Reproduce “Grounded Human-Attributed Description and Activity Recognition in Videos” 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

Jay N. Paranjape, Jurijs Nazarovs, Rohan Sarkar, Yue Guo, Lipeng Ke, Sankar Venkataraman, Vishal M. Patel, Nataraj Jammalamadaka

The paper introduces GHADAR, which asks a model to produce open-set appearance and activity captions for every person in a crowded video from bounding boxes. CAMP first restricts visual attention to an expanded person region, then removes that restriction during fine-tuning to restore scene context. On HC-STVG activity captions, InternVL3-1B’s METEOR score rises from 32.66 to 35.67; this measures reference-text agreement, not factual accuracy.

Summary by alphaXiv.

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