Learning class-specific descriptors for deformable shapes using localized spectral convolutional networks

被引:158
作者
Boscaini, D. [1 ]
Masci, J. [1 ]
Mezi, S. [2 ]
Bronstein, M. M. [1 ]
Castellani, U. [2 ]
Vandergheynst, P. [3 ]
机构
[1] Univ Lugano USI, Fac Informat, Inst Computat Sci, Lugano, Switzerland
[2] Univ Verona, Dept Informat, I-37100 Verona, Italy
[3] Ecole Polytech Fed Lausanne, Dept Elect Engn, CH-1015 Lausanne, Switzerland
基金
欧洲研究理事会;
关键词
D O I
10.1111/cgf.12693
中图分类号
TP31 [计算机软件];
学科分类号
081202 ; 0835 ;
摘要
In this paper, we propose a generalization of convolutional neural networks (CNN) to non-Euclidean domains for the analysis of deformable shapes. Our construction is based on localized frequency analysis (a generalization of the windowed Fourier transform to manifolds) that is used to extract the local behavior of some dense intrinsic descriptor, roughly acting as an analogy to patches in images. The resulting local frequency representations are then passed through a bank of filters whose coefficient are determined by a learning procedure minimizing a task-specific cost. Our approach generalizes several previous methods such as HKS, WKS, spectral CNN, and GPS embeddings. Experimental results show that the proposed approach allows learning class-specific shape descriptors significantly outperforming recent state-of-the-art methods on standard benchmarks.
引用
收藏
页码:13 / 23
页数:11
相关论文
共 41 条
[1]   SCAPE: Shape Completion and Animation of People [J].
Anguelov, D ;
Srinivasan, P ;
Koller, D ;
Thrun, S ;
Rodgers, J ;
Davis, J .
ACM TRANSACTIONS ON GRAPHICS, 2005, 24 (03) :408-416
[2]  
[Anonymous], 2015, ARXIV150106297
[3]  
[Anonymous], P NORDIA
[4]  
[Anonymous], P SCIPY JUN
[5]  
[Anonymous], PAMI
[6]  
[Anonymous], 2003, Visualization and Mathematics III. Mathematics and Visualization
[7]  
[Anonymous], 2008, P CVPR
[8]  
[Anonymous], P SGP
[9]  
[Anonymous], 2014, P ICLR
[10]  
[Anonymous], ARXIV13075708