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
曾经的那一瞬间
Models
提交
f13be764
M
Models
项目概览
曾经的那一瞬间
/
Models
大约 1 年 前同步成功
通知
1
Star
0
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
M
Models
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
f13be764
编写于
1月 18, 2022
作者:
L
Liangzhe Yuan
提交者:
A. Unique TensorFlower
1月 18, 2022
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
internal change.
PiperOrigin-RevId: 422665603
上级
27fb855b
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
31 addition
and
8 deletion
+31
-8
official/projects/movinet/modeling/movinet_model.py
official/projects/movinet/modeling/movinet_model.py
+29
-7
official/projects/movinet/train.py
official/projects/movinet/train.py
+2
-1
未找到文件。
official/projects/movinet/modeling/movinet_model.py
浏览文件 @
f13be764
...
...
@@ -88,14 +88,13 @@ class MovinetClassifier(tf.keras.Model):
# Move backbone after super() call so Keras is happy
self
.
_backbone
=
backbone
def
_build_
network
(
def
_build_
backbone
(
self
,
backbone
:
tf
.
keras
.
Model
,
input_specs
:
Mapping
[
str
,
tf
.
keras
.
layers
.
InputSpec
],
state_specs
:
Optional
[
Mapping
[
str
,
tf
.
keras
.
layers
.
InputSpec
]]
=
None
,
)
->
Tuple
[
Mapping
[
str
,
tf
.
keras
.
Input
],
Union
[
Tuple
[
Mapping
[
# pytype: disable=invalid-annotation # typed-keras
str
,
tf
.
Tensor
],
Mapping
[
str
,
tf
.
Tensor
]],
Mapping
[
str
,
tf
.
Tensor
]]]:
"""Builds the model network.
)
->
Tuple
[
Mapping
[
str
,
Any
],
Any
,
Any
]:
"""Builds the backbone network and gets states and endpoints.
Args:
backbone: the model backbone.
...
...
@@ -104,9 +103,9 @@ class MovinetClassifier(tf.keras.Model):
layer, will overwrite the contents of the buffer(s).
Returns:
Inputs and outputs as a tuple. Inputs are expected to be a dict with
base input and states. Outputs are expected to be a dict of endpoints
and (optionally) output
states.
inputs: a dict of input specs.
endpoints: a dict of model endpoints.
states: a dict of model
states.
"""
state_specs
=
state_specs
if
state_specs
is
not
None
else
{}
...
...
@@ -145,7 +144,30 @@ class MovinetClassifier(tf.keras.Model):
mismatched_shapes
))
else
:
endpoints
,
states
=
backbone
(
inputs
)
return
inputs
,
endpoints
,
states
def
_build_network
(
self
,
backbone
:
tf
.
keras
.
Model
,
input_specs
:
Mapping
[
str
,
tf
.
keras
.
layers
.
InputSpec
],
state_specs
:
Optional
[
Mapping
[
str
,
tf
.
keras
.
layers
.
InputSpec
]]
=
None
,
)
->
Tuple
[
Mapping
[
str
,
tf
.
keras
.
Input
],
Union
[
Tuple
[
Mapping
[
# pytype: disable=invalid-annotation # typed-keras
str
,
tf
.
Tensor
],
Mapping
[
str
,
tf
.
Tensor
]],
Mapping
[
str
,
tf
.
Tensor
]]]:
"""Builds the model network.
Args:
backbone: the model backbone.
input_specs: the model input spec to use.
state_specs: a dict of states such that, if any of the keys match for a
layer, will overwrite the contents of the buffer(s).
Returns:
Inputs and outputs as a tuple. Inputs are expected to be a dict with
base input and states. Outputs are expected to be a dict of endpoints
and (optionally) output states.
"""
inputs
,
endpoints
,
states
=
self
.
_build_backbone
(
backbone
=
backbone
,
input_specs
=
input_specs
,
state_specs
=
state_specs
)
x
=
endpoints
[
'head'
]
x
=
movinet_layers
.
ClassifierHead
(
...
...
official/projects/movinet/train.py
浏览文件 @
f13be764
...
...
@@ -46,7 +46,8 @@ from official.modeling import performance
# Import movinet libraries to register the backbone and model into tf.vision
# model garden factory.
# pylint: disable=unused-import
# the followings are the necessary imports.
from
official.projects.movinet.google.configs
import
movinet_google
from
official.projects.movinet.google.modeling
import
movinet_model_google
from
official.projects.movinet.modeling
import
movinet
from
official.projects.movinet.modeling
import
movinet_model
# pylint: enable=unused-import
...
...
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