Here are the features used in the Exemplar-SVM framework.
Compile with features_compile.m script
** esvm_features.m: a Matlab function which calles one of the mexed functions. It returns the dimensionality of the per-cell features when called without arguments. The idea is that you can add your own features and 31 is never hard-coded in the codebase.
** features_pedro.cc: the 31-D features from voc-release-3.1
** features_raw.cc: the same features as features_pedro.cc but without the contrast normalization. They are definitely worse than when using normalization, but the effect has not been fully studied.
** fconvblas.cc: a fast convolution procedure from from voc-release-4