AI and DS Skills

 View Only
  • 1.  Any deep learning algorithm for unique face clustering ?

    Posted Thu August 20, 2020 03:54 PM

    Hi guys, I am working on a task to detect new unique face or person given any image. 

    I  came across different models that could do this clustering. Example is 

    • DBSCAN

    However i am interested in a deep learning approach. 
    Any deep learning algorithm for unique face clustering ?

    Thanks in advance. 



    ------------------------------
    Dami Sparks Omifare
    Data Scientist
    ------------------------------

    #AIandDSSkills
    #DataandAILearning


  • 2.  RE: Any deep learning algorithm for unique face clustering ?

    Posted Fri August 21, 2020 09:57 AM
    Probably try opencv (open source computer vision), I did a short course on it and I think it is a really great tool for face recognition /clustering. 

    ------------------------------
    Victoria Amadin
    ------------------------------



  • 3.  RE: Any deep learning algorithm for unique face clustering ?
    Best Answer

    Posted Wed August 26, 2020 05:14 AM

    @Victoria Amadin Thanks for your response.

    Apparently, I ​am already using that for the image manipulation.

    Thanks for the suggestions.

    After some research I found the solution.

    I used DBSCAN for clustering faces.

    I used Pyfacy in combination with dlib for the unique face detection.

    Hopefully this helps others.

    Cheers.



    ------------------------------
    Dami Sparks Omifare
    Data Scientist
    ------------------------------



  • 4.  RE: Any deep learning algorithm for unique face clustering ?

    Posted Thu August 27, 2020 09:13 AM

    In addition to this, Here was my process flow : 

    • First detect faces (using neural networks) and opencv
    • Encode feature extracted.
    • cluster the features using DBSCAN
    • implement auto increment for no of persons found on an image.


    ------------------------------
    Dami Sparks Omifare
    Data Scientist
    ------------------------------