Research map / Vision-language and multimodal / Multimodal LLMs
DIVA: Instruction-Aware Vision Token Pruning via Dual-Probe Attention Discrepancy
Hyunwoo Kim, Kisu Lee, Yuna Shin, Ha Young Kim
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 DIVA-Pruner (CS score selection) (ablation) | DIVA · LLaVA-1.5-7B, 64 tokens kept (ablation) | relative performance vs unpruned (%): 87.9% -> 97.8% | Improved |
| add VF-Calibrator (ablation) | DIVA · LLaVA-1.5-7B, 64 tokens kept (ablation) | relative performance vs unpruned (%): 87.9% -> 95.6% | Improved |
| add VF-Calibrator on top of DIVA-Pruner (ablation) | DIVA · LLaVA-1.5-7B, 64 tokens kept (ablation) | relative performance vs unpruned (%): 97.8% -> 98.8% | Improved |
| use suffix plus similarity-selected instruction tokens as contextual probes instead of last text token (ablation) | DIVA · LLaVA-1.5-7B, 64 tokens kept (ablation) | relative performance vs unpruned (%): 95.1% -> 98.8% | Improved |
| use all instruction plus suffix tokens instead of similarity-selected subset (ablation) | DIVA · LLaVA-1.5-7B, 64 tokens kept (ablation) | relative performance vs unpruned (%): 98.6% -> 98.8% | No clear effect |
| forward KLD instead of reverse KLD (ablation) | DIVA · LLaVA-1.5-7B, 64 tokens kept (ablation) | relative performance vs unpruned (%): 95.9% -> 98.8% | Improved |
| forward KLD instead of JSD or CFG variants (ablation) | DIVA · LLaVA-1.5-7B, 64 tokens kept (ablation) | relative performance vs unpruned (%): 98.1% -> 98.8% | No clear effect |
| raise gamma from 0.1 to 0.5 (ablation) | DIVA · LLaVA-1.5-7B, 64 tokens kept (ablation) | relative performance vs unpruned (%): 98.8% -> 92.7% | Hurt |
| raise gamma from 0.1 to 0.3 (ablation) | DIVA · LLaVA-1.5-7B, 64 tokens kept (ablation) | relative performance vs unpruned (%): 98.8% -> 98.4% | No clear effect |
| DIVA vs CDPruner at 64 tokens | DIVA · LLaVA-1.5-7B, 64 tokens kept | relative performance vs unpruned (%): 98.6% -> 99.8% | Improved |
| DIVA vs PDrop at 10% tokens on Qwen3-VL-8B | DIVA · Qwen3-VL-8B, 10% tokens kept | relative performance vs unpruned (%): 68.1% -> 82.6% | Improved |
| DIVA vs CDPruner on video at 16 tokens per frame | DIVA · LLaVA-Video-7B, 64x16 | relative performance vs unpruned (%): 90.2% -> 94.5% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- VisionTrim — VisionTrim: Unified Vision Token Compression for Training-Free MLLM Acceleration (ICLR 2026) · better on 2 of 2 shared results
- DivPrune — DivPrune: Diversity-based Visual Token Pruning for Large Multimodal Models (CVPR 2025) · better on 1 of 1 shared result
- VisionZip — VisionZip: Longer is Better but Not Necessary in Vision Language Models (CVPR 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
- LearnPruner: Rethinking Attention-based Token Pruning in Vision Language Models (ICLR 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)
- Balanced Token Pruning: Accelerating Vision Language Models Beyond Local Optimization (NeurIPS 2025 · similarity 0.92)
- TransPrune: Token Transition Pruning for Efficient Large Vision-Language Model (CVPR 2026 · similarity 0.91)
- ATP-LLaVA: Adaptive Token Pruning for Large Vision Language Models (CVPR 2025 · similarity 0.91)
- An Empirical Study of Attention and Diversity for Adaptive Visual Token Pruning in Large Vision-Language Models (ICLR 2026 · similarity 0.91)
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.
