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
OPTHREE
Paddle
提交
8ee866bf
P
Paddle
项目概览
OPTHREE
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
1
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
1
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
1
Issue
1
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
8ee866bf
编写于
2月 27, 2019
作者:
D
dengkaipeng
提交者:
ceci3
3月 06, 2019
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
fix format. test=develop
上级
9c47f36d
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
4 addition
and
4 deletion
+4
-4
paddle/fluid/operators/spectral_norm_op.cc
paddle/fluid/operators/spectral_norm_op.cc
+2
-2
paddle/fluid/operators/spectral_norm_op.h
paddle/fluid/operators/spectral_norm_op.h
+2
-2
未找到文件。
paddle/fluid/operators/spectral_norm_op.cc
浏览文件 @
8ee866bf
...
...
@@ -94,7 +94,7 @@ class SpectralNormOpMaker : public framework::OpProtoAndCheckerMaker {
"W is the 2nd dimentions of Weight after reshape"
"corresponding by Attr(dim). As for Attr(dim) = 1"
"in conv2d layer with weight shape [M, C, K1, K2]"
"Weight will be reshape to [C, M*K1*K
w
], V will"
"Weight will be reshape to [C, M*K1*K
2
], V will"
"be in shape [M*K1*K2, 1]."
);
AddOutput
(
"Out"
,
"The output weight tensor of spectral_norm operator, "
...
...
@@ -105,7 +105,7 @@ class SpectralNormOpMaker : public framework::OpProtoAndCheckerMaker {
"it should be set as 0 if Input(Weight) is the"
"weight of fc layer, and should be set as 1 if"
"Input(Weight) is the weight of conv layer,"
"default is 0."
"default is 0."
)
.
SetDefault
(
0
);
AddAttr
<
int
>
(
"power_iters"
,
"number of power iterations to calculate"
...
...
paddle/fluid/operators/spectral_norm_op.h
浏览文件 @
8ee866bf
...
...
@@ -73,13 +73,13 @@ static inline void CalcMatrixSigmaAndNormWeight(
const
int
w
=
weight
->
dims
()[
1
];
for
(
int
i
=
0
;
i
<
power_iters
;
i
++
)
{
// V = W^T * U / ||W^T * U||_2
// V = W^T * U / ||W^T * U||_2
blas
.
MatMul
(
*
weight
,
true
,
*
u
,
false
,
T
(
1
),
v
,
T
(
0
));
auto
v_t_norm
=
v_t
.
square
().
sum
().
sqrt
().
eval
().
reshape
(
Array1
(
1
)).
broadcast
(
Array1
(
w
));
v_t
.
device
(
place
)
=
v_t
/
(
v_t_norm
+
v_t_norm
.
constant
(
eps
));
// U = W^T * V / ||W^T * V||_2
// U = W^T * V / ||W^T * V||_2
blas
.
MatMul
(
*
weight
,
false
,
*
v
,
false
,
T
(
1
),
u
,
T
(
0
));
auto
u_t_norm
=
u_t
.
square
().
sum
().
sqrt
().
eval
().
reshape
(
Array1
(
1
)).
broadcast
(
...
...
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