Widening siamese architectures for stereo matching

TL;DR

This paper enhances stereo matching by improving feature extraction with CNNs and simplifying correlation learning, leading to promising results on benchmark datasets without the need for refinement.

Contribution

It introduces a novel approach that uses standard CNNs for better feature extraction and a simple space aggregation to simplify correlation learning in stereo matching.

Findings

Improved feature quality with CNNs enhances stereo matching accuracy.

Simplified space aggregation reduces complexity of correlation learning.

Results show strong performance on benchmark datasets without refinement.

Abstract

Computational stereo is one of the classical problems in computer vision. Numerous algorithms and solutions have been reported in recent years focusing on developing methods for computing similarity, aggregating it to obtain spatial support and finally optimizing an energy function to find the final disparity. In this paper, we focus on the feature extraction component of stereo matching architecture and we show standard CNNs operation can be used to improve the quality of the features used to find point correspondences. Furthermore, we propose a simple space aggregation that hugely simplifies the correlation learning problem. Our results on benchmark data are compelling and show promising potential even without refining the solution.