Created by: FlyingQianMM

PR types

Others

PR changes

APIs

Describe

Code example of sigmoid_focal_loss op before is too simple, so rewrite it to show:

• get the fg_num tensor needed by sigmoid_focal_loss during training phase
• loss output from sigmoid_focal_loss should be connected to a reduce_sum layer
• prediction should be connected to a sigmoid layer during testing phase

Before

After