C24 | Calculating Average Precision from Precision-Recall curves | Object Detection | EvODN

Поділитися
Вставка
  • Опубліковано 15 лис 2024

КОМЕНТАРІ • 15

  • @Cogneethi
    @Cogneethi  5 років тому

    Full course: ua-cam.com/play/PL1GQaVhO4f_jLxOokW7CS5kY_J1t1T17S.html
    If you found this useful, please:
    a. Subscribe to this channel b. Share on twitter.com and tag @cogneethi c. Share on other Social and Messaging apps

  • @digitaldrive1192
    @digitaldrive1192 3 роки тому +5

    So far the best break down of this topic. Nobody else seems to explain it in better detail than this.

  • @deeps-n5y
    @deeps-n5y 2 роки тому +1

    One of the best and earliest resources for object detection!

  • @saurabhband6948
    @saurabhband6948 3 роки тому

    Best explanation of this topic on the internet. It was really very helpful.

    • @Cogneethi
      @Cogneethi  3 роки тому

      Thanks Saurabh. All credit goes to this: github.com/rafaelpadilla/Object-Detection-Metrics

  • @swarnabandi7670
    @swarnabandi7670 4 роки тому +1

    Very good explanation .Thank you

  • @keenchkaat1543
    @keenchkaat1543 3 роки тому +1

    Why are you considering highest value of precision rather than lower? any particular reason?

  • @adarshnarayan
    @adarshnarayan 3 роки тому

    This is one of the best courses I have come across, thank you for that. I had a doubt. So while we are drawing the horizontal line, do we stop at the next vertical line only or the intersection of our horizontal line with the line plot?

  • @coskuoksuz2446
    @coskuoksuz2446 4 роки тому +1

    This is very good video that really helps me. In many tutorial, only the average precision is explained. However in some areas such as medical field, recall value is very important. How can we estimate average recall value? Is it possible for you to give an example on this metric?

    • @Cogneethi
      @Cogneethi  4 роки тому +1

      Thanks for the question, I didnt know about Average Recall. I just googled it.
      As I have explained in one of the previous videos: ua-cam.com/video/ji48Lz6amMc/v-deo.html
      You might get different values of precision and recall by varying the thresholds.
      The threshold itself depends on the use case.
      As a simple example, for cancer, it might be the tumor size.
      For object detection, it might be the IoU.
      So, basically, you vary the thresholds from say 0 to 1 or 0.5 to 1 etc and get different values of recall.
      All these recall values are averaged to get 'Average Recall'.
      If you have multiple classes to test, you take the mean of 'Average Recalls' of all the classes to get 'Mean Average Recall'.
      In case of Object Detection, you may have 10 classes of animals like Cats, Dogs etc.
      You first calculate the AR for Cats, then AR for Dogs and so on.
      Then take the mean of them to get Mean Average Recall.
      Let me know if I have to elaborate.

  • @mattjoe182
    @mattjoe182 4 роки тому

    Awesome content, thanks for helping me understand

  • @ankursahu7704
    @ankursahu7704 3 роки тому +1

    Thank you @Cogneethi

  • @tiasm919
    @tiasm919 4 роки тому

    Hi, i'd like to ask, if the highest confidence (the first row) is a False Positive, it means the first Precision and Recall should be both 0 right ? Then the AP graph might look like from 0 then go way up and slowly going down again. Is it ok ?