Skip to content
master
Switch branches/tags
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 

BoW Translation of Dense Trajectory Features

This is the code that accompanies the paper

A. Gupta, A. Shafaei, J. J. Little and R. J. Woodham. Unlabelled 3D Motion Examples Improve Cross-View Action Recognition. In BMVC, 2014. See project page for more information.

The basic idea is to learn a transformation function for BoW features that translates the feature descriptor as if they were seen from another viewpoint. We use this idea to perform cross-view action recognition.

How to run

In order to run this code you need to have

We use VLFeat to calculate Homogenous Kernel Maps of chi-squared kernel. LibLinear is used to train an SVM. You can easily replace these parts with any other implementations you like.

To avoid potential compatibility problems with the future versions we have also included a copy of the liblinear in the 3rd party folder.

Before you can run this code you need to download and extract the dataset. If you're using Unix based machines you simply need to navigate to the base folder and run ./prepare_dataset.sh.

If you are a windows user you can manually download and extract the dataset from here.

After that you can simply run classify_ixmas.m in Matlab.

What's in the dataset?

The dataset contains

  • IXMAS Precomputed and quantized dense trajectory features. You can also use the vocbulary inside these files to quantize your own dense trajectory features to use with this code.
  • Trained transition matrices for 426 viewpoint changes. The procedure to prepare this data is explained in the paper.

Do you have a question?

You can either contact Alireza Shafaei or Ankur Gupta.

About

BoW translation of dense trajectory features based on the viewpoint.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published