Overview

Stereo matching finds corresponding points in two camera views, usually rectified so their horizontal displacement, or disparity, can be converted into depth. Modern methods build cost volumes or correlation maps and refine disparities with recurrent updates, while attention and transformer models use wider context to handle occlusions, textureless surfaces, and thin structures. Other approaches incorporate monocular depth, geometric or foundation-model priors, self-supervision, and synthetic data to improve transfer across settings such as driving, underwater imaging, satellites, and omnidirectional cameras. The central tension is achieving accurate, generalizable depth under changing appearance and sparse supervision while keeping inference fast and compact enough for deployment.

Latest papers

CardsList

62 papers6 papers in the last four weeks, against 1 the four weeks before. 0.0% of all new papers.

  1. ESMStereo: Enhanced ShuffleMixer Disparity Upsampling for Real-Time and Accurate Stereo Matching

    Jun 26, 2025Mahmoud Tahmasebi, Saif Huq, Kevin Meehan +1Stereo Matching

  2. Motif Channel Opened in a White-Box: Stereo Matching via Motif Correlation Graph

    Nov 19, 2024Ziyang Chen, Yongjun Zhang, Wenting Li +3Representation LearningStereo Matching