English

CRISP: Pre-LLM Yet Text-Driven Visual Token Pruning for Efficient LVLM Inference

Computer Vision and Pattern Recognition 2026-07-15 v1

Abstract

Large Vision-Language Models (LVLMs) typically require processing hundreds to thousands of visual tokens, leading to substantial inference overhead. Existing visual token pruning methods either operate before the LLM using text-agnostic heuristics or prune inside the LLM at the cost of efficiency and noisy cross-modal attention. To address these limitations, we propose CRISP, a pre-LLM yet text-driven visual token pruning framework that preserves both instruction-relevant evidence and essential scene context. CRISP works in a two-stage pipeline: Stage 1 first identifies text-aligned visual tokens, and Stage 2 enhances contextual completeness through semantic diversity. Extensive experiments on LLaVA-1.5 and LLaVA-NeXT demonstrate that CRISP achieves superior performance retention under aggressive pruning ratios, maintaining up to 99.5% accuracy while reducing inference cost and latency by more than 2 times. CRISP serves as a practical solution for efficient LVLM inference, especially in resource-constrained scenarios.

Cite

@article{arxiv.2607.16326,
  title  = {CRISP: Pre-LLM Yet Text-Driven Visual Token Pruning for Efficient LVLM Inference},
  author = {Xu Li and Yi Zheng and Mengyang Zhao and Yuxuan Liang and Zhe Liu and Rui Zhu and Xiaolei Chen and Wei Zhou and Baoquan Zhao and Juncen Guo},
  journal= {arXiv preprint arXiv:2607.16326},
  year   = {2026}
}

Comments

Accepted by the 2026 IEEE International Conference on Multimedia and Expo (ICME 2026) as Oral