How to merge new LBP training to existing one?

asked 2017-03-13 07:51:45 -0500

Sylvio gravatar image

updated 2017-03-13 07:55:32 -0500


I have done several implementations of face recognition and currently working with LBP training models. I can create LBP face training models and call the FaceRecognizer update() to append new faces traning.

Now, considering that I have 2 raspberry boards running the face recognizer service, I would need to replicate the XML training file into both boards evertyime I add a new face.

Let's say I have a server that holds the faces (10 to 20 images per person) and the training XML itself. Now, I see 3 ways to let both raspberry boards updated:

  1. The server service runs the update() to add up new faces to the existing XML training and then the file is downloaded to both boards.

    a) PROS: Only 1 file as training model all the time.

    b) CONS: File gets bigger and bigger all the time

    c) CONS: Raspberry will download bigger files (take time..)

  2. The server service runs a NEW training on new faces and create a NEW XML file. Then the boards keep track of new XML files and APPEND it to its actual file.

    a) PROS: Server will keep lots of versioned .xml files

    b) PROS: Raspberry will download only the last training (small file)

    c) CONS: HOW to merge/append this new XML file to existing one as I cannot use update()?

  3. Boards download the new available server's faces set and each board update() its own XML LBP training.

    a) PROS: Board will download only a few images < 1MB.

    b) PROS: Boards just need to run update() to update the training file

Does any of you have implemented some kind work like this? Any suggestion?


Regards, Sylvio

edit retag flag offensive close merge delete


run the face reco on a decent, central server, and let your raspi's act as shallow clients for that.

(imho, your problem is artificial)

berak gravatar imageberak ( 2017-03-13 07:55:23 -0500 )edit

You mean that I should upload new faces to the server to recognize it? I do want to run it offline for fast recognition. Consider that the number of raspberrys can increase to 100 units..

Sylvio gravatar imageSylvio ( 2017-03-13 08:06:11 -0500 )edit

you want to deploy, mantain and synchonize models on 100 seperate pi's ?

... no dear.

berak gravatar imageberak ( 2017-03-13 08:14:01 -0500 )edit