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
eb0c7e5e
P
Paddle
项目概览
PaddlePaddle
/
Paddle
11 个月 前同步成功
通知
2292
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
体验新版 GitCode,发现更多精彩内容 >>
提交
eb0c7e5e
编写于
6月 21, 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
Move the Im2Col code of the CPU version into the Im2ColOp.cpp file.
上级
1a53cba6
变更
3
隐藏空白更改
内联
并排
Showing
3 changed file
with
253 addition
and
116 deletion
+253
-116
paddle/function/Im2ColOp.cpp
paddle/function/Im2ColOp.cpp
+235
-0
paddle/function/Im2ColOpGpu.cu
paddle/function/Im2ColOpGpu.cu
+18
-8
paddle/function/ImageExpandOp.cpp
paddle/function/ImageExpandOp.cpp
+0
-108
未找到文件。
paddle/function/Im2ColOp.cpp
0 → 100644
浏览文件 @
eb0c7e5e
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#include "Im2Col.h"
namespace
paddle
{
/*
* imShape = [inputChannels, inputHeight, inputWidth]
* colShape =
* [inputChannels, filterHeight, filterWidth, outputHeight, outputWidth]
*/
template
<
class
T
>
class
Im2ColFunctor
<
kCFO
,
DEVICE_TYPE_CPU
,
T
>
{
public:
void
operator
()(
const
T
*
imData
,
const
TensorShape
&
imShape
,
T
*
colData
,
const
TensorShape
&
colShape
,
int
strideHeight
,
int
strideWidth
,
int
paddingHeight
,
int
paddingWidth
)
{
int
inputChannels
=
imShape
[
0
];
int
inputHeight
=
imShape
[
1
];
int
inputWidth
=
imShape
[
2
];
int
filterHeight
=
colShape
[
1
];
int
filterWidth
=
colShape
[
2
];
int
outputHeight
=
colShape
[
3
];
int
outputWidth
=
colShape
[
4
];
int
channelsCol
=
inputChannels
*
filterHeight
*
filterWidth
;
for
(
int
c
=
0
;
c
<
channelsCol
;
++
c
)
{
int
wOffset
=
c
%
filterWidth
;
int
hOffset
=
(
c
/
filterWidth
)
%
filterHeight
;
int
c_im
=
c
/
filterWidth
/
filterHeight
;
for
(
int
h
=
0
;
h
<
outputHeight
;
++
h
)
{
for
(
int
w
=
0
;
w
<
outputWidth
;
++
w
)
{
int
imRowIdx
=
h
*
strideHeight
+
hOffset
;
int
imColIdx
=
w
*
strideWidth
+
wOffset
;
if
((
imRowIdx
-
paddingHeight
)
<
0
||
(
imRowIdx
-
paddingHeight
)
>=
inputHeight
||
(
imColIdx
-
paddingWidth
)
<
0
||
(
imColIdx
-
paddingWidth
)
>=
inputWidth
)
{
colData
[(
c
*
outputHeight
+
h
)
*
outputWidth
+
w
]
=
T
(
0
);
}
else
{
imRowIdx
+=
c_im
*
inputHeight
-
paddingHeight
;
imColIdx
-=
paddingWidth
;
colData
[(
c
*
outputHeight
+
h
)
*
outputWidth
+
w
]
=
imData
[
imRowIdx
*
inputWidth
+
imColIdx
];
}
}
}
}
}
};
/*
* imShape = [inputChannels, inputHeight, inputWidth]
* colShape =
* [inputChannels, filterHeight, filterWidth, outputHeight, outputWidth]
*/
template
<
class
T
>
class
Col2ImFunctor
<
kCFO
,
DEVICE_TYPE_CPU
,
T
>
{
public:
void
operator
()(
T
*
imData
,
const
TensorShape
&
imShape
,
const
T
*
colData
,
const
TensorShape
&
colShape
,
int
strideHeight
,
int
strideWidth
,
int
paddingHeight
,
int
paddingWidth
)
{
int
inputChannels
=
imShape
[
0
];
int
inputHeight
=
imShape
[
1
];
int
inputWidth
=
imShape
[
2
];
int
filterHeight
=
colShape
[
1
];
int
filterWidth
=
colShape
[
2
];
int
outputHeight
=
colShape
[
3
];
int
outputWidth
=
colShape
[
4
];
int
channelsCol
=
inputChannels
*
filterHeight
*
filterWidth
;
for
(
int
c
=
0
;
c
<
