, which improves the sampling efficiency and quality.•FSRDiff achieves nearly 20 times speedup over SRDiff while maintaining comparable performance.