How to check mAP in training?

@jwchoi384 Hi,
It's stated that we should check mAP in the training in several issues (#19 (comment)) which asks why average loss is not converging during training. But, I didn't understand how to do it. Is there a hyperparameter in the config file? If not, how does it work? Does not the network try to converge according to loss function?

@canyilmaz90
The average loss is converging during training.
Printed loss value in our code is not actual loss value, because of the gradient of uncertainty.
I didn't modify the calculation of loss value.
If you want to print the actual loss value, I think you should modify it.

Gaussian_YOLOv3/src/gaussian_yolo_layer.c

Line 279 in 02b91a5

*(l.cost) = pow(mag_array(l.delta, l.outputs * l.batch), 2);

I just calculated the mAP with trained weights after training.
There is no hyper parameter related to this in cfg.

@jwchoi384 thanks for the explanation.