channelsCol
;
++
c
)
{
int
wOffset
=
c
%
filterWidth
;
int
hOffset
=
(
c
/
filterWidth
)
%
filterHeight
;
int
c_im
=
c
/
filterWidth
/
filterHeight
;
for
(
int
h
=
0
;
h
<
outputHeight
;
++
h
)
{
for
(
int
w
=
0
;
w
<
outputWidth
;
++
w
)
{
int
imRowIdx
=
h
*
strideHeight
+
hOffset
;
int
imColIdx
=
w
*
strideWidth
+
wOffset
;
if
((
imRowIdx
-
paddingHeight
)
>=
0
&&
(
imRowIdx
-
paddingHeight
)
<
inputHeight
&&
(
imColIdx
-
paddingWidth
)
>=
0
&&
(
imColIdx
-
paddingWidth
)
<
inputWidth
)
{
imRowIdx
+=
c_im
*
inputHeight
-
paddingHeight
;
imColIdx
-=
paddingWidth
;
imData
[
imRowIdx
*
inputWidth
+
imColIdx
]
+=
colData
[(
c
*
outputHeight
+
h
)
*
outputWidth
+
w
];
}
}
}
}
}
};
template
class
Im2ColFunctor
<
kCFO
,
DEVICE_TYPE_CPU
,
float
>;
template
class
Im2ColFunctor
<
kCFO
,
DEVICE_TYPE_CPU
,
double
>;
template
class
Col2ImFunctor
<
kCFO
,
DEVICE_TYPE_CPU
,
float
>;
template
class
Col2ImFunctor
<
kCFO
,
DEVICE_TYPE_CPU
,
double
>;
/*
* imShape = [inputChannels, inputHeight, inputWidth]
* colShape =
* [outputHeight, outputWidth, inputChannels, filterHeight, filterWidth]
*/
template
<
class
T
>
class
Im2ColFunctor
<
kOCF
,
DEVICE_TYPE_CPU
,
T
>
{
public:
void
operator
()(
const
T
*
imData
,
const
TensorShape
&
imShape
,
T
*
colData
,
const
TensorShape
&
colShape
,
int
strideHeight
,
int
strideWidth
,
int
paddingHeight
,
int
paddingWidth
)
{
int
inputChannels
=
imShape
[
0
];
int
inputHeight
=
imShape
[
1
];
int
inputWidth
=
imShape
[
2
];
int
filterHeight
=
colShape
[
3
];
int
filterWidth
=
colShape
[
4
];
int
outputHeight
=
colShape
[
0
];
int
outputWidth
=
colShape
[
1
];
for
(
int
outputH
=
0
;
outputH
<
outputHeight
;
++
outputH
)
{
for
(
int
outputW
=
0
;
outputW
<
outputWidth
;
++
outputW
)
{
for
(
int
channel
=
0
;
channel
<
inputChannels
;
++
channel
)
{
for
(
int
filterH
=
0
;
filterH
<
filterHeight
;
++
filterH
)
{
for
(
int
filterW
=
0
;
filterW
<
filterWidth
;
++
filterW
)
{
int
imRowOffset
=
outputH
*
strideHeight
+
filterH
-
paddingHeight
;
int
imColOffset
=
outputW
*
strideWidth
+
filterW
-
paddingWidth
;
int
colDataOffset
=
(((
outputH
*
outputWidth
+
outputW
)
*
inputChannels
+
channel
)
*
filterHeight
+
filterH
)
*
filterWidth
+
filterW
;
if
(
imRowOffset
<
0
||
imRowOffset
>=
inputHeight
||
imColOffset
<
0
||
imColOffset
>=
inputWidth
)
{
colData
[
colDataOffset
]
=
float
(
0
);
}
else
{
int
imDataOffset
=
(
channel
*
inputHeight
+
imRowOffset
)
*
inputWidth
+
imColOffset
;
colData
[
colDataOffset
]
=
imData
[
imDataOffset
];
}
}
}
}
}
}
}
};
/*
* imShape = [inputChannels, inputHeight, inputWidth]
* colShape =
* [outputHeight, outputWidth, inputChannels, filterHeight, filterWidth]
*/
template
<
class
T
>
class
Col2ImFunctor
<
kOCF
,
DEVICE_TYPE_CPU
,
T
>
{
public:
void
operator
()(
T
*
imData
,
const
TensorShape
&
imShape
,
const
T
*
colData
,
const
TensorShape
&
colShape
,
int
strideHeight
,
int
strideWidth
,
int
paddingHeight
,
int
paddingWidth
)
{
int
inputChannels
=
imShape
[
0
];
int
inputHeight
=
imShape
[
1
];
int
inputWidth
=
imShape
[
2
];
int
filterHeight
=
colShape
