TU Darmstadt / ULB / TUbiblio

Unsupervised 3D object retrieval with parameter-free hierarchical clustering

Getto, Roman and Kuijper, Arjan and Fellner, Dieter W. (2017):
Unsupervised 3D object retrieval with parameter-free hierarchical clustering.
In: CGI'17 : Proceedings of the Computer Graphics International Conference, pp. 1-6,
New York, ACM, Yokohama, Japan, ISBN 978-1-4503-5228-4/17/06,
DOI: 10.1145/3095140.3095147,
[Conference or Workshop Item]

Abstract

In 3D object retrieval, additional knowledge like user input, classification information or database dependent configured parameters are rarely available in real scenarios. For example, meta data about 3D objects is seldom if the objects are not within a well-known evaluation database. We propose an algorithm which improves the performance of unsupervised 3D object retrieval without using any additional knowledge. For the computation of the distances in our system any descriptor can be chosen; we use the Panorama-descriptor. Our algorithm uses a precomputed parameter-free agglomerative hierarchical clustering and combines the information of the hierarchy of clusters with the individual distances to improve a single object query. Additionally, we propose an adaption algorithm for the cases that new objects are added frequently to the database. We evaluate our approach with 6 databases including a total of 13271 objects in 481 classes. We show that our algorithm improves the average precision in an unsupervised scenario without any parameter configuration.

Item Type: Conference or Workshop Item
Erschienen: 2017
Creators: Getto, Roman and Kuijper, Arjan and Fellner, Dieter W.
Title: Unsupervised 3D object retrieval with parameter-free hierarchical clustering
Language: English
Abstract:

In 3D object retrieval, additional knowledge like user input, classification information or database dependent configured parameters are rarely available in real scenarios. For example, meta data about 3D objects is seldom if the objects are not within a well-known evaluation database. We propose an algorithm which improves the performance of unsupervised 3D object retrieval without using any additional knowledge. For the computation of the distances in our system any descriptor can be chosen; we use the Panorama-descriptor. Our algorithm uses a precomputed parameter-free agglomerative hierarchical clustering and combines the information of the hierarchy of clusters with the individual distances to improve a single object query. Additionally, we propose an adaption algorithm for the cases that new objects are added frequently to the database. We evaluate our approach with 6 databases including a total of 13271 objects in 481 classes. We show that our algorithm improves the average precision in an unsupervised scenario without any parameter configuration.

Title of Book: CGI'17 : Proceedings of the Computer Graphics International Conference
Place of Publication: New York
Publisher: ACM
ISBN: 978-1-4503-5228-4/17/06
Uncontrolled Keywords: 3D Object retrieval, Classifications, Clustering
Divisions: 20 Department of Computer Science
20 Department of Computer Science > Interactive Graphics Systems
20 Department of Computer Science > Mathematical and Applied Visual Computing
Event Location: Yokohama, Japan
Date Deposited: 05 May 2020 14:59
DOI: 10.1145/3095140.3095147
Official URL: https://doi.org/10.1145/3095140.3095147
Export:
Suche nach Titel in: TUfind oder in Google
Send an inquiry Send an inquiry

Options (only for editors)
Show editorial Details Show editorial Details