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
PaddleDetection
提交
179dd0cb
P
PaddleDetection
项目概览
PaddlePaddle
/
PaddleDetection
大约 1 年 前同步成功
通知
694
Star
11112
Fork
2696
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
184
列表
看板
标记
里程碑
合并请求
40
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
PaddleDetection
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
184
Issue
184
列表
看板
标记
里程碑
合并请求
40
合并请求
40
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 >>
未验证
提交
179dd0cb
编写于
7月 25, 2018
作者:
T
tensor-tang
提交者:
GitHub
7月 25, 2018
浏览文件
操作
浏览文件
下载
差异文件
Merge pull request #12337 from tensor-tang/refine/im2col
refine cpu im2col no padding
上级
2d21aa76
b72befc5
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
102 addition
and
5 deletion
+102
-5
paddle/fluid/operators/math/im2col.cc
paddle/fluid/operators/math/im2col.cc
+30
-5
paddle/fluid/operators/math/im2col_test.cc
paddle/fluid/operators/math/im2col_test.cc
+72
-0
未找到文件。
paddle/fluid/operators/math/im2col.cc
浏览文件 @
179dd0cb
...
...
@@ -40,22 +40,47 @@ class Im2ColFunctor<paddle::operators::math::ColFormat::kCFO,
int
im_width
=
im
.
dims
()[
2
];
int
filter_height
=
col
->
dims
()[
1
];
int
filter_width
=
col
->
dims
()[
2
];
int
col
_height
=
col
->
dims
()[
3
];
int
col
_width
=
col
->
dims
()[
4
];
int
output
_height
=
col
->
dims
()[
3
];
int
output
_width
=
col
->
dims
()[
4
];
int
channels_col
=
im_channels
*
filter_height
*
filter_width
;
const
T
*
im_data
=
im
.
data
<
T
>
();
T
*
col_data
=
col
->
data
<
T
>
();
// TODO(TJ): change me to template
// further optimaze:
// 1. padding != 1
// 2. could also support stride_h != 1
if
(
stride
[
0
]
==
1
&&
stride
[
1
]
==
1
&&
dilation
[
0
]
==
1
&&
dilation
[
1
]
==
1
&&
padding
[
0
]
==
0
&&
padding
[
1
]
==
0
)
{
int
col_matrix_width
=
output_width
*
output_height
;
size_t
copy_size
=
sizeof
(
T
)
*
output_width
;
for
(
int
oh
=
0
;
oh
<
output_height
;
++
oh
)
{
const
T
*
im_data_start
=
im_data
+
oh
*
im_width
;
T
*
dst_data
=
col_data
+
oh
*
output_width
;
for
(
int
ic
=
0
;
ic
<
im_channels
;
++
ic
)
{
const
T
*
src_data
=
im_data_start
+
ic
*
im_height
*
im_width
;
for
(
int
kh
=
0
;
kh
<
filter_height
;
++
kh
)
{
for
(
int
kw
=
0
;
kw
<
filter_width
;
++
kw
)
{
std
::
memcpy
(
dst_data
,
src_data
+
kw
,
copy_size
);
dst_data
=
dst_data
+
col_matrix_width
;
}
src_data
=
src_data
+
im_width
;
}
}
}
return
;
}
for
(
int
c
=
0
;
c
<
channels_col
;
++
c
)
{
int
w_offset
=
c
%
filter_width
;
int
h_offset
=
(
c
/
filter_width
)
%
filter_height
;
int
c_im
=
c
/
(
filter_width
*
filter_height
);
for
(
int
h
=
0
;
h
<
col
_height
;
++
h
)
{
for
(
int
h
=
0
;
h
<
output
_height
;
++
h
)
{
int
im_row_idx
=
h
*
stride
[
0
]
-
padding
[
0
]
+
h_offset
*
dilation
[
0
];
for
(
int
w
=
0
;
w
<
col
_width
;
++
w
)
{
for
(
int
w
=
0
;
w
<
output
_width
;
++
w
)
{
int
im_col_idx
=
w
*
stride
[
1
]
-
padding
[
1
]
+
w_offset
*
dilation
[
1
];
int
col_idx
=
(
c
*
col_height
+
h
)
*
col
_width
+
w
;
int
col_idx
=
(
c
*
output_height
+
h
)
*
output
_width
+
w
;
int
im_idx
=
(
im_row_idx
+
c_im
*
im_height
)
*
im_width
+
im_col_idx
;
col_data
[
col_idx
]
=
(
im_row_idx
<
0
||
im_row_idx
>=
im_height
||
...
...
paddle/fluid/operators/math/im2col_test.cc
浏览文件 @
179dd0cb
...
...
