Matlab BoW_SVM_tutorial

This tutorial is intended to be a quick & simple introduction to the popular BoW-SVM 1-vs-all pipeline in action recognition.

The purpose is to illustrate and convey the main ideas practically and not to show state-of-the-art implementation details.

Three actions are considered: We want to decide whether an action is: foo1, foo2 or foo3.

NOTE: Features are generated randomly and bear no relation to action video sequences!

It is however straight-forward to replace the features generated in this tutorial with those extracted from images/videos.

Instructions for usage:

1a. This tutorial should run out-of-the-box for linux 64bit users since libsvm mex files are provided.

1b. Otherwise download LIBSVM from: http://www.csie.ntu.edu.tw/~cjlin/libsvm/ and place files in MATLAB path.

2. run bow_svm_tutorial.m

Code References:

1. CVML 2010 (Grenoble) tutorial: Classification of human actions in video.

2. Q.V. Le, W.Y. Zou, S.Y. Yeung, A.Y. Ng. Learning hierarchical spatio-temporal features for action recognition with independent subspace analysis.

Software available at http://ai.stanford.edu/~wzou/release.tar.gz

3. Chih-Chung Chang and Chih-Jen Lin, LIBSVM : a library for support vector machines. ACM Transactions on Intelligent Systems and Technology,

2:27:1--27:27, 2011. Software available at http://www.csie.ntu.edu.tw/~cjlin/libsvm

4. Synthetic Pinwheel Data (Matlab). Software available at http://hips.seas.harvard.edu/content/synthetic-pinwheel-data-matlab

Download