On Sun, Mar 21, 2010 at 9:51 PM, Roberto -MadBob- Guido
<address@hidden> wrote:
2010/3/21 Laurent Aguerreche <address@hidden>:
> Does somebody plan to add such features in Tracker (for pictures first
> but they may be extended to videos for automatic identification of
> chapters/scenes)?
>
It depends on the availability of existing open licensed code to
provide face/things recognization, since it doesn't seems a function
implementable in short time from scratch.
Any link?
I'd say that that functionality should be out of tracker. Applications should do all that "guessing" and store in tracker only facts. I think it is extremely difficult to do it (right) at extraction time. Anyway, if somebody decides to try that patch and implement an extractor for face recognition... it is more than welcome.
I'd prefer a plugin in EOG (or any other image viewer) that allows the user to easily mark "My contact X is in this picture": more reliable information, and probably easier to implement.
> Would Tracker be able to store appropriate data to
> find someone/something in a picture when this picture has been proceeded
> to extract its content?
>
This is an argument actually discussed in the Nepomuk ontologies
bugtracker: http://sourceforge.net/apps/trac/oscaf/ticket/66
Right now it is possible with the generic nie:related property (IIRC). The idea is to add more specific subproperties to refine the meaning, but we need the concrete use cases (contact is picture is a very good one).
Thanks for the comments, going back to the original topic, to get some extra information from the DCIM file system sounds a good idea (it is a standard, so we can kinda rely on it). I'd like to read the document posted before (permission already requested!)
Regards,
Ivan