TUDSR: Twice Upsampling-Diffusion for Higher Super-Resolution
Researchers have developed TUDSR, a novel framework for image super-resolution that utilizes a two-stage diffusion process to achieve higher resolutions. This method addresses limitations in current diffusion models, which struggle with extreme upsampling ratios and high-resolution outputs. TUDSR, built upon SD2.1-base, demonstrates state-of-the-art performance, generating high-quality images at resolutions up to $2048^2$. Additionally, a separate paper introduces RSD, a distillation method for image super-resolution that achieves single-step restoration and competitive perceptual quality with fewer computational resources. AI
IMPACT These advancements in diffusion models for super-resolution could lead to more efficient and higher-quality image generation tools.