Grouped Correlation Aggregation with Propagation for Stereo Matching

Boning Zhang, Zekai Xu, Zhezhi He, Hongzi Zhu, Yan Li, Limin Xiao, Weifeng Zhang

Keywords: stereo matching, computer vision

Abstract:

Iterative optimization-based methods have dominated the field of stereo matching with extraordinary precision and speed. However, these methods still suffer from low iteration efficiency and insufficient correlation volume with low utilization rates.
As the countermeasure, we propose grouped correlation aggregation with propagation, a novel stereo matching method inspired by traditional methods.
We design an efficient updater to improve the performance of single iteration optimization. To alleviate the problems of correlation volume, a novel grouped window shifting mechanism and a contour-aware aggregation modified from semi-global matching (SGM) have been introduced. Our method outperforms all methods in zero-shot generalization and ranks 1st on ETH3D among published works.
Additionally, we conducted targeted inference optimization on the video stream and demonstrated the improvement in frame rate without sacrificing accuracy through experiments on the simulator.
Finally, a real-world binocular system is deployed to qualitatively demonstrate the practicality of our method.

Primary Area: applications to computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 6577