VScan: Rethinking Visual Token Reduction for Efficient Large Vision-Language Models

TL;DR

VScan introduces a two-stage visual token reduction method that significantly accelerates large vision-language models by intelligently pruning tokens during encoding and decoding, maintaining high performance with reduced computational costs.

Contribution

The paper proposes VScan, a novel framework for visual token reduction that combines global and local scans with token merging, improving efficiency without sacrificing accuracy.

Findings

VScan achieves 2.91× speedup in pre-filling for LLaVA-NeXT-7B.

VScan reduces FLOPs by 10× while retaining 95.4% of original performance.

Extensive experiments validate VScan's effectiveness across multiple benchmarks.

Abstract

Recent Large Vision-Language Models (LVLMs) have advanced multi-modal understanding by incorporating finer-grained visual perception and encoding. However, such methods incur significant computational costs due to longer visual token sequences, posing challenges for real-time deployment. To mitigate this, prior studies have explored pruning unimportant visual tokens either at the output layer of the visual encoder or at the early layers of the language model. In this work, we revisit these design choices and reassess their effectiveness through comprehensive empirical studies of how visual tokens are processed throughout the visual encoding and language decoding stages. Guided by these insights, we propose VScan, a two-stage visual token reduction framework that addresses token redundancy by: (1) integrating complementary global and local scans with token merging during visual encoding, and (2) introducing pruning at intermediate layers of the language model. Extensive experimental results across four LVLMs validate the effectiveness of VScan in accelerating inference and demonstrate its superior performance over current state-of-the-arts on sixteen benchmarks. Notably, when applied to LLaVA-NeXT-7B, VScan achieves a 2.91$\times$ speedup in prefilling and a 10$\times$ reduction in FLOPs, while retaining 95.4\% of the original performance. Code is available at https://github.com/Tencent/SelfEvolvingAgent/tree/main/VScan.