>I asked Stefan if HOG was the kind of stuff he planned to implement in
scikit-image and told me no.

Where is the right place for it then?  OpenCV is an option.  It has a version, but its incredibly complicated and in my view obfuscated code.   I'm also growing less certain about the quality of that codebase as it increases in size.


On 31 August 2011 10:51, Olivier Grisel <olivier.grisel@ensta.org> wrote:
2011/8/31 Gael Varoquaux <gael.varoquaux@normalesup.org>:
> On Wed, Aug 31, 2011 at 11:40:30AM +0200, Olivier Grisel wrote:
>> +1 for integrating simple and mainstream feature extractors (no
>> additional dependency required: just pure numpy, the mathematical
>> formula is a one-liner on Wikipedia and there are plenty of already
>> available implementations to check that our own is correct by
>> comparing to them).
>
> On the other hand, where does the scikit-learn stop and the scikit-image
> start?

I asked Stefan if HOG was the kind of stuff he planned to implement in
scikit-image and told me no. But if brian wants to contribute it there
I think Stefan is an open-minded person :) We could then add examples
in scikit-learn that rely upon scikit-image for feature extraction.

However if it's just 100 lines of numpy with comments, it would be
good to have self contained examples in scikit-learn as well :)