VisPCO: Visual Token Pruning Configuration Optimization via Budget-Aware Pareto-Frontier Learning for Vision-Language Models

TL;DR

VisPCO introduces a gradient-based Pareto optimization framework for visual token pruning in vision-language models, balancing computational efficiency and performance.

Contribution

It formulates token pruning as a Pareto configuration problem and employs continuous relaxation for automatic, optimal pruning configuration discovery.

Findings

Effectively approximates the empirical Pareto frontier across benchmarks.

Generalizes well across different pruning methods and architectures.

Multi-step progressive pruning outperforms single-layer approaches.

Abstract

Visual token pruning methods effectively mitigate the quadratic computational growth caused by processing high-resolution images and video frames in vision-language models (VLMs). However, existing approaches rely on predefined pruning configurations without determining whether they achieve computation-performance optimality. In this work, we introduce , a novel framework that formulates visual token pruning as a Pareto configuration optimization problem to automatically identify optimal configurations. Our approach employs continuous relaxation and straight-through estimators to enable gradient-based search, solved via the Augmented Lagrangian method. Extensive experiments across 8 visual benchmarks demonstrate that effectively approximates the empirical Pareto frontier obtained through grid search and generalizes well across various pruning methods and VLM architectures. Furthermore, through learnable kernel functions, we investigate layer-wise pruning patterns and reveal that multi-step progressive pruning captures VLMs' hierarchical compression structure, achieving superior accuracy-efficiency trade-offs compared to single-layer approaches.