EdgeStereo: A Context Integrated Residual Pyramid Network for Stereo Matching

Recently convolutional neural network (CNN) promotes the development of stereo matching greatly. Especially those end-to-end stereo methods achieve best performance. However less attention is paid on encoding context information, simplifying two-stage disparity learning pipeline and improving details in disparity maps. Differently we focus on these problems. Firstly, we propose an one-stage context pyramid based residual pyramid network (CP-RPN) for disparity estimation, in which a context pyramid is embedded to encode multi-scale context clues explicitly. Next, we design a CNN based multi-task learning network called EdgeStereo to recover missing details in disparity maps, utilizing mid-level features from edge detection task. In EdgeStereo, CP-RPN is integrated with a proposed edge detector HED$_\beta$ based on two-fold multi-task interactions. The end-to-end EdgeStereo outputs the edge map and disparity map directly from a stereo pair without any post-processing or regularization. We discover that edge detection task and stereo matching task can help each other in our EdgeStereo framework. Comprehensive experiments on stereo benchmarks such as Scene Flow and KITTI 2015 show that our method achieves state-of-the-art performance.