Research map / Video / Video-language understanding
VideoAgent: Long-form Video Understanding with Large Language Model as Agent
Xiaohan Wang, Yuhui Zhang, Orr Zohar, Serena Yeung-Levy
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use iterative frame selection instead of uniform sampling of 180 frames (ablation) | VideoAgent · EgoSchema (500-question subset) | accuracy: 59.6% -> 60.2% with 8.4 vs 180 frames | Improved |
| increase iterative rounds from 1 to 3 (ablation) | VideoAgent · EgoSchema (500-question subset) | accuracy: 53.8% -> 58.6% -> 60.2% | Improved |
| increase rounds from 3 to 4 (ablation) | VideoAgent · EgoSchema (500-question subset) | accuracy: 60.2% -> 59.8% | No clear effect |
| use 5 initial frames rather than 3 or 8 (ablation) | VideoAgent · EgoSchema (500-question subset) | accuracy: 58.4% / 57.4% -> 60.2% | Improved |
| add self-evaluation to stop iteration early (ablation) | VideoAgent · EgoSchema (500-question subset) | accuracy and frames: 59.6% -> 60.2%; frames 11.8 -> 8.4 | Improved |
| add segment selection when generating retrieval queries (ablation) | VideoAgent · EgoSchema (500-question subset) | accuracy: +3.6% | Improved |
| use GPT-4 as the agent LLM instead of LLaMA-2-70B, Mixtral-8x7B or GPT-3.5 | VideoAgent · EgoSchema (500-question subset) | accuracy | Improved |
| use CogAgent or LaViLa captions instead of BLIP-2 captions | VideoAgent · EgoSchema (500-question subset) | accuracy | Improved |
| swap among OpenCLIP ViT-G, EVA-CLIP-8B, EVA-CLIP-8B-plus for retrieval | VideoAgent · EgoSchema (500-question subset) | accuracy | No clear effect |
| replace LLoVi with VideoAgent | zero-shot long-video QA · NExT-QA (val) | Acc@All: 67.7 -> 71.3 | Improved |
Closest papers by meaning
- VideoAgent: A Memory-augmented Multimodal Agent for Video Understanding (ECCV 2024 · similarity 0.93)
- Insight-V: Exploring Long-Chain Visual Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.89)
- Deep Video Discovery: Agentic Search with Tool Use for Long-form Video Understanding (NeurIPS 2025 · similarity 0.89)
- ReAgent-V: A Reward-Driven Multi-Agent Framework for Video Understanding (NeurIPS 2025 · similarity 0.88)
- TransAgent: Transfer Vision-Language Foundation Models with Heterogeneous Agent Collaboration (NeurIPS 2024 · similarity 0.87)
- VideoWebArena: Evaluating Long Context Multimodal Agents with Video Understanding Web Tasks (ICLR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
