Progressive Shape-Distribution-Encoder for Learning 3D Shape Representation

Jin Xie, Fan Zhu, Guoxian Dai, Ling Shao, Yi Fang

Research output: Contribution to journalArticle

Abstract

Since there are complex geometric variations with 3D shapes, extracting efficient 3D shape features is one of the most challenging tasks in shape matching and retrieval. In this paper, we propose a deep shape descriptor by learning shape distributions at different diffusion time via a progressive shape-distribution-encoder (PSDE). First, we develop a shape distribution representation with the kernel density estimator to characterize the intrinsic geometry structures of 3D shapes. Then, we propose to learn a deep shape feature through an unsupervised PSDE. Specially, the unsupervised PSDE aims at modeling the complex non-linear transform of the estimated shape distributions between consecutive diffusion time. In order to characterize the intrinsic structures of 3D shapes more efficiently, we stack multiple PSDEs to form a network structure. Finally, we concatenate all neurons in the middle hidden layers of the unsupervised PSDE network to form an unsupervised shape descriptor for retrieval. Furthermore, by imposing an additional constraint on the outputs of all hidden layers, we propose a supervised PSDE to form a supervised shape descriptor. For each hidden layer, the similarity between a pair of outputs from the same class is as large as possible and the similarity between a pair of outputs from different classes is as small as possible. The proposed method is evaluated on three benchmark 3D shape data sets with large geometric variations, i.e., McGill, SHREC'10 ShapeGoogle, and SHREC'14 Human data sets, and the experimental results demonstrate the superiority of the proposed method to the existing approaches.

Original languageEnglish (US)
Article number7812756
Pages (from-to)1231-1242
Number of pages12
JournalIEEE Transactions on Image Processing
Volume26
Issue number3
DOIs
StatePublished - Mar 1 2017

Fingerprint

Electric power distribution
Neurons
Geometry

Keywords

  • 3D shape retrieval
  • denoising auto-encoder
  • heat diffusion
  • heat kernel signature
  • shape descriptor

ASJC Scopus subject areas

  • Software
  • Computer Graphics and Computer-Aided Design

Cite this

Progressive Shape-Distribution-Encoder for Learning 3D Shape Representation. / Xie, Jin; Zhu, Fan; Dai, Guoxian; Shao, Ling; Fang, Yi.

In: IEEE Transactions on Image Processing, Vol. 26, No. 3, 7812756, 01.03.2017, p. 1231-1242.

Research output: Contribution to journalArticle

Xie, Jin ; Zhu, Fan ; Dai, Guoxian ; Shao, Ling ; Fang, Yi. / Progressive Shape-Distribution-Encoder for Learning 3D Shape Representation. In: IEEE Transactions on Image Processing. 2017 ; Vol. 26, No. 3. pp. 1231-1242.
@article{c7f30cfabef7413e8910b15aae559df1,
title = "Progressive Shape-Distribution-Encoder for Learning 3D Shape Representation",
abstract = "Since there are complex geometric variations with 3D shapes, extracting efficient 3D shape features is one of the most challenging tasks in shape matching and retrieval. In this paper, we propose a deep shape descriptor by learning shape distributions at different diffusion time via a progressive shape-distribution-encoder (PSDE). First, we develop a shape distribution representation with the kernel density estimator to characterize the intrinsic geometry structures of 3D shapes. Then, we propose to learn a deep shape feature through an unsupervised PSDE. Specially, the unsupervised PSDE aims at modeling the complex non-linear transform of the estimated shape distributions between consecutive diffusion time. In order to characterize the intrinsic structures of 3D shapes more efficiently, we stack multiple PSDEs to form a network structure. Finally, we concatenate all neurons in the middle hidden layers of the unsupervised PSDE network to form an unsupervised shape descriptor for retrieval. Furthermore, by imposing an additional constraint on the outputs of all hidden layers, we propose a supervised PSDE to form a supervised shape descriptor. For each hidden layer, the similarity between a pair of outputs from the same class is as large as possible and the similarity between a pair of outputs from different classes is as small as possible. The proposed method is evaluated on three benchmark 3D shape data sets with large geometric variations, i.e., McGill, SHREC'10 ShapeGoogle, and SHREC'14 Human data sets, and the experimental results demonstrate the superiority of the proposed method to the existing approaches.",
keywords = "3D shape retrieval, denoising auto-encoder, heat diffusion, heat kernel signature, shape descriptor",
author = "Jin Xie and Fan Zhu and Guoxian Dai and Ling Shao and Yi Fang",
year = "2017",
month = "3",
day = "1",
doi = "10.1109/TIP.2017.2651408",
language = "English (US)",
volume = "26",
pages = "1231--1242",
journal = "IEEE Transactions on Image Processing",
issn = "1057-7149",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
number = "3",

