CoverPrune: Coverage-Driven Token Pruning for 3D VLMs via Optimal Transport

2026-08-13

AI Summary

This work addresses the computational bottleneck in 3D vision-language model inference caused by the large number of visual tokens, where existing diversity-based pruning methods often compromise multi-view consistency and geometric structure. We propose CoverPrune, the first approach to formulate token pruning during inference as an optimal transport problem, prioritizing the preservation of visual evidence coverage over maximizing diversity. Our method introduces a training-agnostic optimal transport framework, a novel Feature-Spatial-Temporal transport cost, an efficient approximation algorithm (SGS), and a lightweight variant, CoverPrune-Lite. Evaluated across multiple 3D visual spatial reasoning benchmarks, CoverPrune maintains strong performance even under high pruning ratios, significantly improving token efficiency and achieving state-of-the-art results.

Abstract

While 3D Vision-Language Models (3D VLMs) have demonstrated remarkable spatial reasoning capabilities, they suffer from massive visual token counts that create severe computational bottlenecks during inference. Existing token pruning methods primarily rely on diversity-based selection, discarding similar tokens to maximize dispersion. However, in 3D environments, this approach frequently drops representative prototype tokens in favor of outliers, breaking the multi-view consistencies and geometric structures essential for spatial reasoning. In this paper, we propose a paradigm shift for 3D VLM token pruning: from maximizing diversity to preserving visual evidence coverage. We introduce CoverPrune, a training-free framework that formulates inference-time token pruning as an Optimal Transport (OT) problem. To overcome the intractable combinatorial subset selection inherent in this formulation, we design the Feature-Spatial-Temporal (FST) transport cost and target capacity, along with an efficient Spatial-Guided Greedy Selection (SGS) algorithm to approximate the OT objective. Furthermore, we propose CoverPrune-Lite, an accelerated variant utilizing spatially structured local matching for minimal overhead. Extensive experiments across multiple 3D visual-spatial reasoning benchmarks demonstrate that our methods achieve state-of-the-art token efficiency, maintaining robust reasoning performance even under highly aggressive pruning budgets. Visit our project website at https://github.com/Brucess/CoverPrune.

Problem

• 3D Vision-Language Models
• Token Pruning
• Spatial Reasoning
• Computational Bottleneck
• Multi-view Consistency

Innovation

• Optimal Transport
• Token Pruning
• 3D Vision-Language Models
• Coverage Preservation
• Spatial Consistency