Aggregating local descriptors into a compact representation
We provide a simple Matlab implementation of the method described
in this paper:
author = "Herv\'e J\'egou and Matthijs Douze and Cordelia Schmid and Patrick P\'erez",
title = "Aggregating local descriptors into a compact image representation",
booktitle = "IEEE Conference on Computer Vision \& Pattern Recognition",
month = "jun",
year = "2010",
url = "http://lear.inrialpes.fr/pubs/2010/JDSP10"
This technique allows very large scale image indexing with a very compact image representation
(typically 16 to 40 bytes per image).
Download Matlab package (71MB)
This package was written by Herve Jegou, June 2010.
Copyright (C) INRIA 2010.
This version is provided for research purposes only, without any support/guarantee.
The package makes use of Mex-files.
It has been tested on Linux and MacOS.
We will not provide a Windows version.
Remark: This is an independent implementation of the method, not the one we used to generate the results of our the paper.
- It is sufficient to reproduce the results in terms of accuracy and memory
- but the timings are not as good as the one we obtained with our optimized C implementation.
- Alhtough the results are the same as those reported in our paper,
Using powerlaw component-wise normalization further improves the results.
See the improved journal version.
This paper makes use of several distinct packages. All instructions are given in the README file,
however you might be interested at looking at them:
- This package requires the Yael library.
- The indexing/encoding method is built upon the one described
in this paper
and available from this webpage
- The evaluation dataset used in the package is the Holidays dataset.
This package is outdated now. The webpage containing the improved version of the proposed image search system is available on this webpage.