Reproduce “TEASR: Training-Efficient Any-Step Diffusion Transformer for Real-World Image Super-Resolution” 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
Xiang Gao, Chenxin Zhu, Yushun Fang, Qiang Hu, Xiaoyun Zhang
TEASR trains one diffusion model for real-world image super-resolution that can restore in one step or sample for up to 30 steps without retraining or an external teacher. On ×4 RealSR, its one-step LPIPS and FID beat OSEDiff, while PSNR and SSIM do not. More steps offer a choice, but worsen RealSR full-reference metrics.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.