Reproduce “Divide, then Ground: Adapting Frame Selection to Query Types for Long-Form 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

Jialuo Li, Bin Li, Jiahao Li, Yan Lu

DIG adapts frame sampling for long-form video question answering by routing global questions to uniform sampling and localized questions to a targeted selection pipeline. The pipeline identifies representative frames, scores relevance, and refines video before inference. On MLVU, Qwen2.5-VL-7B with 32 frames reaches 67.20% accuracy, versus 59.52% for uniform sampling; this result does not establish lower compute or universal gains.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.