Reproduce “Where, How, and When: Query-Guided Grounding for 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

Yogesh Kumar

The authors present three query-guided video-grounding methods for locating objects, relationships, and moments, rather than one unified system. TOViT reaches 30.6 AP on FSVOD-500 in a five-shot setup, versus 25.1 for FSVOD. MATR reaches 59.2 mIoU and 56.5 Recall@1 on SportsMoments; its dual-stage alignment ablation improves both metrics over no alignment, without pretraining, on SportsMoments.

Summary by alphaXiv.

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