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
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?
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?
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.
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 ?
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
So far the best break down of this topic. Nobody else seems to explain it in better detail than this.
Thank you!
One of the best and earliest resources for object detection!
Best explanation of this topic on the internet. It was really very helpful.
Thanks Saurabh. All credit goes to this: github.com/rafaelpadilla/Object-Detection-Metrics
Very good explanation .Thank you
Why are you considering highest value of precision rather than lower? any particular reason?
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?
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?
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.
Awesome content, thanks for helping me understand
Thank you!
Thank you @Cogneethi
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 ?