-
Abstract
-
We present a new method that addresses the problem of approximate nearest neighbor search via partitioning the feature space. The proposed random exemplar hashing algorithm can be used to generate binary codes of data to facilitate nearest neighbor search within large datasets.
Inspired by the idea of using an ensemble of classifiers for discriminative learning, we devise an unsupervised learning algorithm to explore the feature space with respect to randomly selected exemplars.
Experimental results on three large datasets show that our method outperforms the state-of-the-art, especially on the cases of longer binary codes.
-
Paper
- Tao-Yen Tang, Tzu-Wei Huang, and Hwann-Tzong Chen: Random Exemplar Hashing. DDS 2013.
-
Code
-
Download Matlab implementation
Run 'demo.m' in Matlab
It takes about one minute to train and generate hash codes for the first time.
Hit 'Enter' after training or run 'demo.m' again to try random queries.
Required data and tools:
1.
CIFAR 100 dataset
[Ref]
"Learning multiple layers of features from tiny images",
Alex Krizhevsky, 2009.
2.
Matlab code for generating Gist descriptors
[Ref]
"Modeling the shape of the scene: a holistic representation of the spatial envelope",
Aude Oliva, Antonio Torralba, IJCV, Vol. 42(3): 145-175, 2001.
3.
LIBSVM
[Ref]
"LIBSVM : a library for support vector machines",
C.-C. Chang and C.-J. Lin.
ACM Transactions on Intelligent Systems and Technology, 2:27:1--27:27, 2011.
4.
Matlab code 'subaxis' for subplot
Last updated: 07 November 2013