-
-
Notifications
You must be signed in to change notification settings - Fork 16.2k
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Precision-Recall Curve feature update (#1206)
* Precision-Recall Curve feature update * sentinel value update
- Loading branch information
1 parent
7fe4a6b
commit ed85038
Showing
1 changed file
with
10 additions
and
8 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
ed85038
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Why change the way to calculate the AP? I find most of the codes calculate the ap using the old way
np.concatenate(([0.], recall, [min(recall[-1] + 1E-3, 1.)]))
, and the new way can make the AP higher, but I don't think it is more reasonable, because there should be a 0 in recall and precision in actual.ed85038
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@klyjm this update helps close the gap between the repo mAP and the pycocotools mAP, so now users will receive more accurate mAP results. See #1206 (comment)
There still exists a gap in mAP results that we are not able to account for however, so the displayed YOLOv5 mAP is still under-representing the true mAP as calculated by pycocotools.
ed85038
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
OK, I know why. Thank you for your explanation!