Research map / Video / Video-language understanding
FlashVID: Efficient Video Large Language Models via Training-free Tree-based Spatiotemporal Token Merging
Ziyang Fan, Keyu Chen, Ruilong Xing, Yulin Li, Li Jiang, Zhuotao Tian
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 ADTS (attention + diversity) selection instead of attention-only (ATS) (ablation) | FlashVID · 4 video benchmarks, R=10% | relative accuracy: 96.9 -> 99.1 | Improved |
| use ADTS instead of diversity-only (DTS) (ablation) | FlashVID · 4 video benchmarks, R=10% | relative accuracy: 97.1 -> 99.1 | Improved |
| add [CLS] attention calibration to diversity selection (ablation) | ADTS · 4 video benchmarks, R=10% | relative accuracy: 97.1 -> 98.5 | Improved |
| add event relevance calibration to diversity selection (ablation) | ADTS · 4 video benchmarks, R=10% | relative accuracy: 97.1 -> 97.6 | Improved |
| combine ADTS and TSTM (alpha=0.7) rather than TSTM alone (ablation) | FlashVID · 4 video benchmarks, R=10% | relative accuracy: 97.4 -> 99.1 | Improved |
| combine ADTS and TSTM (alpha=0.7) rather than ADTS alone (ablation) | FlashVID · 4 video benchmarks, R=10% | relative accuracy: 98.5 -> 99.1 | Improved |
| constrain spanning tree depth and breadth in TSTM (ablation) | TSTM · video benchmarks | accuracy | No clear effect |
| use FlashVID instead of FastVID at 10% retention | LLaVA-OneVision · video benchmarks avg | relative accuracy: 97.8 -> 99.1 | Improved |
| use FlashVID instead of VisionZip at low retention | LLaVA-OneVision · video benchmarks avg | relative accuracy: 92.5 -> 99.1 (R=10%) | Improved |
| prune tokens with FlashVID at 25% retention vs full tokens | LLaVA-OneVision · video benchmarks avg | relative accuracy: 100.0 -> 100.3 | Improved |
| extend input from 16 to 160 frames with FlashVID under same token budget | Qwen2.5-VL · 5 video benchmarks avg | relative accuracy: 100.0 -> 108.6 | Improved |
| use FlashVID instead of FastVID on LLaVA-Video at 20% retention, VideoMME overall | LLaVA-Video · VideoMME Overall | accuracy: 62.6 (FastVID) vs 62.2 (FlashVID) | Hurt |
Later papers that compared with this method
2 later accepted papers report FlashVID as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- EchoPrune: Interpreting Redundancy as Temporal Echoes for Efficient VideoLLMs (NeurIPS 2026 · better on 9 of 9 shared results)
- V-CAST: Video Curvature-Aware Spatio-Temporal Pruning for Efficient Video Large Language Models (NeurIPS 2026 · better on 2 of 2 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FastVID — FastVID: Dynamic Density Pruning for Fast Video Large Language Models (NeurIPS 2025) · better on 8 of 8 shared results
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
- FastVID: Dynamic Density Pruning for Fast Video Large Language Models (NeurIPS 2025 · similarity 0.92)
- V-CAST: Video Curvature-Aware Spatio-Temporal Pruning for Efficient Video Large Language Models (NeurIPS 2026 · similarity 0.92)
- HoliTom: Holistic Token Merging for Fast Video Large Language Models (NeurIPS 2025 · similarity 0.91)
- DyCoke: Dynamic Compression of Tokens for Fast Video Large Language Models (CVPR 2025 · similarity 0.91)
- Video-LaVIT: Unified Video-Language Pre-training with Decoupled Visual-Motional Tokenization (ICML 2024 · similarity 0.91)
- One Token per Highly Selective Frame: Towards Extreme Compression for Long Video Understanding (NeurIPS 2025 · similarity 0.89)
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.
