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
Mr.Vain
Mace
提交
7bb7468c
Mace
项目概览
Mr.Vain
/
Mace
与 Fork 源项目一致
Fork自
Xiaomi / Mace
通知
1
Star
1
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
DevOps
流水线
流水线任务
计划
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
Mace
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
DevOps
DevOps
流水线
流水线任务
计划
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
流水线任务
提交
Issue看板
体验新版 GitCode,发现更多精彩内容 >>
提交
7bb7468c
编写于
4月 27, 2018
作者:
李
李寅
浏览文件
操作
浏览文件
下载
差异文件
Merge branch 'feature_wuch' into 'master'
fix memory optimization See merge request !430
上级
1a86f7b8
78f9a398
变更
4
隐藏空白更改
内联
并排
Showing
4 changed file
with
26 addition
and
3 deletion
+26
-3
mace/core/tensor.h
mace/core/tensor.h
+1
-1
mace/core/workspace.cc
mace/core/workspace.cc
+5
-1
mace/python/tools/caffe_converter_lib.py
mace/python/tools/caffe_converter_lib.py
+16
-1
mace/python/tools/tf_converter_lib.py
mace/python/tools/tf_converter_lib.py
+4
-0
未找到文件。
mace/core/tensor.h
浏览文件 @
7bb7468c
...
...
@@ -212,7 +212,7 @@ class Tensor {
image_shape_
.
clear
();
if
(
buffer_
!=
nullptr
)
{
MACE_CHECK
(
!
has_opencl_image
(),
"Cannot resize image, use ResizeImage."
);
buffer_
->
Resize
(
raw_size
());
if
(
raw_size
()
>
buffer_
->
size
())
buffer_
->
Resize
(
raw_size
());
}
else
{
MACE_CHECK
(
is_buffer_owner_
);
buffer_
=
new
Buffer
(
allocator_
,
raw_size
());
...
...
mace/core/workspace.cc
浏览文件 @
7bb7468c
...
...
@@ -170,7 +170,6 @@ void Workspace::CreateOutputTensorBuffer(const NetDef &net_def,
std
::
unique_ptr
<
Tensor
>
tensor
(
new
Tensor
(
preallocated_allocator_
.
GetBuffer
(
mem_ids
[
i
]),
dtype
));
tensor
->
SetSourceOpName
(
op
.
name
());
tensor_map_
[
op
.
output
(
i
)]
=
std
::
move
(
tensor
);
if
(
device_type
==
DeviceType
::
OPENCL
)
{
VLOG
(
3
)
<<
"Tensor: "
<<
op
.
name
()
<<
"("
<<
op
.
type
()
<<
")"
<<
" Mem: "
<<
mem_ids
[
i
]
...
...
@@ -180,7 +179,12 @@ void Workspace::CreateOutputTensorBuffer(const NetDef &net_def,
<<
", "
<<
dynamic_cast
<
Image
*>
(
tensor
->
UnderlyingBuffer
())
->
image_shape
()[
1
];
}
else
if
(
device_type
==
DeviceType
::
CPU
)
{
VLOG
(
3
)
<<
"Tensor: "
<<
op
.
name
()
<<
"("
<<
op
.
type
()
<<
")"
<<
" Mem: "
<<
mem_ids
[
i
]
<<
", Buffer size: "
<<
tensor
->
UnderlyingBuffer
()
->
size
();
}
tensor_map_
[
op
.
output
(
i
)]
=
std
::
move
(
tensor
);
}
}
}
...
...
mace/python/tools/caffe_converter_lib.py
浏览文件 @
7bb7468c
...
...
@@ -529,7 +529,8 @@ class CaffeConverter(object):
# Add filter
weight_tensor_name
=
op
.
name
+
'_weight:0'
weight_data
=
op
.
data
[
0
]
# OIHW
input_shape
=
op
.
data
[
1
].
shape
input_shape
=
op
.
get_single_parent
().
output_shape_map
[
op
.
layer
.
bottom
[
0
]]
if
input_shape
[
2
]
>
16
and
input_shape
[
3
]
>
16
:
G
=
np
.
array
([
[
1.0
,
0.0
,
0.0
],
...
...
@@ -1085,6 +1086,13 @@ class CaffeConverter(object):
arg
.
name
=
'T'
arg
.
i
=
self
.
dt
input_op
=
self
.
ops_map
[
name
]
if
input_op
.
layer
is
not
None
:
output_shape
=
input_op
.
output_shape_map
[
input_op
.
layer
.
top
[
0
]]
else
:
output_shape
=
input_op
.
output_shape_map
[
input_op
.
name
]
self
.
add_output_shape
(
op_def
,
output_shape
)
def
add_cpu_output_transform
(
self
,
names
):
for
name
in
names
:
output_name
=
MACE_OUTPUT_NODE_NAME
+
'_'
+
name
+
":0"
...
...
@@ -1098,6 +1106,13 @@ class CaffeConverter(object):
dims_arg
.
name
=
'dims'
dims_arg
.
ints
.
extend
([
0
,
2
,
3
,
1
])
# NCHW -> NHWC
input_op
=
self
.
ops_map
[
name
]
if
input_op
.
layer
is
not
None
:
output_shape
=
input_op
.
output_shape_map
[
input_op
.
layer
.
top
[
0
]]
else
:
output_shape
=
input_op
.
output_shape_map
[
input_op
.
name
]
self
.
add_output_shape
(
op_def
,
output_shape
)
def
convert
(
self
,
input_nodes
,
input_shapes
,
output_nodes
):
assert
self
.
ops
[
0
].
type
==
'Input'
self
.
add_input_op_shape
(
input_nodes
,
input_shapes
)
...
...
mace/python/tools/tf_converter_lib.py
浏览文件 @
7bb7468c
...
...
@@ -174,6 +174,8 @@ class TFConverter(object):
arg
.
name
=
'T'
arg
.
i
=
self
.
dt
self
.
add_output_shape
(
self
.
ops
[
name
].
outputs
,
op_def
)
def
add_gpu_output_transform
(
self
,
names
):
for
name
in
names
:
output_name
=
MACE_OUTPUT_NODE_NAME
+
'_'
+
name
+
":0"
...
...
@@ -200,6 +202,8 @@ class TFConverter(object):
dims_arg
.
name
=
'dims'
dims_arg
.
ints
.
extend
([
0
,
2
,
3
,
1
])
self
.
add_output_shape
(
self
.
ops
[
name
].
outputs
,
op_def
)
@
staticmethod
def
add_output_shape
(
outputs
,
op
):
output_shapes
=
[]
...
...
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