Can mAP be calculated per class for multiple classes on COCO or custom dataset?

[Eldad27]

Inference result (mAP) for COCO dataset is typically reported for all classes. I stand to be corrected if I'm wrong. Can you please guide me on the procedure to produce the mAP per class on either COCO or custom dataset for Mask RCNN?

I successfully trained Mask RCNN (R-50 FPN) on a custom dataset of 2 classes. During the evaluation, inference results display one AP for detection (Bbox) results and another for segmentation results (Segm), which I assume represents mAP for my two classes combined. How can I display mAP for each specific class (class specific mAP)? Is this possible? Please help.

[jpainam]

@sotte any comment on this issue? I'm also confronted with the problem. Is there any command for producing mAP for each class?

[SkeletonOne]

#524