Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
wmsofts
Paddle
提交
83f2e2c9
P
Paddle
项目概览
wmsofts
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
0
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
0
列表
看板
标记
里程碑
合并请求
0
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
Paddle
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
0
Issue
0
列表
看板
标记
里程碑
合并请求
0
合并请求
0
Pages
分析
分析
仓库分析
DevOps
Wiki
0
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
提交
Issue看板
体验新版 GitCode,发现更多精彩内容 >>
提交
83f2e2c9
编写于
12月 28, 2018
作者:
S
shippingwang
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
rewrite the comments, test=develop
上级
9322d340
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
44 addition
and
19 deletion
+44
-19
paddle/fluid/operators/shuffle_channel_op.cc
paddle/fluid/operators/shuffle_channel_op.cc
+3
-8
python/paddle/fluid/layers/nn.py
python/paddle/fluid/layers/nn.py
+41
-11
未找到文件。
paddle/fluid/operators/shuffle_channel_op.cc
浏览文件 @
83f2e2c9
...
...
@@ -55,17 +55,12 @@ class ShuffleChannelOpMaker : public framework::OpProtoAndCheckerMaker {
AddComment
(
R"DOC(
Shuffle Channel operator
This operator obtains the group convolutional layer with channels shuffled.
Firstly, divide the input channels in each group into several subgroups,
then, feed each group in the next layer with different subgroups.
According to the paper, "Suppose a convolution layer with G groups
whose output has (G * N) channels, first reshape the output channel dimension into(G,N),
transposing and then flattening it back as the input of next layer. "
This opearator shuffles the channels of input x.
It divide the input channels in each group into several subgroups,
and obtain a new order by selecting element from every subgroup one by one.
Shuffle channel operation makes it possible to build more powerful structures
with multiple group convolutional layers.
please get more information from the following paper:
https://arxiv.org/pdf/1707.01083.pdf
)DOC"
);
...
...
python/paddle/fluid/layers/nn.py
浏览文件 @
83f2e2c9
...
...
@@ -9338,27 +9338,57 @@ def get_tensor_from_selected_rows(x, name=None):
def
shuffle_channel
(
x
,
group
,
name
=
None
):
"""
**Shuffle Channel Operator**
This operator obtains the group convolutional layer with channels shuffled.
First, divide the input channels in each group into several subgroups,
then, feed each group in the next layer with different subgroups.
Channel shuffling operation makes it possible to build more powerful structures
with multiple group convolutional layers.
This operator shuffles the channels of input x.
It divide the input channels in each group into :attr:`group` subgroups,
and obtain a new order by selecting element from every subgroup one by one.
Please refer to the paper
https://arxiv.org/pdf/1707.01083.pdf
.. code-block:: text
Given a 4-D tensor input with the shape (N, C, H, W):
input.shape = (1, 4, 2, 2)
input.data =[[[[0.1, 0.2],
[0.2, 0.3]],
[[0.3, 0.4],
[0.4, 0.5]],
[[0.5, 0.6],
[0.6, 0.7]],
[[0.7, 0.8],
[0.8, 0.9]]]]
Given group: 2
then we get a 4-D tensor out whth the same shape of input:
out.shape = (1, 4, 2, 2)
out.data = [[[[0.1, 0.2],
[0.2, 0.3]],
[[0.5, 0.6],
[0.6, 0.7]],
[[0.3, 0.4],
[0.4, 0.5]],
[[0.7, 0.8],
[0.8, 0.9]]]]
Args:
x(Variable): The input tensor variable.
group(
Integer): The num of group
.
x(Variable): The input tensor variable.
It should be a 4-D tensor with shape [N, C, H, W]
group(
int): Indicating the conuts of subgroups, It should divide the number of channels
.
Returns:
Variable: channels shuffled tensor variable.
out(Variable): the channels shuffling result is a tensor variable with the
same shape and same type as the input.
Raises:
ValueError: If group is not an int type variable.
Examples:
.. code-block:: python
out = fluid.layers.shuffle_channel(x=group_conv,group=4)
input = fluid.layers.data(name='input', shape=[1,4,2,2], dtype='float32')
out = fluid.layers.shuffle_channel(x=input, group=2)
"""
helper
=
LayerHelper
(
"shuffle_channel"
,
**
locals
())
...
...
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录