Prune Redundancy, Preserve Essence: Vision Token Compression in VLMs via Synergistic Importance-Diversity
About
Vision-language models (VLMs) face significant computational inefficiencies caused by excessive generation of visual tokens. While prior work shows that a large fraction of visual tokens are redundant, existing compression methods struggle to balance importance preservation and information diversity. To address this, we propose PruneSID, a training-free Synergistic Importance-Diversity approach featuring a two-stage pipeline: (1) Principal Semantic Components Analysis (PSCA) for clustering tokens into semantically coherent groups, ensuring comprehensive concept coverage, and (2) Intra-group Non-Maximum Suppression (NMS) for pruning redundant tokens while preserving key representative tokens within each group. Additionally, PruneSID incorporates an information-aware dynamic compression ratio mechanism that optimizes token compression rates based on image complexity, enabling more effective average information preservation across diverse scenes. Extensive experiments demonstrate state-of-the-art performance, achieving 96.3% accuracy on LLaVA-1.5 with only 11.1% token retention, and 92.8% accuracy at extreme compression rates (5.6%) on LLaVA-NeXT, outperforming prior methods by 2.5% with 7.8 $\times$ faster prefilling speed compared to the original model. Our framework generalizes across diverse VLMs and both image and video modalities, showcasing strong cross-modal versatility. Code is available at https://github.com/ZhengyaoFang/PruneSID.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Object Hallucination Evaluation | POPE | Accuracy87.1 | 2019 | |
| Visual Question Answering | TextVQA | Accuracy73.46 | 1453 | |
| Visual Question Answering | VQA v2 | Accuracy74.8 | 1429 | |
| Visual Question Answering | GQA | Accuracy60.2 | 1425 | |
| Multimodal Understanding | MMBench | Accuracy63.8 | 847 | |
| Visual Question Answering | ChartQA | Accuracy76.92 | 519 | |
| Visual Question Answering | ScienceQA | Accuracy71.1 | 446 | |
| Optical Character Recognition | OCRBench | Score726 | 433 | |
| Massive Multi-discipline Multimodal Understanding | MMMU | Accuracy37.2 | 216 | |
| Visual Question Answering | TextVQA | TextVQA Accuracy65.89 | 210 |