Unified Spatio-Temporal Token Scoring for Efficient Video VLMs

TL;DR

This paper introduces STTS, a lightweight, unified token pruning method for video vision-language models that significantly improves efficiency with minimal performance loss across various video QA tasks.

Contribution

STTS is a novel, architecture-wide token scoring and pruning method that operates across both ViT and LLM without complex conditioning, enhancing efficiency in video VLMs.

Findings

Prunes 50% of vision tokens with 62% efficiency gain

Achieves only 0.7% performance drop across 13 video QA tasks

Test-time scaling further improves long-video QA performance

Abstract

Token pruning is essential for enhancing the computational efficiency of vision-language models (VLMs), particularly for video-based tasks where temporal redundancy is prevalent. Prior approaches typically prune tokens either (1) within the vision transformer (ViT) exclusively for unimodal perception tasks such as action recognition and object segmentation, without adapting to downstream vision-language tasks; or (2) only within the LLM while leaving the ViT output intact, often requiring complex text-conditioned token selection mechanisms. In this paper, we introduce Spatio-Temporal Token Scoring (STTS), a simple and lightweight module that prunes vision tokens across both the ViT and the LLM without text conditioning or token merging, and is fully compatible with end-to-end training. By learning how to score temporally via an auxiliary loss and spatially via LLM downstream gradients, aided by our efficient packing algorithm, STTS prunes 50% of vision tokens throughout the entire architecture, resulting in a 62% improvement in efficiency during both training and inference with only a 0.7% drop in average performance across 13 short and long video QA tasks. Efficiency gains increase with more sampled frames per video. Applying test-time scaling for long-video QA further yields performance gains of 0.5-1% compared to the baseline. Overall, STTS represents a novel, simple yet effective technique for unified, architecture-wide vision token pruning.