Reproduce “Reviso: Not All Visual Evidence Deserves Equal Attention in Video Captioning” 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
Tengfei Liu, Yang Shi, Yuran Wang, Yuqi Tang, Yizhuo Jia, Zhuoran Zhang, Qixun Wang, Shujie Li and 7 more
Reviso prioritizes video evidence by making a coarse preview, choosing a relevant focus, then revisiting segments, cropping frames, or reading visible text with OCR. It interprets observations in context and updates its plan before producing a caption. On MCTS-VCB, it scores 65.7/56.8/60.9% precision/recall/F1, versus 57.6/37.7/45.5% for its backbone; these are benchmark-specific results.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.