% -------------------------------------------
%
%  Learning Hierarchical Invariant
%  Sptio-temporal Features for 
%  Action Recognition with
%  Independent Subspace Analysis
%  Quoc V. Le, Will Y. Zou
%  Serena Y. Yeung, Andrew Y. Ng
%  [CVPR2011]
%
%	     Video features
% -------------------------------------------

This package is aimed at providing read-to-use
video features the stacked convolutional ISA 
model described in the paper, and functions to 
directly extract features from video.

The tool functions take in gray-scale video 
blocks in batches and compute corresponding 
features. Option is provided to use Jacket GPU 
implmentation for MATLAB.

- 'features/bases/' contains video features learned from the 
  Hollywood2 dataset

- Functions compute_features_1layer, compute_features
  are used to directly extract our features for local video blocks

  - compute_feature_1layer extract features using
    1 layer ISA

  - compute_feature extract features using
    2 layer stacked convolutional ISA

%Questions about code: 
%Will Zou (wzou *A*T* cs.stanford.edu)
