VisPCO: Visual Token Pruning Configuration Optimization via Budget-Aware Pareto-Frontier Learning for Vision-Language Models
For vision-language model practitioners, VisPCO automates the search for optimal pruning configurations, eliminating manual tuning and improving performance-efficiency balance.
VisPCO formulates visual token pruning as a Pareto configuration optimization problem, automatically identifying optimal pruning configurations that achieve superior accuracy-efficiency trade-offs across 8 visual benchmarks, outperforming single-layer approaches.
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.