Deep Parametric Continuous Convolutional Neural Networks

被引:2408
作者
Wang, Shenlong [1 ,3 ]
Suo, Simon [2 ,3 ]
Ma, Wei-Chiu [3 ]
Pokrovsky, Andrei [3 ]
Urtasun, Raquel [1 ,3 ]
机构
[1] Univ Toronto, Toronto, ON, Canada
[2] Univ Waterloo, Waterloo, ON, Canada
[3] Uber Adv Technol Grp, Pittsburgh, PA 15201 USA
来源
2018 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR) | 2018年
关键词
D O I
10.1109/CVPR.2018.00274
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Standard convolutional neural networks assume a grid structured input is available and exploit discrete convolutions as their fundamental building blocks. This limits their applicability to many real-world applications. In this paper we propose Parametric Continuous Convolution, a new learnable operator that operates over non-grid structured data. The key idea is to exploit parameterized kernel functions that span the full continuous vector space. This generalization allows us to learn over arbitrary data structures as long as their support relationship is computable. Our experiments show significant improvement over the state-of-the-art in point cloud segmentation of indoor and outdoor scenes, and lidar motion estimation of driving scenes.
引用
收藏
页码:2589 / 2597
页数:9
相关论文
共 30 条
[1]  
[Anonymous], 2017, CVPR
[2]  
[Anonymous], 2017, SUBMANIFOLD SPARSE C
[3]  
[Anonymous], 2017, CVPR
[4]  
[Anonymous], 2017, Pointnet: Deep
[5]  
[Anonymous], 2016, CVPR
[6]  
[Anonymous], 2016, NIPS
[7]  
[Anonymous], 2017, CVPR
[8]  
[Anonymous], 2014, ICLR
[9]  
[Anonymous], CONTINUOUS FILTER CO
[10]  
[Anonymous], 2016, POINTNET DEEP LEARNI