Re: downloading files for face detection / red-eye removal

Hi Steven,

The kind of binary data files are IA model computed and coming from
3rd party projects (mostly taken on github repositories).

They are computed by teaching neural networks with real data
introduced in mathematical definitions. The process takes ages to do.
On github, this kind of model is pre-computed and ready to use with a
deep learning engine. The format of binary files are standardized but
not the same especially as there are many formats to store this kind
of data, HDF5 is the most common for ex (i use it in my office).
Another one is ONNX :

https://en.wikipedia.org/wiki/Open_Neural_Network_Exchange

HDF5 is Berkeley based licensing
ONNX is MIT Licensing.

For all models files hosted in the repository, all README files are
now update with the places where files have been taken and the
licensing of data.

Best

Gilles Caulier



Le dim. 20 août 2023 à 05:18, Steven Robbins <[email protected]> a écrit :
>
> On Saturday, May 6, 2023 7:12:12 A.M. CDT Maik Qualmann wrote:
> > With this commit the face engine data can now also be stored centrally in
> > the /usr/share/digikam/ directory.
>
> Thank you Maik!  I'm now looking at creating a package of these files for
> Debian.
>
> I located the needed files on the KDE site https://files.kde.org/digikam/
> facesengine/
>
> I wasn't able to locate a copyright or license statement for these files.  Is
> there one?
>
> These files are mostly binary data blobs.  Is there a "source" for them?  Or a
> description of how they were generated?
>
> Thanks,
> -Steve