attribute-features is a Python wrapper around Matlab/C++ attribute-based object recognition code from http://vision.cs.uiuc.edu/attributes/.
To use the python wrapper, you need to install Octave, Octave's img package, and
Python's oct2py package. Attribute features can then be extracted from image
by passing the filename to attributes.detect
.
In case of segmentation faults during detection, re-building MEX-files is the
first thing to try. Building MEX-files requires mkoctfile
Octave utility, if
it's not part of default Octave package for your distribution it can be found
under octave-devel or such. There are four files to build:
- feature_extraction/code/textons/anigauss_mex.c (the result .mex then needs to be linked to feature_extraction/code/textons/anigauss.mex)
- feature_extraction/code/features.c
- feature_extraction/code/getNearest_mex.c
- feature_extraction/code/resize.cc
prediction_demo.m needs to know where the VOC 2008 data is. This data is best downloaded from http://pascallin.ecs.soton.ac.uk/challenges/VOC/voc2008/VOCtrainval_14-Jul-2008.tar.