Hierarchical vs. flat classification package
--------------------------------------------
Written by Thomas Hoyoux, 2016
Check our paper "Can computer vision problems benefit from structured hierarchical
classification?" at http://link.springer.com/article/10.1007/s00138-016-0763-9

Install
-------
1. Copy this package in some place where you can access it within MATLAB
2. Launch MATLAB and cd to the root folder of this package
3. cd to the svm-struct-matlab folder
4. Under Linux and Mac: type 'make'
   Does not work: type '!make -f Makefile MEX=/Applications/MATLAB_R2014b.app/bin/mex'
     (change your MATLAB version accordingly)
   Under Windows or if the mex compilation of svm-struct-matlab gave errors,
     - refer to http://www.robots.ox.ac.uk/~vedaldi/svmstruct.html
     - update your system, mex compiler
     - google your errors
     - ... if you're stuck, well, at least SkNN works out-of-the-box with this package.
5. cd back to the root folder of this package
6. you're good to go.

Warning
-------
SVM-Struct is much slower to train than SkNN (yet it gives better results).

Top functions
-------------
- Evaluation on the PSB dataset: 'evalPSB.m'
- Evaluation on the CK+ dataset: 'evalCKplus.m'
- Evaluation on simulation data: 'evalSimulation.m'
  N.B. You may need to generate data before using 'evalSimulation.m'
       -> use 'generateSimulationData.m'

To customize your hierarchical vs. flat classification evaluations,
check the general-purpose function 'evalStructLearning.m'

Where is the doc?
-----------------
The documentation is written in the source files.
Three possibilities:
1. Read the source files with your favorite text editor
2. Type 'help <some-function>' in the MATLAB prompt
3. Type 'doc <some-function>' in the MATLAB prompt

Send all questions/remarks to thomas.hoyoux@gmail.com
