Menu

imglab question

Help
2016-06-01
2016-06-02
  • Jungsae Yim

    Jungsae Yim - 2016-06-01

    hi, Davis !
    i want to make my own classifier. for example, like upper body.
    so, i manually draw square and circle.
    Is there any more convenient way?
    how many sample should i do for good detecting?
    thanks.

     
    • Davis

      Davis - 2016-06-02

      That's the only way there is to do it. If there was an automatic way then
      you wouldn't need the training data in the first place :)

      You keep making the dataset bigger until you either give up or notice that
      the performance of the learned model stops increasing.

       
      • Jungsae Yim

        Jungsae Yim - 2016-06-02

        Thank you, Davis !!

         

        Last edit: Jungsae Yim 2016-06-02
  • Kenny Low

    Kenny Low - 2016-06-02

    Hope your day's good Davis, I like to learn from you in imglab, should we also include negative samples? How can we do this with imglab? :)

    Much appreciation to all here :)

     

    Last edit: Kenny Low 2016-06-02
    • Jungsae Yim

      Jungsae Yim - 2016-06-02

      i am newbie,
      i think you are already know how to use imglab.
      i think you have to train simillar target (like face)
      i think you have to keep same aspect ratio in imglab.

       
  • Pawan Yadav

    Pawan Yadav - 2017-03-25

    @Kenny Anything you don't label is assumed to be a negative sample automatically.

     

Log in to post a comment.

MongoDB Logo MongoDB