Reproduce “FastV-RAG: Towards Fast and Fine-Grained Video QA with Retrieval-Augmented Generation” 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
Gen Li, Peiyu Liu
FastV-RAG answers knowledge-intensive video questions by retrieving text and generating a separate draft for each document with a smaller vision-language model. A stronger model checks reasoning support, while CLIP compares each draft’s entity with video frames. On VideoSimpleQA, it reports 91.12% accuracy and 25.74 seconds, versus 91.30% and 47.72 seconds for standard RAG; evaluation is limited to named datasets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.