mAP not the official COCO mAP measurement?

[starlipernl]

According to COCO dataset, the mAP metric should be calculated as: AP at IoU=.50:.05:.95 (primary challenge metric) The mAP calculated in this implementation is actually mAP @ IOU=0.5. Wouldn't this cause the model to get higher mAP than the true value?

[ujsyehao]

you just compare with [email protected]