Research map / Video / Video-language understanding
MMG-Vid: Maximizing Marginal Gains at Segment-level and Token-level for Efficient Video LLMs
Junpeng Ma, Qizhe Zhang, Ming Lu, Zhibin Wang, Qiang Zhou, Jun Song, Shanghang Zhang
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 MMG-Vid pruning at 10% retention versus FastV | LLaVA-Video-7B · Average over 4 benchmarks, 10% retention | relative performance (%): 77.1 -> 93.8 | Improved |
| prune 75% of tokens with MMG-Vid | LLaVA-OneVision-7B · Average over 4 benchmarks, 25% retention | relative performance (%): 100.0 -> 99.5 | Improved |
| use TG-DPC (temporal-guided) instead of standard DPC-KNN for token pruning (ablation) | MMG-Vid · MVBench, 25% retention | relative performance (%): 94.6 -> 97.3 | Improved |
| use marginal gain-based dynamic budget instead of uniform static budget (ablation) | MMG-Vid · LongVideoBench and MLVU, 25% retention | relative performance (%): +0.8% LongVideoBench, +1.0% MLVU | Improved |
| use balanced lambda=0.5 instead of representativeness only (lambda=1.0) (ablation) | MMG-Vid segment budgeting · Average, 25% retention | relative performance (%): 96.6 -> 97.6 | Improved |
| use balanced lambda=0.5 instead of diversity only (lambda=0) (ablation) | MMG-Vid segment budgeting · Average, 25% retention | relative performance (%): 97.2 -> 97.6 | Improved |
| diversity-only budgeting versus representativeness-only budgeting (ablation) | MMG-Vid segment budgeting · Average, 25% retention | relative performance (%): 96.6 -> 97.2 | Improved |
| MMG-Vid versus FrameFusion at 25% retention | LLaVA-Video-7B · Average, 25% retention | relative performance (%): 96.8 -> 98.1 | Improved |
| MMG-Vid versus VisionZip at 25% retention on LLaVA-OneVision | LLaVA-OneVision-7B · Average, 25% retention | relative performance (%): 99.2 -> 99.5 | Improved |
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.
- FrameFusion — FrameFusion: Combining Similarity and Importance for Video Token Reduction on Large Vision Language Models (ICCV 2025) · better on 1 of 1 shared result
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)
- HoliTom: Holistic Token Merging for Fast Video Large Language Models (NeurIPS 2025 · similarity 0.91)
- V-CAST: Video Curvature-Aware Spatio-Temporal Pruning for Efficient Video Large Language Models (NeurIPS 2026 · similarity 0.90)
- One Token per Highly Selective Frame: Towards Extreme Compression for Long Video Understanding (NeurIPS 2025 · similarity 0.90)
- EchoPrune: Interpreting Redundancy as Temporal Echoes for Efficient VideoLLMs (NeurIPS 2026 · similarity 0.90)
- SlowFocus: Enhancing Fine-grained Temporal Understanding in Video LLM (NeurIPS 2024 · 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.
