Only found just one loss about pytorch-yolov3 HOT 2 CLOSED

Hi,

The model consists of three YOLO layers. In each of these the loss of the predictions with respect to the target annotations are calculated and returned as output. The model is then updated with respect to the loss computed in these three layers.
In DarkNet.forward under if module_def['type'] == 'yolo' you can see that the output (loss in the case when target annotations are fed as input to the forward function) of each yolo layer is added to the accumulated output (loss) of the forward pass, and then returned.

Hope this clarifies how the three losses are computed.

Cheers,
Erik

from pytorch-yolov3.

@eriklindernoren
Roger that. Thanks for your wonderful work 👍

from pytorch-yolov3.