Research map / Vision-language and multimodal / Multimodal LLMs
Structured Redundancy Modeling for Efficient Visual Token Pruning in High-Resolution MLLMs
Jouwon Song, Woohyeong Kim, Kyeongbo Kong
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 directional masking to SG Top-K selection (ablation) | SFPruner · LLaVA-NeXT-7B ablation, 5x64 tokens (320) | relative performance (%): 96.9 -> 97.9 | Improved |
| add directional masking to SG Top-K selection (ablation) | SFPruner · LLaVA-NeXT-7B ablation, 5x64 tokens (320) | selection time (ms): 0.7 -> 0.8 | Hurt |
| use directional masking instead of sequential selection (ablation) | SFPruner · LLaVA-NeXT-7B ablation, 5x64 tokens (320) | selection time (ms): 18.6 -> 0.8 | Improved |
| use directional masking instead of sequential selection (ablation) | SFPruner · LLaVA-NeXT-7B ablation, 5x64 tokens (320) | relative performance (%): 98.0 -> 97.9 | No clear effect |
| add SG-RLS to Top-K selection (replace SG) (ablation) | SFPruner · LLaVA-NeXT-7B ablation, 5x64 tokens (320) | relative performance (%): 96.9 -> 97.6 | Improved |
| add SG-RLS to directional masking (ablation) | SFPruner · LLaVA-NeXT-7B ablation, 5x64 tokens (320) | relative performance (%): 97.9 -> 98.3 | Improved |
| add SG-RLS to directional masking (ablation) | SFPruner · LLaVA-NeXT-7B ablation, 5x64 tokens (320) | selection time (ms): 0.8 -> 2.3 | Hurt |
| use SFPruner instead of CDPruner | SFPruner · Qwen2.5-VL-7B retain 40% tokens | selection time (ms): 112.4 -> 2.5 | Improved |
| use SFPruner instead of CDPruner | SFPruner · Qwen2.5-VL MME stress test (9216 tokens, retain 30%) | pruning time (ms): 576 -> 28 | Improved |
| apply SFPruner token pruning (30% retention) | SFPruner · Qwen2.5-VL MME stress test (9216 tokens, retain 30%) | prefill latency (ms): 1813 -> 1114 | Improved |
| apply SFPruner token pruning (30% retention) | SFPruner · Qwen2.5-VL MME stress test (9216 tokens, retain 30%) | total inference time (s): 6101 -> 3601 | Improved |
| use SFPruner instead of learned VisionSelector | SFPruner · Qwen2.5-VL-7B retain 20% tokens | relative performance (%): 95.7 -> 96.5 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DivPrune — DivPrune: Diversity-based Visual Token Pruning for Large Multimodal Models (CVPR 2025) · better on 7 of 7 shared results
- VisionZip — VisionZip: Longer is Better but Not Necessary in Vision Language Models (CVPR 2025) · better on 2 of 2 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
- Beyond Attention or Similarity: Maximizing Conditional Diversity for Token Pruning in MLLMs (NeurIPS 2025 · similarity 0.94)
- Hi-Lo Prune: Look at What You'll Lose before Pruning with Hierarchical Token Selection (CVPR 2026 · similarity 0.93)
- TopV: Compatible Token Pruning with Inference Time Optimization for Fast and Low-Memory Multimodal Vision Language Model (CVPR 2025 · similarity 0.92)
- What Kind of Visual Tokens Do We Need? Training-Free Visual Token Pruning for Multi-Modal Large Language Models from the Perspective of Graph (AAAI 2025 · similarity 0.92)
- Fit and Prune: Fast and Training-free Visual Token Pruning for Multi-modal Large Language Models (AAAI 2025 · similarity 0.92)
- Balanced Token Pruning: Accelerating Vision Language Models Beyond Local Optimization (NeurIPS 2025 · similarity 0.92)
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.
