Lite Any Stereo V2: Faster and Stronger Efficient Zero-Shot Stereo Matching

2026-06-23

AI Summary

Deploying high-accuracy stereo matching models on resource-constrained platforms remains challenging, and existing efficient methods often suffer from insufficient zero-shot generalization. This work proposes LAS2, a family of ultra-fast stereo matching models featuring a purely 2D cost aggregation architecture optimized for real-world inference latency. To enhance synthetic-to-real transfer performance, we introduce a three-stage training strategy that integrates synthetic supervision, self-distillation, and real-world knowledge distillation, augmented with pseudo-label filtering and error clamping mechanisms. LAS2 achieves state-of-the-art accuracy among efficient methods while maintaining extremely low latency; notably, LAS2-H runs 1.8× and 2.7× faster than Fast-FoundationStereo on H200 and Orin platforms, respectively.

Abstract

Recent advances in stereo matching have achieved remarkable accuracy, but often rely on large models, heavy computation, or additional foundation-model priors, making them difficult to deploy on resource-constrained platforms. In contrast, efficient stereo models offer faster inference but are commonly considered less capable of strong zero-shot generalization. In this paper, we challenge this assumption by introducing Lite Any Stereo V2 (LAS2), an ultra-fast model series designed for efficient zero-shot stereo matching. LAS2 is developed from both architecture and training perspectives. Architecturally, we revisit efficient stereo design under practical deployment settings and propose a 2D-only cost aggregation framework, optimized for real inference latency rather than theoretical MACs alone. For training, we develop a three-stage strategy that combines synthetic supervision, self-distillation, and real-world knowledge distillation. To improve the reliability of real-world pseudo supervision, we further introduce pseudo-label filtering and an error-clamping operation, enabling smoother synthetic-to-real transfer. We instantiate LAS2 as a family of models, including feed-forward variants for different efficiency budgets and an iterative variant for higher accuracy. Extensive experiments show that LAS2 achieves state-of-the-art accuracy among efficient stereo methods while maintaining significantly lower latency. Specifically, LAS2-H achieves stronger overall zero-shot performance than the iterative method Fast-FoundationStereo, with 1.8x and 2.7x faster inference on H200 and Orin, respectively. The project page, demos, and code are available at https://tomtomtommi.github.io/LiteAnyStereoV2/.

Problem

• stereo matching
• zero-shot generalization
• efficient models
• resource-constrained deployment
• real-world transfer

Innovation

• zero-shot stereo matching
• efficient architecture
• 2D cost aggregation
• knowledge distillation
• pseudo-label filtering