Research map / Video / Video-language understanding
Keep It Simple: Multi-Key Episodic Memory Retrieval for Ultra-Long Video Understanding
Yeeun Choi, Youngbeom Yoo, Joon-Young Lee, Hyolim Kang, Seon Joo Kim
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 neighbor filtering (ablation) | MERIT (Qwen3-VL-8B) · EgoLifeQA | average accuracy (%): 48.0 -> 54.2 | Improved |
| add neighbor filtering (ablation) | MERIT (GPT-5) · EgoLifeQA | average accuracy (%): 70.0 -> 71.2 | Improved |
| add neighbor filtering to oracle clip (ablation) | Target Time Oracle (Qwen3-VL-8B) · EgoLifeQA | average accuracy (%): 64.0 -> 65.8 | Improved |
| add neighbor filtering (ablation) | MERIT (GPT-5) · EgoLifeQA | hit rate: 0.40 -> 0.56 | Improved |
| use all four keys instead of the summary key alone (ablation) | MERIT (Qwen3-VL-8B) · EgoLifeQA | average accuracy (%): 49.4 -> 54.2 | Improved |
| use summary key instead of event key as single key (ablation) | MERIT (Qwen3-VL-8B) · EgoLifeQA | accuracy (%): 44.6 -> 49.4 | Improved |
| add multi-key representation to flat retrieval (ablation) | flat retrieval (Qwen3-VL-8B) · EgoLifeQA | average accuracy (%): 46.6 -> 48.0 | Improved |
| add neighbor filtering to flat retrieval (ablation) | flat retrieval (Qwen3-VL-8B) · EgoLifeQA | average accuracy (%): 46.6 -> 52.8 | Improved |
| add both multi-key and neighbor filtering to flat retrieval (ablation) | flat retrieval (Qwen3-VL-8B) · EgoLifeQA | average accuracy (%): 46.6 -> 54.2 | Improved |
| retrieve top 10 clips instead of 5 (ablation) | MERIT (Qwen3-VL-8B) · EgoLifeQA | average accuracy (%): 54.2 -> 52.2 | Hurt |
| retrieve top 10 clips instead of 5 (ablation) | MERIT (GPT-5) · EgoLifeQA | average accuracy (%): 67.4 -> 71.2 | Improved |
| retrieve top 10 clips instead of 5 (ablation) | MERIT (Gemini 2.5 Pro) · EgoLifeQA | average accuracy (%): 61.4 -> 64.2 | Improved |
Closest papers by meaning
- AdaVideoRAG: Omni-Contextual Adaptive Retrieval-Augmented Efficient Long Video Understanding (NeurIPS 2025 · similarity 0.89)
- Streaming Long Video Understanding with Large Language Models (NeurIPS 2024 · similarity 0.89)
- $\infty$-Video: A Training-Free Approach to Long Video Understanding via Continuous-Time Memory Consolidation (ICML 2025 · similarity 0.89)
- Video-RAG: Visually-aligned Retrieval-Augmented Long Video Comprehension (NeurIPS 2025 · similarity 0.88)
- VideoLucy: Deep Memory Backtracking for Long Video Understanding (NeurIPS 2025 · similarity 0.88)
- Vgent: Graph-based Retrieval-Reasoning-Augmented Generation For Long Video Understanding (NeurIPS 2025 · similarity 0.88)
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.