[
3
];
int
filterWidth
=
colShape
[
4
];
int
outputHeight
=
colShape
[
0
];
int
outputWidth
=
colShape
[
1
];
for
(
int
outputH
=
0
;
outputH
<
outputHeight
;
++
outputH
)
{
for
(
int
outputW
=
0
;
outputW
<
outputWidth
;
++
outputW
)
{
for
(
int
channel
=
0
;
channel
<
inputChannels
;
++
channel
)
{
for
(
int
filterH
=
0
;
filterH
<
filterHeight
;
++
filterH
)
{
for
(
int
filterW
=
0
;
filterW
<
filterWidth
;
++
filterW
)
{
int
imRowOffset
=
outputH
*
strideHeight
+
filterH
-
paddingHeight
;
int
imColOffset
=
outputW
*
strideWidth
+
filterW
-
paddingWidth
;
int
colDataOffset
=
(((
outputH
*
outputWidth
+
outputW
)
*
inputChannels
+
channel
)
*
filterHeight
+
filterH
)
*
filterWidth
+
filterW
;
if
(
imRowOffset
>=
0
&&
imRowOffset
<
inputHeight
&&
imColOffset
>=
0
&&
imColOffset
<
inputWidth
)
{
int
imDataOffset
=
(
channel
*
inputHeight
+
imRowOffset
)
*
inputWidth
+
imColOffset
;
imData
[
imDataOffset
]
+=
colData
[
colDataOffset
];
}
}
}
}
}
}
}
};
template
class
Im2ColFunctor
<
kOCF
,
DEVICE_TYPE_CPU
,
float
>;
template
class
Im2ColFunctor
<
kOCF
,
DEVICE_TYPE_CPU
,
double
>;
template
class
Col2ImFunctor
<
kOCF
,
DEVICE_TYPE_CPU
,
float
>;
template
class
Col2ImFunctor
<
kOCF
,
DEVICE_TYPE_CPU
,
double
>;
}
// namespace paddle
paddle/function/Im2ColOpGpu.cu
浏览文件 @
eb0c7e5e
...
...
@@ -57,6 +57,11 @@ void im2col(const T* data_im, int numOuts, int height, int width,
}
}
/*
* imShape = [inputChannels, inputHeight, inputWidth]
* colShape =
* [inputChannels, filterHeight, filterWidth, outputHeight, outputWidth]
*/
template
<
class
T
>
class
Im2ColFunctor
<
kCFO
,
DEVICE_TYPE_GPU
,
T
>
{
public:
...
...
@@ -71,10 +76,10 @@ public:
int
inputChannels
=
imShape
[
0
];
int
inputHeight
=
imShape
[
1
];
int
inputWidth
=
imShape
[
2
];
int
filterHeight
=
colShape
[
3
];
int
filterWidth
=
colShape
[
4
];
int
outputHeight
=
colShape
[
0
];
int
outputWidth
=
colShape
[
1
];
int
filterHeight
=
colShape
[
1
];
int
filterWidth
=
colShape
[
2
];
int
outputHeight
=
colShape
[
3
];
int
outputWidth
=
colShape
[
4
];
int
numKernels
=
inputChannels
*
outputHeight
*
outputWidth
;
int
blocks
=
(
numKernels
+
1024
-
1
)
/
1024
;
...
...
@@ -135,6 +140,11 @@ void col2im(size_t n, const T* data_col, size_t height,
}
}
/*
* imShape = [inputChannels, inputHeight, inputWidth]
* colShape =
* [inputChannels, filterHeight, filterWidth, outputHeight, outputWidth]
*/
template
<
class
T
>
class
Col2ImFunctor
<
kCFO
,
DEVICE_TYPE_GPU
,
T
>
{
public:
...
...
@@ -149,10 +159,10 @@ public:
int
inputChannels
=
imShape
[
0
];
int
inputHeight
=
imShape
[
1
];
int
inputWidth
=
imShape
[
2
];
int
filterHeight
=
colShape
[
3
];
int
filterWidth
=
colShape
[
4
];
int
outputHeight
=
colShape
[
0
];
int
outputWidth
=
colShape
[
1
];
int
filterHeight
=
colShape
[
1
];
int
filterWidth
=
colShape
[
2
];
int
outputHeight
=
colShape
[
3
];
int
outputWidth
=
colShape
[
4
];
size_t
numKernels
=
inputChannels
*
(
inputHeight
+
2
*
paddingHeight
)
*
(
inputWidth
+
2
*
paddingWidth
);
...
...
paddle/function/ImageExpandOp.cpp
浏览文件 @
eb0c7e5e
...
...
