Research map / Video / Video-language understanding
VideoAgent: A Memory-augmented Multimodal Agent for Video Understanding
Yue Fan, Xiaojian Ma, Rujie Wu, yuntao du, Jiaqi Li, Zhi Gao, Qing Li
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 |
|---|---|---|---|
| add object memory database to the agent (with VQA, grounding, captions) (ablation) | VideoAgent · NExT-QA ablation subset (150 questions) | average accuracy: 62.7 -> 71.3 | Improved |
| add object re-identification to the object memory (ablation) | VideoAgent captions + object memory · NExT-QA ablation subset (150 questions) | average accuracy: 48.7 -> 56.0 | Improved |
| add visual question answering and segment localization tools to caption-only agent (ablation) | VideoAgent · NExT-QA ablation subset (150 questions) | average accuracy: 40.7 -> 62.7 | Improved |
| switch VQA tool from Video-LLaVA to GPT-4V (ablation) | VideoAgent · NExT-QA ablation subset (150 questions) | average accuracy: 71.3 -> 74.7 | Improved |
| use caption-query similarity instead of video-query similarity for segment localization (ablation) | VideoAgent segment localization · Ego4D NLQ val (zero-shot) | R1@0.3: 8.40 -> 10.07 | Improved |
| use higher-quality captions (Ego4D narrations instead of LaViLa) (ablation) | VideoAgent segment localization · Ego4D NLQ val (zero-shot) | R1@0.3: 10.07 -> 16.41 | Improved |
| ensemble textual caption features with ViCLIP visual features for localization (ablation) | VideoAgent segment localization · Ego4D NLQ val (zero-shot) | R1@0.3: 16.41 -> 17.39 | Improved |
| use similarity-based segment localization instead of LLM-based localization (LifeLongMemory) | VideoAgent · Ego4D NLQ val | R1@0.3 vs R@0.3: 15.99 -> 16.41 | Improved |
| wrap Video-LLaVA as a tool inside the agent instead of using it alone | Video-LLaVA · NExT-QA val subset (600 questions) | average accuracy: 53.5 -> 70.8 | Improved |
| agent versus GPT-4V on open-ended QA | VideoAgent · WorldQA Open-Ended | score: 35.37 (GPT-4V) vs 32.53 | Hurt |
| zero-shot agent versus supervised GroundNLQ | VideoAgent · Ego4D NLQ val | R1@0.3: 27.20 vs 17.39 | Hurt |
Later papers that compared with this method
18 later accepted papers report VideoAgent2 as a baseline in their results tables; 16 of them report a better number on most of the results they share with it.
- Think, Then Verify: A Hypothesis-Verification Multi-Agent Framework for Long Video Understanding (CVPR 2026 · better on 8 of 8 shared results)
- SciEducator: Scientific Video Understanding and Educating via Deming-Cycle Multi-Agent System (CVPR 2026 · better on 4 of 4 shared results)
- Bridging VideoQA and Video-Guided Agentic Tasks via Generalized Keyframe Extraction (ECCV 2026 · better on 4 of 4 shared results)
- A Benchmark and Multi-Agent System for Instruction-driven Cinematic Video Compilation (ECCV 2026 · better on 4 of 4 shared results)
- Ego-R1: Agentic Chain-of-Tool-Thought for Ultra-Long Egocentric Video Reasoning (TPAMI 2026 · better on 4 of 4 shared results)
- Towards Sparse Video Understanding and Reasoning (CVPR 2026 · better on 1 of 2 shared results)
- Chain-of-Glimpse: Search-Guided Progressive Object-Grounded Reasoning for Video Understanding (ICML 2026 · better on 1 of 1 shared result)
- VideoSEAL: Mitigating Evidence Misalignment in Agentic Long Video Understanding by Decoupling Answer Authority (ICML 2026 · better on 1 of 1 shared result)
10 more not listed.
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- VideoAgent: Long-form Video Understanding with Large Language Model as Agent (ECCV 2024 · similarity 0.93)
- ReAgent-V: A Reward-Driven Multi-Agent Framework for Video Understanding (NeurIPS 2025 · similarity 0.89)
- Seeing, Listening, Remembering, and Reasoning: A Multimodal Agent with Long-Term Memory (ICLR 2026 · similarity 0.89)
- Deep Video Discovery: Agentic Search with Tool Use for Long-form Video Understanding (NeurIPS 2025 · similarity 0.88)
- VideoWebArena: Evaluating Long Context Multimodal Agents with Video Understanding Web Tasks (ICLR 2025 · similarity 0.88)
- VideoLucy: Deep Memory Backtracking for Long Video Understanding (NeurIPS 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.
