How to calculate the mAP?

pqhieu / jsis3d

[CVPR'19] JSIS3D: Joint Semantic-Instance Segmentation of 3D Point Clouds

https://pqhieu.com/research/cvpr19/

MIT License

175 stars 36 forks source link

How to calculate the mAP? #28

Closed tpzou closed 2 years ago

tpzou commented 3 years ago

I find there is no score and only one pair of RP, so how to calculate the mAP?

pqhieu commented 3 years ago

We use the same AP as defined in the COCO dataset, which means we change the IoU threshold from 0.5 to 0.95 with 0.05 step size and calculate precision at these intervals.mAP is the average of all the classes.

tpzou commented 3 years ago

Thanks for your reply. But I stll don't understand how to calculate a single AP. I don't know if I've made my point clear. I mean that if I set the IoU as 0.5, as far as I am concerned, the AP is to calculate the area under the PR polyline. But in this work, there is no score, and it only has one pair of PR. So how to calculate to area?

pqhieu commented 3 years ago

You calculate the precision at several IoU threshold and take the average. I suggest that you take a look at this article https://jonathan-hui.medium.com/map-mean-average-precision-for-object-detection-45c121a31173

haffybra commented 3 years ago

@pqhieu I had the same doubts on how to calculate the AUC. Check this repository, it might help https://github.com/rafaelpadilla/review_object_detection_metrics