-
Notifications
You must be signed in to change notification settings - Fork 1.8k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Interpolation error on computing average precision #1103
Comments
Hey @illian01 👋 Thank you for reporting the issue! If anyone in the community stumbles upon this - would you like to help us out? 🙂 |
Hey I was just playing around with this and had a couple questions:
|
For reference here's what I was doing: https://colab.research.google.com/drive/1vT_9Q2RwQnCFqlcz-ptpmeZaQomGY-WB?usp=sharing |
Hi @illian01, @Griffin-Sullivan, Having had a few conversations and looking around, we've decided to keep the current method. Prepending the 0 and 1 for precision and recall seems to be the standard approach when implementing the COCO 101 mAP algorithm, and we wouldn't like to gravitate away from that. |
Search before asking
Bug
I have precision and recall curve as below.
So, I can plot PR curve as below.
However, according to
compute_average_precision
method, always pad0.0
and1.0
to precision and recall vectors.supervision/supervision/metrics/detection.py
Lines 727 to 749 in 4729e20
I think these extensions are dangerous and makes pr curve inaccurate. Below is curve of
interpolated_precision
andinterpolated_recall_levels
.The bent part of right bottom in the plot seems unnatural and effects on the accuracy of the average precision calculation.
Environment
No response
Minimal Reproducible Example
Additional
No response
Are you willing to submit a PR?
The text was updated successfully, but these errors were encountered: