Research map / Video / Video-language understanding
Threading Keyframe with Narratives: MLLMs as Strong Long Video Comprehenders
Bo Fang, YuXin Song, Haoyuan Sun, Qiangqiang Wu, Wenhao Wu, Antoni B. Chan
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 KFC+narratives (Nar-KFC) to uniform sampling | LLaVA-Video-7B · Video-MME (no sub.) | accuracy: 55.9 -> 61.6 | Improved |
| add Nar-KFC | InternVL2-8B · Video-MME (no sub.) | accuracy: 51.9 -> 56.3 | Improved |
| add Nar-KFC | Qwen2-VL-7B · Video-MME (no sub.) | accuracy: 55.0 -> 56.7 | Improved |
| add Nar-KFC vs KFC alone | Qwen2-VL-7B · LongVideoBench val | accuracy: 53.4 -> 53.6 (KFC 54.6) | Mixed |
| add Nar-KFC | InternVL2-8B · MLVU | accuracy: 54.3 -> 64.4 | Improved |
| insert narratives between uniformly sampled frames (ablation) | InternVL2-8B · Video-MME (sub.) | accuracy: 52.5 -> 55.4 | Improved |
| replace exact IQP with greedy search (ablation) | KFC · Video-MME (sub.) | accuracy: 55.1 -> 55.0 | No clear effect |
| remove query-relevance score (ablation) | KFC · Video-MME (sub.) | accuracy: 51.8 -> 55.0 | Improved |
| add frame-diversity score (ablation) | KFC · MLVU | accuracy: 60.9 -> 62.2 | Improved |
| add frame-diversity score (ablation) | KFC · Video-MME (sub.) | accuracy: 52.5 -> 55.0 | Improved |
| add initialization with most query-relevant frame to greedy search (ablation) | Vanilla GS · Video-MME (sub.) | accuracy: 52.3 -> 53.3 | Improved |
| add low-rank plus downsampling to score matrix (ablation) | Greedy search · Video-MME (sub.) | accuracy: 53.3 -> 54.7 | Improved |
Closest papers by meaning
- Adaptive Keyframe Sampling for Long Video Understanding (CVPR 2025 · similarity 0.92)
- One Token per Highly Selective Frame: Towards Extreme Compression for Long Video Understanding (NeurIPS 2025 · similarity 0.91)
- Logic-in-Frames: Dynamic Keyframe Search via Visual Semantic-Logical Verification for Long Video Understanding (NeurIPS 2025 · similarity 0.90)
- LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding (ICML 2025 · similarity 0.90)
- Leveraging Textual Memory and Key Frame Reasoning for Full Video Understanding Using Off-the-Shelf LLMs and VLMs (Student Abstract) (AAAI 2025 · similarity 0.90)
- BOLT: Boost Large Vision-Language Model Without Training for Long-form Video Understanding (CVPR 2025 · similarity 0.90)
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.