@@ -17,114 +17,6 @@ limitations under the License. */
namespace
paddle
{
/*
* imShape = [inputChannels, inputHeight, inputWidth]
* colShape =
* [outputHeight, outputWidth, inputChannels, filterHeight, filterWidth]
*/
template
<
class
T
>
class
Im2ColFunctor
<
kOCF
,
DEVICE_TYPE_CPU
,
T
>
{
public:
void
operator
()(
const
T
*
imData
,
const
TensorShape
&
imShape
,
T
*
colData
,
const
TensorShape
&
colShape
,
int
strideHeight
,
int
strideWidth
,
int
paddingHeight
,
int
paddingWidth
)
{
int
inputChannels
=
imShape
[
0
];
int
inputHeight
=
imShape
[
1
];
int
inputWidth
=
imShape
[
2
];
int
filterHeight
=
colShape
[
3
];
int
filterWidth
=
colShape
[
4
];
int
outputHeight
=
colShape
[
0
];
int
outputWidth
=
colShape
[
1
];
for
(
int
outputH
=
0
;
outputH
<
outputHeight
;
++
outputH
)
{
for
(
int
outputW
=
0
;
outputW
<
outputWidth
;
++
outputW
)
{
for
(
int
channel
=
0
;
channel
<
inputChannels
;
++
channel
)
{
for
(
int
filterH
=
0
;
filterH
<
filterHeight
;
++
filterH
)
{
for
(
int
filterW
=
0
;
filterW
<
filterWidth
;
++
filterW
)
{
int
imRowOffset
=
outputH
*
strideHeight
+
filterH
-
paddingHeight
;
int
imColOffset
=
outputW
*
strideWidth
+
filterW
-
paddingWidth
;
int
colDataOffset
=
(((
outputH
*
outputWidth
+
outputW
)
*
inputChannels
+
channel
)
*
filterHeight
+
filterH
)
*
filterWidth
+
filterW
;
if
(
imRowOffset
<
0
||
imRowOffset
>=
inputHeight
||
imColOffset
<
0
||
imColOffset
>=
inputWidth
)
{
colData
[
colDataOffset
]
=
float
(
0
);
}
else
{
int
imDataOffset
=
(
channel
*
inputHeight
+
imRowOffset
)
*
inputWidth
+
imColOffset
;
colData
[
colDataOffset
]
=
imData
[
imDataOffset
];
}
}
}
}
}
}
}
};
/*
* imShape = [inputChannels, inputHeight, inputWidth]
* colShape =
* [outputHeight, outputWidth, inputChannels, filterHeight, filterWidth]
*/
template
<
class
T
>
class
Col2ImFunctor
<
kOCF
,
DEVICE_TYPE_CPU
,
T
>
{
public:
void
operator
()(
T
*
imData
,
const
TensorShape
&
imShape
,
const
T
*
colData
,
const
TensorShape
&
colShape
,
int
strideHeight
,
int
strideWidth
,
int
paddingHeight
,
int
paddingWidth
)
{
int
inputChannels
=
imShape
[
0
];
int
inputHeight
=
imShape
[
1
];
int
inputWidth
=
imShape
[
2
];
int
filterHeight
=
colShape
[
3
];
int
filterWidth
=
colShape
[
4
];
int
outputHeight
=
colShape
[
0
];
int
outputWidth
=
colShape
[
1
];
for
(
int
outputH
=
0
;
outputH
<
outputHeight
;
++
outputH
)
{
for
(
int
outputW
=
0
;
outputW
<
outputWidth
;
++
outputW
)
{
for
(
int
channel
=
0
;
channel
<
inputChannels
;
++
channel
)
{
for
(
int
filterH
=
0
;
filterH
<
filterHeight
;
++
filterH
)
{
for
(
int
filterW
=
0
;
filterW
<
filterWidth
;
++
filterW
)
{
int
imRowOffset
=
outputH
*
strideHeight
+
filterH
-
paddingHeight
;
int
imColOffset
=
outputW
*
strideWidth
+
filterW
-
paddingWidth
;
int
colDataOffset
=
(((
outputH
*
outputWidth
+
outputW
)
*
inputChannels
+
channel
)
*
filterHeight
+
filterH
)
*
filterWidth
+
filterW
;
if
(
imRowOffset
>=
0
&&
imRowOffset
<
inputHeight
&&
imColOffset
>=
0
&&
imColOffset
<
inputWidth
)
{
int
imDataOffset
=
(
channel
*
inputHeight
+
imRowOffset
)
*
inputWidth
+
imColOffset
;
imData
[
imDataOffset
]
+=
colData
[
colDataOffset
];
}
}
}
}
}
}
}
};
/*
* \brief Converts the image data of four dimensions(NCHW) into
* a sequence data of three dimensions(NST) in the forward calculation,
...
...
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