Reproduce “Efficient One-Step Diffusion Restoration Model with Compact Token Compression and Linear Attention” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Bingtian Qiao, Yue Shi, Yingjie Zhou, Yong Guo, Guangtao Zhai, Jiezhang Cao

SANA-SR performs real-world image super-resolution by compressing images into a 32× latent grid, using linear attention and one transformer update. It adapts a pretrained SANA model with LoRA and a prompt. On DRealSR, it scores 70.69, above LinearSR’s 69.22, on MUSIQ under an HQ-derived prompt protocol. Its reported 0.019-second time measures the pruned backbone, not the full pipeline.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.