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
PaddlePaddle
Paddle
提交
5b9450ae
P
Paddle
项目概览
PaddlePaddle
/
Paddle
大约 1 年 前同步成功
通知
2299
Star
20931
Fork
5422
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
1423
列表
看板
标记
里程碑
合并请求
543
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
1,423
Issue
1,423
列表
看板
标记
里程碑
合并请求
543
合并请求
543
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
提交
5b9450ae
编写于
1月 22, 2017
作者:
H
hedaoyuan
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
follow comments
上级
4426573a
变更
1
隐藏空白更改
内联
并排
Showing
1 changed file
with
24 addition
and
14 deletion
+24
-14
paddle/function/CrossMapNormalOp.cpp
paddle/function/CrossMapNormalOp.cpp
+24
-14
未找到文件。
paddle/function/CrossMapNormalOp.cpp
浏览文件 @
5b9450ae
...
...
@@ -119,14 +119,14 @@ void CrossMapNormalGrad<DEVICE_TYPE_CPU>(real* inputsGrad,
*
* The original formula is:
*
*
Input(
x, y)
* Output(
x, y) =
---------------------------------------------
* -- upper
*
(k + alpha * > (Input(
x, y))^2) ^ (beta)
*
--
lower
*
Input(i,
x, y)
* Output(
i, x, y) = -
---------------------------------------------
*
-- upper
*
(k + alpha * > (Input(j,
x, y))^2) ^ (beta)
*
-- j =
lower
*
* upper is `min(
F, f-[N/2] + N
)`
* lower if `max(0,
f-[N/2]
)`
* upper is `min(
C, c + N/2
)`
* lower if `max(0,
c - N/2
)`
*
* Function implementation:
*
...
...
@@ -134,8 +134,12 @@ void CrossMapNormalGrad<DEVICE_TYPE_CPU>(real* inputsGrad,
* And the meaning of each dimension(0-3) is respectively batch size,
* feature maps, rows and columns.
*
* Input and Output in the above formula is for each map of one image, and
* Input(x, y), Output(x, y) represents an element in an image.
* Input and Output in the above formula is for each map(i) of one image, and
* Input(i, x, y), Output(i, x, y) represents an element in an image.
*
* C is the number of feature maps of one image, and N is a hyper-parameters
* is configured when Function is initialized. The sum in the denominator
* is the sum of the same position in the neighboring maps.
*
* In the implementation of Function, k is equal to 1,
* so Function has no argument for k.
...
...
@@ -199,20 +203,26 @@ private:
/**
* \brief Backward calculation for normalization with across maps.
*
* Function implementation:
*
* The implementation of this Function is derived from the
* CrossMapNormalFunc implementation.
*
* InputGrad = OutputGrad * denoms ^ (-beta)
*
/
* +
|
(OutputGrad * OutputValue * (-2 * alpha * beta) / denoms) * InputValue
*
/
*
-- upper
* +
>
(OutputGrad * OutputValue * (-2 * alpha * beta) / denoms) * InputValue
*
-- lower
*
* Argument in the Function:
* The data of inputs/outputs format is the same as the forward interface
* and is NCHW.
*
* The upper and lower is the same as forward. The logic of the sum
* is also the same as forward.
*
* Function Arguments:
*
* \param size_ represent N
* \param scale_ represent alpha
/ N
* \param scale_ represent alpha
* \param pow_ represent beta
* \param inputs[0] represent InputValue, inputs[0] of CrossMapNormalFunc
* \param inputs[1] represent OutputValue, outputs[0] of CrossMapNormalFunc
...
...
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