
Matlab scripts for Hierarchical Models of Scenes, Objects, & Parts
******************************************************************

Variants of these Matlab scripts were used to run the experiments 
reported in the following journal publication:
  Describing Visual Scenes Using Transformed Objects and Parts
  E. B. Sudderth, A. Torralba, W. T. Freeman, and A. S. Willsky
  Int. J. Comp. Vision, vol. 77, pp. 291-330, 2008.

These scripts provide an example of how to use the learning code with image 
databases.  The first part of the script name indicates the type of model:
* shareLDA:  Parametric, latent Dirichlet allocation (LDA) model for sharing 
             parts among multiple object categories.  IJCV Sec. 3.2, 5.3.
* shareHDP:  Nonparametric, hierarchical Dirichlet process (HDP) model for 
             sharing parts among multiple object categories.  IJCV Sec. 4, 5.
* sceneLDA:  Parametric, contextual model for scenes containing a fixed,
             known set of object categories.  IJCV Sec. 6, 9.
* scenepartTDP:  Nonparametric, transformed Dirichlet process (TDP) model
             for scenes with unknown sets of objects.  IJCV Sec. 8, 9.

Files *_script.m both train models, and evaluate test performance, using
parameters set at the top of the script file.  For code which computes features
needed by these scripts, see subdirectories databaseSharing & databaseScenes.

Files *_vis*.m visualize either trained models or test performance in various
ways - see comments for details.  In many cases, variables at the top of these
scripts will need to be changed to match the filenames used by training scripts.

