Created by: prakharg24

I am looking to train MobileNet-SSD as present here, https://github.com/PaddlePaddle/models/tree/develop/PaddleCV/ssd

However, instead of using the SSD Loss (which samples negative gt anchor boxes to maintain 3:1 ratio), I want to use Focal Loss (as described in RetinaNet).

Is there an inbuilt loss function for such focal loss calculations? If not, what would be the easiest way to implement a custom loss function like that, using the existing paddlepaddle functions?

Any direction or help is appreciated.

Thank You