Stereo Anywhere: Robust Zero-Shot Deep Stereo Matching Even Where Either Stereo or Mono Fail

TL;DR

Stereo Anywhere is a new deep stereo-matching framework that combines geometric constraints with monocular depth priors, achieving robust zero-shot performance across diverse challenging scenarios.

Contribution

It introduces a dual-branch architecture with novel cost volume fusion mechanisms, integrating stereo and monocular cues for improved robustness and generalization.

Findings

State-of-the-art zero-shot generalization on multiple benchmarks

Robust performance in textureless, occluded, and non-Lambertian regions

Effective handling of challenging cases like mirrors and transparencies

Abstract

We introduce Stereo Anywhere, a novel stereo-matching framework that combines geometric constraints with robust priors from monocular depth Vision Foundation Models (VFMs). By elegantly coupling these complementary worlds through a dual-branch architecture, we seamlessly integrate stereo matching with learned contextual cues. Following this design, our framework introduces novel cost volume fusion mechanisms that effectively handle critical challenges such as textureless regions, occlusions, and non-Lambertian surfaces. Through our novel optical illusion dataset, MonoTrap, and extensive evaluation across multiple benchmarks, we demonstrate that our synthetic-only trained model achieves state-of-the-art results in zero-shot generalization, significantly outperforming existing solutions while showing remarkable robustness to challenging cases such as mirrors and transparencies.