Dense SIFT implementation in Python based on/copied from Svetlana Lazebnik
and Yangqing Jia's code
- http://www.cs.illinois.edu/homes/slazebni/research/SpatialPyramid.zip
- https://github.com/Yangqing/dsift-python

Not as fast as MATLAB implementation but should be faster than other
Python implementation due to use of convolutions.

If find a way to make feature computation faster or add computation of
features w.r.t. a start orientation please let me know :).