ReaSon: Reinforced Causal Search with Information Bottleneck for Video Understanding

Benchmark Model Rank Results
zero-shot-video-question-answer-on-next-qaReaSon (GPT-4o)–Accuracy: 77.6
zero-shot-video-question-answer-on-next-qaReaSon (LLaVA-Video-7B)–Accuracy: 81.4
zero-shot-video-question-answer-on-next-qaReaSon (Qwen2.5-VL-7B)–Accuracy: 80.4