Re: [Libface-general] Some questions

[Aditya B. <adi...@gm...>]

Hi Richard,

I think I fixed that single face updating problem in the last commit.

- Face recognition is implemented - it used to work fine, but a few (2-3)
commits back there some design changes, it is still in the transitional
phase till we change the design. What we're trying to do is, if a face with
the same ID as an existing one in the DB is trained, then that face is
averaged with the existing one, instead of training it as a new face,

- Fisherfaces is not used at the moment. Once we polish the Eigenfaces
implementation, I plan to concentrate solely on digiKam work for my GSoC.
After that, I'll come back to libface to implement a new, better recognition
algorithm. If you want to implement another algorithm, it's awesome!

- I'm not particularly opposed to using Boost, because I think KDE/digiKam
use boost anyway. But could you mention what are the benefits of boost if
used here instead of the standard STL...

- We're a bit short of manpower at the moment, so any help is welcome :)
Alex can give you commit rights.

- If you get write permissions, all you have to keep in mind is not to
change the interface (of course :) ). And if you implement a new recognition
algorithm, just make sure that it complies with the necessary abstractions.

- OpenCV has a C++ interface, but IMHO it is in some ways C-like too. For
ex, when the OpenCV folk say "C++ interface", it usually translates to
having default arguments in some functions. I wouldn't say it's entirely OO
yet.

Thanks for the help, meanwhile it'd be cool if you could send some diffs
while you wait for Alex to give you rights :)

Regards,
Aditya

On Wed, Jun 2, 2010 at 1:01 AM, Richard Ulrich <ri...@pa...> wrote:

> Hi there,
>
> I want to use libface in a new project
> http://sourceforge.net/projects/receptiongreet/
>
> I understand that libface is not quite finished.
> Thus I would be happy to participate a bit in the development.
> So, I have a few questions:
> - Face Recognition doesn't seem to be implemented in the main interface
> class (libface::LibFace).
> - If I use the Eigenfaces class directly I run into some problems:
>  - Training a person with only one face (image) fails. This seems to be
> by design, but why is it necessary? -> Eigenfaces::doPCA() -> nEigens =
> nTrainFaces - 1;
>  - Finding a face in an already cropped image gives me a face with no
> coordinates, which leads to problems later on in FaceDetect::finalFaces
> in cvCopy near the end.
> - FisherCore.cpp misses some includes, but it doesn't seem to be used
> anyway.
> - would you generally be opposed to using boost in some areas. I find it
> to help a lot.
> - LibFace.cpp line 110 -> filename.c_str() would be better than
> filename.data().
> - If I have fixes, shoudl I send diffs, or would you give me write
> permission on the repository?
> - If I get write permissions, what do I have to look for not to
> interfere with the GSOC project?
> - I saw that opencv has also a C++ interface. I don't know the status of
> that. Is it incomplete? If it is usable enough, it might contribute to
> readability and quality of the libface code.
>
>
> Rgds
> Richard
>
>
>
> ------------------------------------------------------------------------------
>
> _______________________________________________
> Libface-general mailing list
> Lib...@li...
> https://lists.sourceforge.net/lists/listinfo/libface-general
>



-- 
Aditya Bhatt
Blog : http://adityabhatt.wordpress.com
Bitbucket: http://bitbucket.org/aditya_bhatt
Face Recognition Library : http://libface.sourceforge.net