Learning binary codes for maximum inner product search

Shen, Fumin, Liu, Wei, Zhang, Shaoting, Yang, Yang and Shen, Heng Tao (2015). Learning binary codes for maximum inner product search. In: 2015 IEEE International Conference on Computer Vision: ICCV 2015. IEEE International Conference on Computer Vision (ICCV), Santiago, Chile, (4148-4156). 13-16 December 2015. doi:10.1109/ICCV.2015.472

Attached Files (Some files may be inaccessible until you login with your UQ eSpace credentials)
Name Description MIMEType Size Downloads
UQ377684_OA.pdf Full text (open access) application/pdf 16.78MB 0

Author Shen, Fumin
Liu, Wei
Zhang, Shaoting
Yang, Yang
Shen, Heng Tao
Title of paper Learning binary codes for maximum inner product search
Conference name IEEE International Conference on Computer Vision (ICCV)
Conference location Santiago, Chile
Conference dates 13-16 December 2015
Convener IEEE
Proceedings title 2015 IEEE International Conference on Computer Vision: ICCV 2015   Check publisher's open access policy
Journal name 2015 Ieee International Conference On Computer Vision (Iccv)   Check publisher's open access policy
Place of Publication Piscataway, NJ, United States
Publisher IEEE
Publication Year 2015
Sub-type Fully published paper
DOI 10.1109/ICCV.2015.472
Open Access Status File (Publisher version)
ISBN 9781467383912
ISSN 1550-5499
Volume 11-18-December-2015
Start page 4148
End page 4156
Total pages 9
Language eng
Formatted Abstract/Summary
Binary coding or hashing techniques are recognized to accomplish efficient near neighbor search, and have thus attracted broad interests in the recent vision and learning studies. However, such studies have rarely been dedicated to Maximum Inner Product Search (MIPS), which plays a critical role in various vision applications. In this paper, we investigate learning binary codes to exclusively handle the MIPS problem. Inspired by the latest advance in asymmetric hashing schemes, we propose an asymmetric binary code learning framework based on inner product fitting. Specifically, two sets of coding functions are learned such that the inner products between their generated binary codes can reveal the inner products between original data vectors. We also propose an alternative simpler objective which maximizes the correlations between the inner products of the produced binary codes and raw data vectors. In both objectives, the binary codes and coding functions are simultaneously learned without continuous relaxations, which is the key to achieving high-quality binary codes. We evaluate the proposed method, dubbed Asymmetric Innerproduct Binary Coding (AIBC), relying on the two objectives
on several large-scale image datasets. Both of them are superior to the state-of-the-art binary coding and hashing methods in performing MIPS tasks.
Q-Index Code E1
Q-Index Status Provisional Code
Institutional Status UQ

 
Versions
Version Filter Type
Citation counts: TR Web of Science Citation Count  Cited 16 times in Thomson Reuters Web of Science Article | Citations
Scopus Citation Count Cited 23 times in Scopus Article | Citations
Google Scholar Search Google Scholar
Created: Mon, 25 Jan 2016, 21:27:09 EST by Dr Heng Tao Shen on behalf of School of Information Technol and Elec Engineering