FoPru: Focal Pruning for Efficient Large Vision-Language Models

TL;DR

FoPru is a training-free token pruning method for large vision-language models that significantly improves inference efficiency by removing redundant visual tokens without sacrificing accuracy.

Contribution

We introduce FoPru, a novel attention-based token pruning technique that enhances LVLM inference efficiency without additional training.

Findings

Reduces visual tokens substantially while maintaining accuracy.

Improves inference speed across various LVLMs and datasets.

Offers two pruning strategies: rank and row.

Abstract

Large Vision-Language Models (LVLMs) represent a significant advancement toward achieving superior multimodal capabilities by enabling powerful Large Language Models (LLMs) to understand visual input. Typically, LVLMs utilize visual encoders, such as CLIP, to transform images into visual tokens, which are then aligned with textual tokens through projection layers before being input into the LLM for inference. Although existing LVLMs have achieved significant success, their inference efficiency is still limited by the substantial number of visual tokens and the potential redundancy among them. To mitigate this issue, we propose Focal Pruning (FoPru), a training-free method that prunes visual tokens based on the attention-based token significance derived from the vision encoder. Specifically, we introduce two alternative pruning strategies: 1) the rank strategy, which leverages all token significance scores to retain more critical tokens in a global view; 2) the row strategy, which focuses on preserving continuous key information in images from a local perspective. Finally, the selected tokens are reordered to maintain their original positional relationships. Extensive experiments across various LVLMs and multimodal datasets demonstrate that our method can prune a large number of redundant tokens while maintaining high accuracy, leading to significant improvements in inference efficiency.