regularizer is not included in total loss

[hgjung97]

The current train loop calculates the total loss as below:
total_loss = giou_loss + conf_loss + prob_loss

This implementation doesn't include regularizer's loss, such as kernel_regularizer of Conv2D.
In my opinion, the above sentence should be modified as below to reflect regularizers.
total_loss = giou_loss + conf_loss + prob_loss + tf.add_n(yolo.losses)

Reference: tensorflow/tensorflow#21587

[pythonlessons]

Hmmm, this looks interesting, need to look at it

[hgjung97]

When I applied such a modification, I got different results.

[pythonlessons]

When I applied such a modification, I got different results.

What were the results, were they better, have you compared the mAP?