How to calculate mAP when training on DOTA dataset? Should I write the code myself?

After training the model, how should I calculate the mAP since there is no standard to evaluate besides cls_loss in tensorboard.
Hoping to your reply : )
Best regards~