@@ -160,8 +160,80 @@ void testIm2col() {
delete
context
;
}
void
testIm2colCPU
(
int
ic
,
int
ih
,
int
iw
,
int
fh
,
int
fw
,
int
ph
,
int
pw
)
{
paddle
::
framework
::
Tensor
input
;
paddle
::
framework
::
Tensor
output
;
paddle
::
framework
::
Tensor
ref_output
;
std
::
vector
<
int
>
padding
({
ph
,
pw
});
std
::
vector
<
int
>
stride
({
1
,
1
});
// stride_y, stride_x
std
::
vector
<
int
>
dilation
({
1
,
1
});
// dilation_y, dilation_x
int
output_height
=
(
ih
-
fh
+
padding
[
0
]
*
2
)
/
stride
[
0
]
+
1
;
int
output_width
=
(
iw
-
fw
+
padding
[
1
]
*
2
)
/
stride
[
1
]
+
1
;
float
*
input_ptr
=
input
.
mutable_data
<
float
>
({
ic
,
ih
,
iw
},
paddle
::
platform
::
CPUPlace
());
for
(
int
i
=
0
;
i
<
input
.
numel
();
++
i
)
{
input_ptr
[
i
]
=
static_cast
<
float
>
(
i
+
1
);
}
paddle
::
platform
::
CPUPlace
place
;
paddle
::
platform
::
CPUDeviceContext
context
(
place
);
output
.
mutable_data
<
float
>
({
ic
,
fh
,
fw
,
output_height
,
output_width
},
place
);
ref_output
.
mutable_data
<
float
>
({
ic
,
fh
,
fw
,
output_height
,
output_width
},
place
);
paddle
::
operators
::
math
::
Im2ColFunctor
<
paddle
::
operators
::
math
::
ColFormat
::
kCFO
,
paddle
::
platform
::
CPUDeviceContext
,
float
>
im2col
;
im2col
(
context
,
input
,
dilation
,
stride
,
padding
,
&
output
);
auto
ref_im2col
=
[
&
](
const
paddle
::
framework
::
Tensor
&
im
,
const
std
::
vector
<
int
>&
dilation
,
const
std
::
vector
<
int
>&
stride
,
const
std
::
vector
<
int
>&
padding
,
paddle
::
framework
::
Tensor
*
col
)
{
int
im_channels
=
im
.
dims
()[
0
];
int
im_height
=
im
.
dims
()[
1
];
int
im_width
=
im
.
dims
()[
2
];
int
filter_height
=
col
->
dims
()[
1
];
int
filter_width
=
col
->
dims
()[
2
];
int
output_height
=
col
->
dims
()[
3
];
int
output_width
=
col
->
dims
()[
4
];
int
channels_col
=
im_channels
*
filter_height
*
filter_width
;
const
float
*
im_data
=
im
.
data
<
float
>
();
float
*
col_data
=
col
->
data
<
float
>
();
for
(
int
c
=
0
;
c
<
channels_col
;
++
c
)
{
int
w_offset
=
c
%
filter_width
;
int
h_offset
=
(
c
/
filter_width
)
%
filter_height
;
int
c_im
=
c
/
(
filter_width
*
filter_height
);
for
(
int
h
=
0
;
h
<
output_height
;
++
h
)
{
int
im_row_idx
=
h
*
stride
[
0
]
-
padding
[
0
]
+
h_offset
*
dilation
[
0
];
for
(
int
w
=
0
;
w
<
output_width
;
++
w
)
{
int
im_col_idx
=
w
*
stride
[
1
]
-
padding
[
1
]
+
w_offset
*
dilation
[
1
];
int
col_idx
=
(
c
*
output_height
+
h
)
*
output_width
+
w
;
int
im_idx
=
(
im_row_idx
+
c_im
*
im_height
)
*
im_width
+
im_col_idx
;
col_data
[
col_idx
]
=
(
im_row_idx
<
0
||
im_row_idx
>=
im_height
||
im_col_idx
<
0
||
im_col_idx
>=
im_width
)
?
0.
f
:
im_data
[
im_idx
];
}
}
}
};
ref_im2col
(
input
,
dilation
,
stride
,
padding
,
&
ref_output
);
float
*
out_cfo_ptr
=
output
.
data
<
float
>
();
float
*
out_ref_ptr
=
ref_output
.
data
<
float
>
();
for
(
int
i
=
0
;
i
<
output
.
numel
();
++
i
)
{
EXPECT_EQ
(
out_cfo_ptr
[
i
],
out_ref_ptr
[
i
]);
}
}
TEST
(
math
,
im2col
)
{
testIm2col
<
paddle
::
platform
::
CPUDeviceContext
,
paddle
::
platform
::
CPUPlace
>
();
testIm2colCPU
(
/*ic*/
3
,
/*ih*/
5
,
/*iw*/
5
,
/*fh*/
3
,
/*fw*/
2
,
/*ph*/
0
,
/*pw*/
0
);
testIm2colCPU
(
/*ic*/
2
,
/*ih*/
5
,
/*iw*/
4
,
/*fh*/
3
,
/*fw*/
3
,
/*ph*/
1
,
/*pw*/
1
);
#ifdef PADDLE_WITH_CUDA
testIm2col
<
paddle
::
platform
::
CUDADeviceContext
,
paddle
::
platform
::
CUDAPlace
>
();
...
...
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