sigmoid_focal_loss¶

dragon.losses.sigmoid_focal_loss(
  inputs,
  axis=- 1,
  alpha=0.25,
  gamma=2.0,
  start_index=0,
  reduction='valid',
  **kwargs
)[source]¶

Compute the focal loss of sigmoid cross entropy. [Lin et.al, 2017].

Examples:

x = dragon.constant([[0.5, 0.5], [0.3, 0.7]])
y = dragon.constant([1, 0])
print(dragon.losses.sigmoid_focal_loss([x, y]))  # 0.3472295

Parameters:

• inputs (Sequence[dragon.Tensor]) – The tensor input and target.
• axis (int, optional, default=-1) – The reduction axis, can be negative.
• alpha (float, optional, default=0.25) – The scale factor to the positive classes.
• gamma (float, optional, default=2.0) – The exponential decay factor on the easy examples.
• start_index (int, optional, default=0) – The start value of target.
• reduction ({'none', 'sum', 'mean', 'valid'}, optional) – The reduction method.

Returns:

dragon.Tensor – The output tensor.