Subhransu Maji (smaji@cs.berkeley.edu)
Nov 13, 2009
--------------------------------------------------------
Contains code for experiments on the MNIST dataset.

rawfeatdemo.m - experiments using raw pixel features 
and SVMs trained using linear, intersection, polynomial
and rbf kernels. uses 200 training examples as default.

sphogdemo.m - experiments using spatial pyramid histograms
of oriented gradients and SVMs training using linear and
intersection kernel SVMs.
--------------------------------------------------------
Folder description:
+classifiers
    /liblinear-1.5-dense - LIBLINEAR modified for dense data
    /libsvm-mat-2.84-1-fast.v3 - LIBSVM with support for 
     intersection kernel SVM and fast prediction using the 
     technique proposed in our CVPR'08 paper.

+data 
    MNIST dataset downloaded from 

+io
    code to read/normalize the data

+feat-code
    code to compute sphog features 

+training
    code to train/test models
----------------------------------------------------------