}

TY - JOUR

T1 - Progressive Shape-Distribution-Encoder for Learning 3D Shape Representation

AU - Xie, Jin

AU - Zhu, Fan

AU - Dai, Guoxian

AU - Shao, Ling

AU - Fang, Yi

PY - 2017/3/1

Y1 - 2017/3/1

N2 - Since there are complex geometric variations with 3D shapes, extracting efficient 3D shape features is one of the most challenging tasks in shape matching and retrieval. In this paper, we propose a deep shape descriptor by learning shape distributions at different diffusion time via a progressive shape-distribution-encoder (PSDE). First, we develop a shape distribution representation with the kernel density estimator to characterize the intrinsic geometry structures of 3D shapes. Then, we propose to learn a deep shape feature through an unsupervised PSDE. Specially, the unsupervised PSDE aims at modeling the complex non-linear transform of the estimated shape distributions between consecutive diffusion time. In order to characterize the intrinsic structures of 3D shapes more efficiently, we stack multiple PSDEs to form a network structure. Finally, we concatenate all neurons in the middle hidden layers of the unsupervised PSDE network to form an unsupervised shape descriptor for retrieval. Furthermore, by imposing an additional constraint on the outputs of all hidden layers, we propose a supervised PSDE to form a supervised shape descriptor. For each hidden layer, the similarity between a pair of outputs from the same class is as large as possible and the similarity between a pair of outputs from different classes is as small as possible. The proposed method is evaluated on three benchmark 3D shape data sets with large geometric variations, i.e., McGill, SHREC'10 ShapeGoogle, and SHREC'14 Human data sets, and the experimental results demonstrate the superiority of the proposed method to the existing approaches.

AB - Since there are complex geometric variations with 3D shapes, extracting efficient 3D shape features is one of the most challenging tasks in shape matching and retrieval. In this paper, we propose a deep shape descriptor by learning shape distributions at different diffusion time via a progressive shape-distribution-encoder (PSDE). First, we develop a shape distribution representation with the kernel density estimator to characterize the intrinsic geometry structures of 3D shapes. Then, we propose to learn a deep shape feature through an unsupervised PSDE. Specially, the unsupervised PSDE aims at modeling the complex non-linear transform of the estimated shape distributions between consecutive diffusion time. In order to characterize the intrinsic structures of 3D shapes more efficiently, we stack multiple PSDEs to form a network structure. Finally, we concatenate all neurons in the middle hidden layers of the unsupervised PSDE network to form an unsupervised shape descriptor for retrieval. Furthermore, by imposing an additional constraint on the outputs of all hidden layers, we propose a supervised PSDE to form a supervised shape descriptor. For each hidden layer, the similarity between a pair of outputs from the same class is as large as possible and the similarity between a pair of outputs from different classes is as small as possible. The proposed method is evaluated on three benchmark 3D shape data sets with large geometric variations, i.e., McGill, SHREC'10 ShapeGoogle, and SHREC'14 Human data sets, and the experimental results demonstrate the superiority of the proposed method to the existing approaches.

KW - 3D shape retrieval

KW - denoising auto-encoder

KW - heat diffusion

KW - heat kernel signature

KW - shape descriptor

UR - http://www.scopus.com/inward/record.url?scp=85015321850&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85015321850&partnerID=8YFLogxK

U2 - 10.1109/TIP.2017.2651408

DO - 10.1109/TIP.2017.2651408

M3 - Article

AN - SCOPUS:85015321850

VL - 26

SP - 1231

EP - 1242

JO - IEEE Transactions on Image Processing

JF - IEEE Transactions on Image Processing

SN - 1057-7149

IS - 3

M1 - 7812756

ER -