Reproduce “ReMem: Streaming Video Understanding With Long Context Retention” 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
Li Yiheng, He Xu, Wang Shaobo, Shao Ling, Lu Shijian
ReMem enables offline video-language models to process continuing video with bounded memory. Streaming Context Memory compresses incoming visual tokens independently of the question, while Retrieved Vision Memory selects relevant past context when a question arrives. On StreamingBench, Qwen2.5-VL-7B at 1 frame per second improves ACU accuracy from 24.8% to 44.8%; gains cover selected tasks, not every evaluation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.