Video Active Perception: Effective Inference-Time Long-Form Video Understanding with Vision-Language Models

Benchmark Model Rank Results
zero-shot-video-question-answer-on-next-qaVAP (GPT-4)–Accuracy: 73.2
zero-shot-video-question-answer-on-next-qaVAP (GPT-4o)–Accuracy: 73.8
zero-shot-video-question-answer-on-next-qaVAP (Gemini 1.5 Pro)–Accuracy: 71.1
zero-shot-video-question-answer-on-next-qaVAP (LLaVA-OV-72B)–Accuracy: 81.4