search

Cartucho / mAP

mean Average Precision - This code evaluates the performance of your neural net for object recognition.
Apache License 2.0
2.93k stars 910 forks source link

About the P-R curve #93

Open Qingfeng0623 opened 5 years ago

Qingfeng0623 commented 5 years ago

Hello,I have one problem about the P-R curve. image The curve in the above picture is normal,but my picture didn't like this. image Why is there no content in the red circle?

Cartucho commented 4 years ago

They are assuming that the recall goes to 1 if the precision goes to 0. Meaning that if your object detector was bad (precision 0), it would detect an infinite amount of objects which would bring the recall value to 1.

Cartucho commented 4 years ago

Do you think we should draw it anyway? Even if it does not relate to your real data?

Qingfeng0623 commented 4 years ago

Do you think we should draw it anyway? Even if it does not relate to your real data?

Then why are they different?I mean the the x-coordinate of P-R curves are different according to my classess,one is up to about 0.78,another is up to about 0.64,even 0.5*. image image

Qingfeng0623 commented 4 years ago

Is it related the amount of some class?The larger the number, the larger the abscissa can be？

Cartucho commented 4 years ago

The two classes do not need to have the same recall score. Your object detector may be good at detecting humans but not at detecting bananas for example.

creabob commented 4 years ago

Do you think we should draw it anyway? Even if it does not relate to your real data?

if I want to draw this parts, how to draw it?

Yonio-Lee commented 4 years ago

Do you think we should draw it anyway? Even if it does not relate to your real data?

I don‘t get it.With out drawing it,how can we caculate the area under the P-R curve？The curve is not closed any way.Sorry，my English is poor. I don't know if you get me.

praneet195 commented 3 years ago

Can we extend your code to somehow get closed PR curves? Maybe compute recall at other values as well? Or extrapolate?