Reproduce “FlashBack: Knowing When to Remember in Streaming Vision-Language Models” 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
Yi Chen, MingMing Yu, Rui-Qi Wang, Boran Wang, Xiaohang Cao, Chu Tang, Jingmin Chen, Jie Gu
FlashBack routes questions to a live-answer path or a history-retrieval path by classifying their temporal evidence needs before retrieval. It combines recent context with selected past visual representations only when routed to recall. On StreamingVLM, structured routing scores 60.83 versus 41.67 on Clues Reveal Responding, but hallucination-sensitive accuracy declines, and the historical archive is unbounded.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.