未验证 提交 67f5eaff 编写于 作者: W whs 提交者: GitHub

Add dice loss (#10717)

* Add dice loss.

* Fix comments.

* Remove unused code.
上级 62559ace
...@@ -828,4 +828,10 @@ topk ...@@ -828,4 +828,10 @@ topk
.. autofunction:: paddle.fluid.layers.topk .. autofunction:: paddle.fluid.layers.topk
:noindex: :noindex:
dice_loss
----
.. autofunction:: paddle.fluid.layers.dice_loss
:noindex:
...@@ -80,6 +80,7 @@ __all__ = [ ...@@ -80,6 +80,7 @@ __all__ = [
'pad', 'pad',
'label_smooth', 'label_smooth',
'roi_pool', 'roi_pool',
'dice_loss',
] ]
...@@ -3816,3 +3817,43 @@ def roi_pool(input, rois, pooled_height=1, pooled_width=1, spatial_scale=1.0): ...@@ -3816,3 +3817,43 @@ def roi_pool(input, rois, pooled_height=1, pooled_width=1, spatial_scale=1.0):
"spatial_scale": spatial_scale "spatial_scale": spatial_scale
}) })
return pool_out return pool_out
def dice_loss(input, label, epsilon=0.00001):
"""
**Dice loss Layer**
Dice loss for comparing the similarity of two batch of data,
usually is used for binary image segmentation i.e. labels are binary.
The dice loss can be defined as below equation:
.. math::
dice\_loss &= 1 - \\frac{2 * intersection\_area}{total\_area} \\\\
&= \\frac{(total\_area - intersection\_area) - intersection\_area}{total\_area} \\\\
&= \\frac{(union\_area - intersection\_area)}{total\_area}
Args:
input (Variable): The predictions with rank>=2. The first dimension is batch size,
and the last dimension is class number.
label (Variable): The groud truth with the same rank with input. The first dimension
is batch size, and the last dimension is 1.
epsilon (float): The epsilon will be added to the numerator and denominator.
If both input and label are empty, it makes sure dice is 1.
Default: 0.00001
Returns:
dice_loss (Variable): The dice loss with shape [1].
Examples:
predictions = fluid.layers.softmax(x)
loss = fluid.layers.dice_loss(input=predictions, label=label, 2)
"""
label = one_hot(label, depth=input.shape[-1])
reduce_dim = range(1, len(input.shape))
inse = reduce_sum(input * label, dim=reduce_dim)
dice_denominator = reduce_sum(
input, dim=reduce_dim) + reduce_sum(
label, dim=reduce_dim)
dice_score = 1 - inse * 2 / (dice_denominator + epsilon)
return reduce_mean(dice_score)
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册