LinStereo: Linear-Complexity Global Attention for Multi-Scale Iterative Stereo Matching

2026-06-24

AI Summary

This work addresses the limitations of existing vision foundation model–based stereo matching methods, which struggle with multi-scale feature utilization, geometric initialization, and contextual propagation—particularly under severe photometric degradation in underwater environments, leading to poor generalization. To overcome these challenges, we propose LinStereo, built upon Depth Anything V3, which introduces Position-Aware Linear Attention (PALA) for efficient global modeling. LinStereo further integrates a Hierarchical Semantic Cost Volume (HSCV) and a monocular Depth Prior Initialization (DPI) strategy to significantly enhance matching accuracy and cross-domain robustness. Experimental results demonstrate that our method reduces AbsRel error by 28% on TartanAir-UW and by 26% on the SQUID underwater dataset, while also achieving state-of-the-art performance on standard benchmarks.

Abstract

Existing Vision Foundation Model (VFM)-based iterative stereo pipelines under-exploit three information pathways: multi-scale backbone features are collapsed into single-level correlations, geometric priors remain untapped at initialization, and context propagates only locally. These gaps widen under degraded photometric cues, making underwater scenes a stringent generalization test. To address this, we propose LinStereo, built upon Depth Anything V3, whose core is a Position-Aware Linear Attention (PALA) module that replaces local recurrence with global aggregation at linear cost, propagating reliable estimates from well-matched regions into degraded areas while preserving disparity structure. PALA is made effective by two enabling components: Hierarchical Semantic Cost Volumes (HSCV), which supply scale-aligned correlations from the VFM feature hierarchy, and a Depth Prior Initialization (DPI) that converts monocular depth into a metrically calibrated warm start. LinStereo achieves state-of-the-art-level accuracy on standard benchmarks and strong cross-domain generalization, particularly on underwater scene where severe photometric degradation makes stereo matching particularly challenging, attaining the best overall accuracy with consistent gains 28% lower AbsRel on TartanAir-UW, 26% on SQUID, a real-world underwater dataset).

Problem

• stereo matching
• multi-scale features
• photometric degradation
• underwater scenes
• global context propagation

Innovation

• Linear Attention
• Global Aggregation
• Multi-Scale Stereo Matching
• Depth Prior Initialization
• Cross-Domain Generalization