Reproduce “RelayVSR: Large-Small Model Collaboration for Efficient Real-World Video 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

Xijun Wang, Xin Li, Zirui Lang, Suhang Yao, Haoran Li, Zhibo Chen

RelayVSR restores degraded low-resolution video by using a large generative model to create references for occasional frames and a lightweight network to reconstruct every frame. Video-Aware Reference Optimization trains it using decoded-reference and full-video feedback. On 201-frame 1080p clips, dual-endpoint mode reaches 29.29 FPS versus 7.80 for FlashVSR-Tiny; this measured throughput requires lookahead and is not startup latency.

Summary by alphaXiv.

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