提交 54a85b7b 编写于 作者: D dragonwarrior 提交者: qingqing01

Add lrn layer (#9157)

* add LRN layer for fluid

* add LRN layer for fluid

* add documentation for LRN layer

* add paper reference for LRN layer

* add seperate documentation for LRN layer

* rm lrn.py in doc/fluid/dev/src

* change code style in lrn

* fix style of comments in lrn
上级 3941c2dd
......@@ -74,6 +74,7 @@ __all__ = [
'one_hot',
'autoincreased_step_counter',
'lod_reset',
'lrn',
]
......@@ -3410,3 +3411,73 @@ def lod_reset(x, y=None, target_lod=None):
raise ValueError("y and target_lod should not be both None.")
return out
def lrn(input, n=5, k=1.0, alpha=1e-4, beta=0.75, name=None):
"""
Local Response Normalization Layer. This layer performs a type of
"lateral inhibition" by normalizing over local input regions.
The formula is as follows:
.. math::
Output(i, x, y) = Input(i, x, y) / \left(
k + \alpha \sum\limits^{\min(C, c + n/2)}_{j = \max(0, c - n/2)}
(Input(j, x, y))^2 \right)^{\beta}
In the above equation:
* :math:`n`: The number of channels to sum over.
* :math:`k`: The offset (avoid being divided by 0).
* :math:`alpha`: The scaling parameter.
* :math:`beta`: The exponent parameter.
Refer to `ImageNet Classification with Deep Convolutional Neural Networks
<https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf>`_
Args:
input (Variable): The input tensor of this layer, and the dimension of input tensor must be 4.
n (int, default 5): The number of channels to sum over.
k (float, default 1.0): An offset (usually positive to avoid dividing by 0).
alpha (float, default 1e-4): The scaling parameter.
beta (float, default 0.75): The exponent.
name (str, default None): A name for this operation.
Raises:
ValueError: If rank of the input tensor is not 4.
Returns:
A tensor variable storing the transformation result.
Examples:
.. code-block:: python
data = fluid.layers.data(name="data", shape=[3, 112, 112], dtype="float32")
lrn = fluid.layers.lrn(input=data)
"""
helper = LayerHelper('lrn', **locals())
dtype = helper.input_dtype()
input_shape = input.shape
dims = len(input_shape)
if dims != 4:
raise ValueError(
"dims of input must be 4(not %d), and it's order must be NCHW" %
(dims))
mid_out = helper.create_tmp_variable(dtype=dtype, stop_gradient=True)
lrn_out = helper.create_tmp_variable(dtype)
helper.append_op(
type="lrn",
inputs={"X": input},
outputs={
"Out": lrn_out,
"MidOut": mid_out,
},
attrs={"n": n,
"k": k,
"alpha": alpha,
"beta": beta})
return lrn_out
......@@ -231,6 +231,13 @@ class TestBook(unittest.TestCase):
self.assertIsNotNone(layers.softmax(hid))
print(str(program))
def test_lrn(self):
program = Program()
with program_guard(program):
data = layers.data(name='data', shape=[6, 2, 2], dtype='float32')
self.assertIsNotNone(layers.lrn(data))
print(str(program))
def test_get_places(self):
program = Program()
with program_guard(program):
......
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