Gradual interaction network for stereo matching

Aixin Chong, Hui Yin, Qianqian Du, Yanting Liu, Ming Han

Abstract: Highlights•For obtaining the features that are beneficial to improving the accuracy of stereo matching, we propose the first stereo matching model which can realize the bidirectional interaction between feature extraction and disparity estimation.•The Stereo Space Grade Module (SSGM) is proposed, which utilizes probability volume and disparity map to estimate the disparity confidence map and refine the disparity search space.•The Disparity Feedback Filter Module (DFFM) is proposed, which can feedback disparity information to feature extraction.•We explore a new disparity regression approach named double branch up-sampling strategy, which is conducive to improve the accuracy of disparity estimation in edge regions.