❔Question

Hi,

I've seen you have update the repo (#1474) and add the Confusion Matrix. When I run it with my own dataset I get this Confusion Matrix:

My mAP is 0.84 for this dataset, and for example for Airplane class, AP = 0.95.

So I can't understand how is possible that in the Confusion Matrix [row=Airplane, column=Airplane] = 0.26, [row=Airplane, column=FN] = 0 and [row=FP, column=Airplane] = 0.73.

As I can understand this means that just 26% of detections are correct (TP), but it's not possible because AP(Airplane) = 0.95 and when I run detections Airplanes are well detected.

Additional context

I'm getting this weird result not just with this dataset, if not with other 2.