Deep Stereo Matching Research Articles

Multiple Frequency Inputs and Context-Guided Attention Network for Stereo Disparity Estimation

Deep learning-based methods have made remarkable progress for stereo matching in terms of accuracy. However, two issues still hinder producing a perfect disparity map: (1) blurred boundaries and the discontinuous disparity of a continuous region on disparity estimation maps, and (2) a lack of effective means to restore resolution precisely. In this paper, we propose to utilize multiple frequency inputs and an attention mechanism to construct the deep stereo matching model. Specifically, high-frequency and low-frequency information of the input image together with the RGB image are fed into a feature extraction network with 2D convolutions. It is conducive to produce a distinct boundary and continuous disparity of the smooth region on disparity maps. To regularize the 4D cost volume for disparity regression, we propose a 3D context-guided attention module for stacked hourglass networks, where high-level cost volumes as context guide low-level features to obtain high-resolution yet precise feature maps. The proposed approach achieves competitive performance on SceneFlow and KITTI 2015 datasets.

Deep Stereo Matching With Hysteresis Attention and Supervised Cost Volume Construction.

Stereo matching disparity prediction for rectified image pairs is of great importance to many vision tasks such as depth sensing and autonomous driving. Previous work on the end-to-end unary trained networks follows the pipeline of feature extraction, cost volume construction, matching cost aggregation, and disparity regression. In this paper, we propose a deep neural network architecture for stereo matching aiming at improving the first and second stages of the matching pipeline. Specifically, we show a network design inspired by hysteresis comparator in the circuit as our attention mechanism. Our attention module is multiple-block and generates an attentive feature directly from the input. The cost volume is constructed in a supervised way. We try to use data-driven to find a good balance between informativeness and compactness of extracted feature maps. The proposed approach is evaluated on several benchmark datasets. Experimental results demonstrate that our method outperforms previous methods on SceneFlow, KITTI 2012, and KITTI 2015 datasets.