Adaptive Deconvolution-Based Stereo Matching Net for Local Stereo Matching

In deep learning-based local stereo matching methods, larger image patches usually bring better accuracy. However, it is unrealistic to increase the size of patch without restriction. Arbitrarily extending will change method into global method, and accuracy be saturated. We simplified existing Siamese convolutional network by reducing number parameters propose an efficient CNN based structure, namely adaptive deconvolution-based disparity net (ADSM net) adding deconvolution layers learn how enlarge input feature map for following convolution layers. Experimental results on KITTI2012 2015 datasets demonstrate that proposed can achieve a good trade-off between complexity.