Ask Your Question
0

is there a way to use a GUID as a label in FaceRecognizer?

asked 2013-07-13 15:36:26 -0600

jdavidthomson gravatar image

Hi,

I've been playing with the FaceRecognizer stuff in opencv 2.4.5 and was wondering if there is a way to use a GUID like 25892e17-80f6-415f-9c65-7395632f0223 as a label in an lbph model?

I want to build a very scalable solution and don't want to be limited to a label based on an Integer. Please correct me if I'm wrong, but to my understanding this would limit the number of subjects to something like 32767: (http://www.cplusplus.com/reference/climits/). I was theorizing to build a system that would hold something like 7 billion unique subjects :) What a yml file that would be :P

Ideas?

edit retag flag offensive close merge delete

Comments

Using a little program I was able to see that the max value for an Integer on my system is: 2147483647; approximately 2 billion...right ? lol

jdavidthomson gravatar imagejdavidthomson ( 2013-07-14 02:03:46 -0600 )edit

1 answer

Sort by ยป oldest newest most voted
1

answered 2013-07-14 01:43:59 -0600

berak gravatar image

updated 2013-07-14 01:45:19 -0600

for 7 billions of images ( and probably a billion of labels ), you might want to run several (lbp-like) facerecognizers in parallel, and do a nn-search again on their results.

sure, you can hack it, to have a vector<GUID> instead of a vector<int> as your labels, but a (local) lookup-table int_label <-> GUID will work as well without modding it.

edit flag offensive delete link more

Comments

that's what a buddy and i were talking about earlier tonight. Thank's for the input :) running multiple in parallel. Might take a stab at updating the native OpenCV datatypes to a long int and see if that works.

jdavidthomson gravatar imagejdavidthomson ( 2013-07-14 02:16:57 -0600 )edit

the lbp-recognition is doing a linear search over all lbp feature vecs. still, you don't want more than ~100000 images per instance of facerecognizer ( and that will definitely fit into an int .. )

berak gravatar imageberak ( 2013-07-14 03:17:28 -0600 )edit

Thanks berak! I was thinking of keeping it much lower than that and running multiple instances of facerecognizer as you noted. and running a sort of search tree amongst all instances of facerecognizer then dynamically building a subsearch matrix out of the results from the initial search. What do you think?

jdavidthomson gravatar imagejdavidthomson ( 2013-07-15 11:34:15 -0600 )edit

i'll be curious about your findings ;)

leave no stone unturned.

berak gravatar imageberak ( 2013-07-15 11:41:43 -0600 )edit

Question Tools

1 follower

Stats

Asked: 2013-07-13 15:36:26 -0600

Seen: 196 times

Last updated: Jul 14 '13