Skip to Main content Skip to Navigation
Conference papers

Searching in one billion vectors: re-rank with source coding

Hervé Jégou 1 Romain Tavenard 1 Matthijs Douze 2, 3 Laurent Amsaleg 1
1 TEXMEX - Multimedia content-based indexing
IRISA - Institut de Recherche en Informatique et Systèmes Aléatoires, Inria Rennes – Bretagne Atlantique
2 LEAR - Learning and recognition in vision
Inria Grenoble - Rhône-Alpes, LJK - Laboratoire Jean Kuntzmann, Grenoble INP - Institut polytechnique de Grenoble - Grenoble Institute of Technology
Abstract : Recent indexing techniques inspired by source coding have been shown successful to index billions of high-dimensional vectors in memory. In this paper, we propose an approach that re-ranks the neighbor hypotheses obtained by these compressed-domain indexing methods. In contrast to the usual post-verification scheme, which performs exact distance calculation on the short-list of hypotheses, the estimated distances are refined based on short quantization codes, to avoid reading the full vectors from disk. We have released a new public dataset of one billion 128-dimensional vectors and proposed an experimental setup to evaluate high dimensional indexing algorithms on a realistic scale. Experiments show that our method accurately and efficiently re-ranks the neighbor hypotheses using little memory compared to the full vectors representation.
Complete list of metadata

Cited literature [15 references]  Display  Hide  Download
Contributor : Hervé Jégou Connect in order to contact the contributor
Submitted on : Thursday, February 17, 2011 - 6:39:31 PM
Last modification on : Tuesday, October 19, 2021 - 11:13:04 PM
Long-term archiving on: : Thursday, June 30, 2011 - 1:51:24 PM


Files produced by the author(s)



Hervé Jégou, Romain Tavenard, Matthijs Douze, Laurent Amsaleg. Searching in one billion vectors: re-rank with source coding. ICASSP 2011 - International Conference on Acoustics, Speech and Signal Processing, May 2011, Prague, Czech Republic. pp.861-864, ⟨10.1109/ICASSP.2011.5946540⟩. ⟨inria-00566883⟩



Record views


Files downloads