2009
09.17
09.17
As my thesis project I’ll implement face detection (and hopefully also recognition) features for F-Spot. This task happens to be one of my favourite iWuds, so I’m glad that for the next months I’ll be working on something that I really wanted for a long time now.
For a couple of days I’ll be making my mind and schedule things a bit, only after that actual work will begin!
Here’s a quick list of tasks to get there (more or less in order to priorities):
- Region tagging: at the moment you can tag photos in F-Spot, but you can’t apply’em to no regions in particular, like facebook does for instance for your friends.
- Tags UI: I don’t like very much how F-Spot shows your tags currently, so I’m thinking of something more iPhoto-ish. Expect somewhat different UI experience as far as tags are concerned (thumbnail grids?, mouse hover thumbnail selection?).
- Face detection: (optionally enabled in options), goal is to detect as much people as possibile in any given photo, possibly one photo at a time (batch detection for your entire archive is something that will come with face recognition instead) and of course an user friendly why to tag those face regions (auto tag name completion?).
- Face recognition: (optionally enabled in options), one could enable face recognition and wait for some (long) time in order to automagically detect & recognize & group people in his entire photo archive, only after that parent “People” tag area would show detected people’s faces waiting for you to tag’em.
- Machine learning: of course you can untag any of your photos at any given time, but wouldn’t it be cool if F-Spot would “learn” about your actions? Here F-Spot should get better and better recognizing people.
Please comment your thoughts!
I’ll keep you posted…
ahaaa!!!! awesome. I heard that picasa just got this feature. (no linux version yet) I fear we’ll lose a lot of users for this feature alone. have you seen apart-dev’s previous work on this for GSoC?
Thanks, hopefully we’ll get this thing working asap and yes I have seen his previous work, unfortunately though it has to be done from scracth. Keep your eyes here please.
Hey great that you’re taking up such an important feature in F-spot as you know Google’s Picasa just came out with facial-recognition, so F-spot certainly needs to have something soon. I’m sure you’ve seen previous attempts at this (for instance: http://apart-dev.blogspot.com/2007/07/svn-rev-21.html).
Iyi sanslar!
Yes, that’s what I thought eventhough Picasa’s feature wasn’t announced yet. Anway I’ll see what can I do. Thanks.
go donmez!!! whooo!
Look forward to this…Picasa have stolen me away from F-Spot now with face detection but I don’t mind doing it again if I could have it in F-Spot. The way Picasa does it is so easy – it might be worth looking to see how Picasa does it to give you some idea of how you can make yours easy to use too.
Although Picasa 3.5 was not released for Linux, I just installed the Picasa 3.0 linux version and then installed 3.5 via wine and moved files into /opt and volia I have intergrated Picasa 3.5.
Good luck with your work and hope you get it done easily and quickly :)
Yeah, that’s the whole idea. Hopefully we will have you back! =)
I would really like to see this implemented in f-spot. The OpenCV libraries do a good job of feature recognition. Perhaps a C# wrapper can be written over those modules? OpenCV also has features training.
Hey, thanks for your comment. The most complete C# implementation for OpenCV that I could found was Emgu CV. Unfortunately I wasn’t able to compile it, nor there are packages for GNU/Linux distribution that I’m using. Now I’m PInvoking OpenCV library calls that I needed instead, one by one, manually.