Practical Deep Stereo (PDS): Toward applications-friendly deep stereo matching

TL;DR

The paper introduces Practical Deep Stereo (PDS), a memory-efficient and adaptable deep stereo matching network that performs well on large images and across various disparity ranges, suitable for real-world applications.

Contribution

PDS presents a novel architecture with bottleneck modules for reduced memory use and a new loss function for disparity range flexibility, advancing practical deep stereo matching.

Findings

Outperforms recent state-of-the-art methods on FlyingThings3D and KITTI datasets.

Handles larger images during training due to design optimizations.

Remains effective across different disparity ranges without re-training.

Abstract

End-to-end deep-learning networks recently demonstrated extremely good perfor- mance for stereo matching. However, existing networks are difficult to use for practical applications since (1) they are memory-hungry and unable to process even modest-size images, (2) they have to be trained for a given disparity range. The Practical Deep Stereo (PDS) network that we propose addresses both issues: First, its architecture relies on novel bottleneck modules that drastically reduce the memory footprint in inference, and additional design choices allow to handle greater image size during training. This results in a model that leverages large image context to resolve matching ambiguities. Second, a novel sub-pixel cross- entropy loss combined with a MAP estimator make this network less sensitive to ambiguous matches, and applicable to any disparity range without re-training. We compare PDS to state-of-the-art methods published over the recent months, and demonstrate its superior performance on FlyingThings3D and KITTI sets.