Reproduce “Query-aligned video frame selection for long video understanding” 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
Md. Safayet Islam, Dilip Sarkar, Liang Liang
QAV selects video frames for multiple-choice video question answering by scoring candidate frames against the question paired with each answer option, then passing the top frames to an MLLM. The selector is training-free and keeps the frame budget fixed. On Video-MME without subtitles, Qwen2-VL-7B with LongCLIP scores 60.90% accuracy, versus FOCUS at 57.07%; results are favorable in many, not all, tested settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.