Reproduce “Long-to-Short Video Evidence Reasoning for Grounded Question Answering” 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
Kaiyan Chen, Junbin Xiao, Xun Yang
LOVER trains grounded video question answering to return answers with supporting intervals. It rewards the predicted span’s overlap with annotated evidence, progresses from long to short evidence, and renders legible timestamps on frames. On NExT-GQA, Time-R1-7B’s Acc@GQA increases from 26.1 to 32.7. On ReXTime, it rises 21.46 to 21.87, while [email protected] remains 27.37 in tested multiple-choice visual-frame GQA.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.