Reproduce “VIVID: Beyond Static Multi-Image Reasoning with Localized Visual Differences and Textual Entity Dynamics in Video MLLMs” 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
Chengqian Ma, Yuanze Zeng, Hongyu Wang, Mingfeng Lin, YiCheng Xiao, Yu Chen, Pengyu Li, WenHao Wang and 3 more
VIVID addresses frame-limited video question answering by retaining changes between nearby frames alongside a small set of selected keyframes. In one football illustration, whole-frame similarity is 99.8%, while the most changed 15% of patches show 20.4% change. On Video-MME overall, VIVID-Small scores 60.74% versus VideoStreaming's 58.11% under a shared setup; this is benchmark accuracy, not latency evidence.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.