Reproduce “VRU-Accident: A Vision-Language Benchmark for Video Question Answering and Dense Captioning for Accident Scene 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
Younggun Kim, Ahmed S. Abdelrahman, Mohamed Abdel-Aty
VRU-Accident evaluates video models’ understanding of real-world dashcam crashes involving pedestrians and cyclists through six multiple-choice question categories and dense captions. Its 1,000 videos yield 6,000 questions, testing scene context, accident type, cause, and prevention. In zero-shot tests, Gemini 1.5-flash averages 66.9%, versus 94.7% for human experts, with weaker accident-cause accuracy; scores assess benchmark tasks, not driving safety.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.