Why the mAP it is calculated for each classes?

[pretiand]

I understood that the mAP is an unique value calculated over all the classes. I have 9 classes and val.py outputs 9 different mAP: what does it means?

[glenn-jocher]

@pretiand AP per class and mAP for all classes is displayed. Be happy, this is a feature requested by users.

[Abhy-007]

While training I got the mAP as 0.38, but while running the best.pt on the val.py I got the mAP as 0.46. Is this normal? Also how can I get the training accuracy? I believe the mAP that we are getting while training and running val.py is the validation accuracy, ryt? Please throw some light into this

[glenn-jocher]

@Abhy-007 val.py defaults to evaluating a dataset val split. You can specify train or test splits with python val.py --task train or --task test