SMD-Nets: Stereo Mixture Density Networks

TL;DR

SMD-Nets introduces a novel stereo matching framework using bimodal mixture densities that enhances boundary sharpness, resolution, and uncertainty modeling, applicable across various architectures and validated on synthetic and real datasets.

Contribution

The paper presents a flexible mixture density approach for stereo disparity estimation that improves boundary accuracy and resolution while modeling uncertainty.

Findings

Enhanced disparity accuracy near object boundaries

Ability to generate ultra high-resolution disparity maps

Improved performance across different stereo architectures

Abstract

Despite stereo matching accuracy has greatly improved by deep learning in the last few years, recovering sharp boundaries and high-resolution outputs efficiently remains challenging. In this paper, we propose Stereo Mixture Density Networks (SMD-Nets), a simple yet effective learning framework compatible with a wide class of 2D and 3D architectures which ameliorates both issues. Specifically, we exploit bimodal mixture densities as output representation and show that this allows for sharp and precise disparity estimates near discontinuities while explicitly modeling the aleatoric uncertainty inherent in the observations. Moreover, we formulate disparity estimation as a continuous problem in the image domain, allowing our model to query disparities at arbitrary spatial precision. We carry out comprehensive experiments on a new high-resolution and highly realistic synthetic stereo dataset, consisting of stereo pairs at 8Mpx resolution, as well as on real-world stereo datasets. Our experiments demonstrate increased depth accuracy near object boundaries and prediction of ultra high-resolution disparity maps on standard GPUs. We demonstrate the flexibility of our technique by improving the performance of a variety of stereo backbones.