Camera Pose Matters: Improving Depth Prediction by Mitigating Pose Distribution Bias

被引:14
作者
Zhao, Yunhan [1 ]
Kong, Shu [2 ]
Fowlkes, Charless [1 ]
机构
[1] UC Irvine, Irvine, CA 92697 USA
[2] Carnegie Mellon Univ, Pittsburgh, PA 15213 USA
来源
2021 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, CVPR 2021 | 2021年
关键词
D O I
10.1109/CVPR46437.2021.01550
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Monocular depth predictors are typically trained on large-scale training sets which are naturally biased w.r.t the distribution of camera poses. As a result, trained predictors fail to make reliable depth predictions for testing examples captured under uncommon camera poses. To address this issue, we propose two novel techniques that exploit the camera pose during training and prediction. First, we introduce a simple perspective-aware data augmentation that synthesizes new training examples with more diverse views by perturbing the existing ones in a geometrically consistent manner. Second, we propose a conditional model that exploits the per-image camera pose as prior knowledge by encoding it as a part of the input. We show that jointly applying the two methods improves depth prediction on images captured under uncommon and even never-before-seen camera poses. We show that our methods improve performance when applied to a range of different predictor architectures. Lastly, we show that explicitly encoding the camera pose distribution improves the generalization performance of a synthetically trained depth predictor when evaluated on real images.
引用
收藏
页码:15754 / 15763
页数:10
相关论文
共 57 条
[1]   Building Rome in a Day [J].
Agarwal, Sameer ;
Furukawa, Yasutaka ;
Snavely, Noah ;
Simon, Ian ;
Curless, Brian ;
Seitz, Steven M. ;
Szeliski, Richard .
COMMUNICATIONS OF THE ACM, 2011, 54 (10) :105-112
[2]  
[Anonymous], 2018, ECCV, DOI DOI 10.1007/978-3-030-01234-2_47
[3]  
[Anonymous], 2017, CVPR, DOI DOI 10.1109/CVPR.2017.82
[4]  
[Anonymous], 2018, CVPR, DOI DOI 10.1109/CVPR.2018.00218
[5]  
[Anonymous], 2018, CVPR, DOI DOI 10.1109/CVPR.2018.00940
[6]  
[Anonymous], 2018, CVPR, DOI DOI 10.1109/CVPR.2018.00219
[7]  
[Anonymous], 2017, CVPR, DOI DOI 10.1109/CVPR.2017.596
[8]  
[Anonymous], 2014, CVPR, DOI DOI 10.1109/CVPR.2014.19
[9]   Simultaneous localization and mapping (SLAM): Part II [J].
Bailey, Tim ;
Durrant-Whyte, Hugh .
IEEE ROBOTICS & AUTOMATION MAGAZINE, 2006, 13 (03) :108-117
[10]  
Baradad Manel, 2020, CVPR