Current stereo matching methods benefit a lot from the precise stereo estimation with Convolutional Neural Networks (CNNs). Nevertheless, patch-based siamese networks rely on the implicit assumption of constant depth within a window, which does not hold for slanted surfaces. Existing methods for handling slanted patches focus on post-processing. In contrast, we propose a novel module for matching cost networks to overcome this bias. Slanted objects appear horizontally stretched between stereo pairs, suggesting that the feature extraction in the horizontal direction should be different from that in the vertical direction. To tackle this distortion, we utilize asymmetric convolutions in our proposed module. Experimental results show that the proposed module in matching cost networks can achieve higher accuracy with fewer parameters compared to conventional methods.
Zhongjian MA
Chinese Academy of Sciences,University of Chinese Academy of Sciences
Dongzhen HUANG
Chinese Academy of Sciences
Baoqing LI
Chinese Academy of Sciences
Xiaobing YUAN
Chinese Academy of Sciences
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copy
Zhongjian MA, Dongzhen HUANG, Baoqing LI, Xiaobing YUAN, "Asymmetric Learning for Stereo Matching Cost Computation" in IEICE TRANSACTIONS on Information,
vol. E103-D, no. 10, pp. 2162-2167, October 2020, doi: 10.1587/transinf.2020EDP7002.
Abstract: Current stereo matching methods benefit a lot from the precise stereo estimation with Convolutional Neural Networks (CNNs). Nevertheless, patch-based siamese networks rely on the implicit assumption of constant depth within a window, which does not hold for slanted surfaces. Existing methods for handling slanted patches focus on post-processing. In contrast, we propose a novel module for matching cost networks to overcome this bias. Slanted objects appear horizontally stretched between stereo pairs, suggesting that the feature extraction in the horizontal direction should be different from that in the vertical direction. To tackle this distortion, we utilize asymmetric convolutions in our proposed module. Experimental results show that the proposed module in matching cost networks can achieve higher accuracy with fewer parameters compared to conventional methods.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2020EDP7002/_p
Copy
@ARTICLE{e103-d_10_2162,
author={Zhongjian MA, Dongzhen HUANG, Baoqing LI, Xiaobing YUAN, },
journal={IEICE TRANSACTIONS on Information},
title={Asymmetric Learning for Stereo Matching Cost Computation},
year={2020},
volume={E103-D},
number={10},
pages={2162-2167},
abstract={Current stereo matching methods benefit a lot from the precise stereo estimation with Convolutional Neural Networks (CNNs). Nevertheless, patch-based siamese networks rely on the implicit assumption of constant depth within a window, which does not hold for slanted surfaces. Existing methods for handling slanted patches focus on post-processing. In contrast, we propose a novel module for matching cost networks to overcome this bias. Slanted objects appear horizontally stretched between stereo pairs, suggesting that the feature extraction in the horizontal direction should be different from that in the vertical direction. To tackle this distortion, we utilize asymmetric convolutions in our proposed module. Experimental results show that the proposed module in matching cost networks can achieve higher accuracy with fewer parameters compared to conventional methods.},
keywords={},
doi={10.1587/transinf.2020EDP7002},
ISSN={1745-1361},
month={October},}
Copy
TY - JOUR
TI - Asymmetric Learning for Stereo Matching Cost Computation
T2 - IEICE TRANSACTIONS on Information
SP - 2162
EP - 2167
AU - Zhongjian MA
AU - Dongzhen HUANG
AU - Baoqing LI
AU - Xiaobing YUAN
PY - 2020
DO - 10.1587/transinf.2020EDP7002
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E103-D
IS - 10
JA - IEICE TRANSACTIONS on Information
Y1 - October 2020
AB - Current stereo matching methods benefit a lot from the precise stereo estimation with Convolutional Neural Networks (CNNs). Nevertheless, patch-based siamese networks rely on the implicit assumption of constant depth within a window, which does not hold for slanted surfaces. Existing methods for handling slanted patches focus on post-processing. In contrast, we propose a novel module for matching cost networks to overcome this bias. Slanted objects appear horizontally stretched between stereo pairs, suggesting that the feature extraction in the horizontal direction should be different from that in the vertical direction. To tackle this distortion, we utilize asymmetric convolutions in our proposed module. Experimental results show that the proposed module in matching cost networks can achieve higher accuracy with fewer parameters compared to conventional methods.
ER -