Reproduce “RECAP: Reciprocal Sparse Attention and Hierarchical Caching for Efficient Video Diffusion” 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
Tiange Xu, Kai Liu, Linghe Kong, Yulun Zhang
RECAP reduces video diffusion cost by combining sparse attention and temporal prediction reuse. It stores dense corrections for information omitted by sparse computation, while corrected predictions refresh reuse history and error feedback selects reuse, corrected computation, or dense refresh. On HunyuanVideo-13B at 720×1280, RECAP reports 29.67 dB PSNR in 1,129 seconds, exceeding the tested direct compositions in fidelity and latency.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.