Reproduce “Streaming Generative Video Compression with Embedding Alignment and In-Context Propagation” 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

Zhitao Wang, Hengyu Man, Han Liu, Wenrui Li, Debin Zhao, Xiaopeng Fan

SGVC adapts a pretrained video diffusion transformer for sequential compression, aligning contextual codec features with its input representation and reconstructing pixels directly. It carries forward attention states to use past frames without clip-wise buffering. Across three datasets, it reports a 71.97% average DISTS BD-rate saving over S²VC, a curve-level quality-bitrate result. YODA retains an average DISTS advantage; this does not establish real-time decoding.

Summary by alphaXiv.

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