LiteMatch: Lightweight Zero-Shot Stereo Matching via Cost Volume Stabilization

2026-06-30

AI Summary

This work addresses the high computational cost and poor cross-domain generalization of current learning-based stereo matching methods, which rely on heavy backbone networks and expensive 3D cost volume processing. To overcome these limitations, we propose LiteMatch, a lightweight stereo matching framework that introduces a Cost Volume Consistency Loss (CVC-Loss) to encourage sharp, unimodal disparity distributions. LiteMatch employs a dual-encoder architecture—comprising a Cross-View Correspondence Encoder (CVCE) and an FFT-based High-Frequency Encoder (HFE)—to effectively fuse global correspondence cues with high-frequency details while avoiding computationally intensive 3D convolutions. Coupled with a lightweight, low-iteration refinement module, LiteMatch achieves competitive end-point error (EPE) and D1 scores across Scene Flow, KITTI, Middlebury, ETH3D, and DrivingStereo benchmarks, using only 3.36M–9.58M parameters and demonstrating significantly improved zero-shot generalization.

Abstract

Despite rapid progress in learning-based stereo matching, high accuracy is often achieved at the cost of heavy backbones and computationally intensive 3D cost volume processing, resulting in substantial memory and runtime overhead. More critically, these methods frequently struggle to generalize across domains, limiting their practical deployment. We present \textit{LiteMatch}, a lightweight stereo matching framework that achieves strong zero-shot generalization through cost volume stabilization-without expensive 3D convolutions. LiteMatch employs two complementary encoders: a Cross-View Correspondence Encoder (CVCE) to capture global cross-view interactions, and a High-Frequency Encoder (HFE) that enhances fine structural details via FFT-based frequency cues. To stabilize the cost volume, we introduce the \textit{Cost Volume Consistency Loss (CVC-Loss)}, a voxel-wise binary cross-entropy objective applied to softmax-normalized cost distributions. By encouraging sharp and unimodal disparity probabilities, CVC-Loss promotes stable cost distributions and enables rapid convergence. A lightweight refinement module further produces sharp full-resolution disparities with low-iteration updates, avoiding heavy recurrent refinement. With a flexible design ranging from 3.36M to 9.58M parameters, LiteMatch achieves exceptional zero-shot generalization, delivering competitive EPE and D1 performance across Scene Flow, KITTI, Middlebury, ETH3D, and DrivingStereo. Our results establish that lightweight architectures can indeed generalize across domains without sacrificing accuracy. \href{https://mdraqibkhan.github.io/Litematch}{\textcolor{blue}{Code}}

Problem

• stereo matching
• zero-shot generalization
• cost volume
• domain generalization
• lightweight architecture

Innovation

• zero-shot stereo matching
• cost volume stabilization
• lightweight architecture
• frequency-domain encoding
• CVC-Loss