VLFeat4Py: A wrapper library for VLFeat

This is a Python wrapper library for the open source computer vision library VLFeat using the Armadillo C++ Matrix Library.

VLFeat4Py targets for both the ease of usage in C++ and Python, samples can be found under test/ directory.

VLFeat4Py is not targeted to be a complete wrapper for VLFeat, it is best used with the combination of sklearn and skimage.

Current wrapped functions are:

• vl_sift
• vl_dsift
• vl_imsmooth (obsoleted, use skimage.filters.gaussian_filter instead)
• vl_phow
• vl_kmeans
• vl_homkermap
• vl_gmm
• vl_fisher

INSTALL

1. Install Armadillo
2. Install VLFeat (Properly setup include files and libvl.so)
3. Run python setup.py build_ext --inplace
4. Copy _vlfeat.so and vl_phow.py under folder vlfeat/, and move to PYTHONPATH directory

Usage

import numpy as np
import time
from scipy.misc import lena
from vlfeat import vl_phow

I = np.asarray(lena(), dtype=np.float32) / 255.0

__TIC = time.time()
# frames, descrs = vl_phow(I, sizes=4, verbose=1) 
frames, descrs = vl_phow(I, sizes=4, color='opponent', verbose=1) 
print time.time() - __TIC

print frames.shape
print descrs.shape