Research map / Video / Video-language understanding
From Content to Knowledge: Lightning Fast Long-Video Understanding with Neural Knowledge Representations
Yuchen Guan, Xiao Li, Zongyu Guo, Xiaoyi Zhang, Xiulian Peng, Chun Yuan, Yan Lu
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 |
|---|---|---|---|
| upgrade NKR backbone from Qwen3-8B to Qwen3-14B (ablation) | NKR · LVBench | overall accuracy: 39.5 -> 48.8 (speed 0.26 -> 0.33 s/query) | Improved |
| raise LoRA rank from 64 to 256 (ablation) | NKR on Qwen3-14B · LVBench | overall accuracy: 44.2 -> 48.8 | Improved |
| add clip-level and video-level QA to dense descriptions (ablation) | NKR training data · LVBench | overall accuracy: 14.3 -> 48.8 | Improved |
| add dense descriptions to QA data (ablation) | NKR training data · LVBench | overall accuracy: 38.8 -> 48.8 | Improved |
| add video-level QA (ablation) | NKR training data · LVBench | overall accuracy: 45.0 -> 48.8 | Improved |
| replace direct video tokens with NKR weights | VLM-based long video QA · LVBench | inference time per query: 31.7s-64.5s -> 0.33s | Improved |
| use NKR instead of video tokens for videos longer than one hour | NKR · LVBench >1h vs <1h | accuracy drop: -0.3 (NKR) vs -6.7 (GPT4.1) | Improved |
| use NKR instead of retrieval agent VideoRAG | NKR vs VideoRAG · LVBench reasoning questions | accuracy: 43.9 -> 47.1 | Improved |
| use NKR instead of retrieval agent VideoRAG | NKR vs VideoRAG · LVBench key information retrieval | accuracy: 57.1 -> 53.0 | Hurt |
| apply NKR to images with Qwen2.5-VL-7B | Qwen2.5-VL-7B · MME-RealWorld subset | overall accuracy: 32.1 -> 35.4 | Improved |
| apply NKR to images with Qwen2.5-VL-7B | Qwen2.5-VL-7B · MME-RealWorld subset perception | accuracy: 39.9 -> 38.4 | Hurt |
| replace NKR with Deep Video Discovery agent | NKR · LVBench | overall accuracy: 48.8 -> 74.2 (0.33 -> 180.0 s/query) | Improved |
Closest papers by meaning
- Streaming Long Video Understanding with Large Language Models (NeurIPS 2024 · similarity 0.89)
- ReWind: Understanding Long Videos with Instructed Learnable Memory (CVPR 2025 · similarity 0.89)
- $\infty$-Video: A Training-Free Approach to Long Video Understanding via Continuous-Time Memory Consolidation (ICML 2025 · similarity 0.88)
- Leveraging Textual Memory and Key Frame Reasoning for Full Video Understanding Using Off-the-Shelf LLMs and VLMs (Student Abstract) (AAAI 2025 · similarity 0.88)
- SEAL: Semantic Attention Learning for Long Video Representation (CVPR 2025 · similarity 0.88)
- Temporal Chain of Thought: Long-Video Understanding by Thinking in Frames (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.
