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
马赛克医院
Paddle
提交
9bfa3013
P
Paddle
项目概览
马赛克医院
/
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
0
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
0
Issue
0
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 >>
未验证
提交
9bfa3013
编写于
12月 20, 2017
作者:
W
Wang Meng
提交者:
GitHub
12月 20, 2017
浏览文件
操作
浏览文件
下载
差异文件
Merge pull request #6631 from will-am/img_separable_conv
Add separable convolution
上级
e445b3ff
7be4f2b7
变更
1
隐藏空白更改
内联
并排
Showing
1 changed file
with
91 addition
and
12 deletion
+91
-12
python/paddle/trainer_config_helpers/networks.py
python/paddle/trainer_config_helpers/networks.py
+91
-12
未找到文件。
python/paddle/trainer_config_helpers/networks.py
浏览文件 @
9bfa3013
...
...
@@ -25,10 +25,10 @@ from paddle.trainer.config_parser import *
__all__
=
[
'sequence_conv_pool'
,
'simple_lstm'
,
"simple_img_conv_pool"
,
"img_conv_bn_pool"
,
'lstmemory_group'
,
'lstmemory_unit'
,
'small_vgg'
,
'img_conv_group'
,
'
vgg_16_network'
,
'gru_unit'
,
'gru_group'
,
'simple_gru
'
,
'
simple_attention'
,
'dot_product_attention'
,
'multi_head
_attention'
,
'
simple_gru2'
,
'bidirectional_gru'
,
'text_conv_pool'
,
'bidirectional_lstm
'
,
'inputs'
,
'outputs'
'img_conv_group'
,
'
img_separable_conv'
,
'vgg_16_network'
,
'gru_unit
'
,
'
gru_group'
,
'simple_gru'
,
'simple_attention'
,
'dot_product
_attention'
,
'
multi_head_attention'
,
'simple_gru2'
,
'bidirectional_gru
'
,
'
text_conv_pool'
,
'bidirectional_lstm'
,
'
inputs'
,
'outputs'
]
######################################################
...
...
@@ -251,13 +251,13 @@ def img_conv_bn_pool(input,
pool_layer_attr
=
None
):
"""
Convolution, batch normalization, pooling group.
Img input => Conv => BN => Pooling => Output.
:param name: group name.
:type name: basestring
:param input: input layer.
:type input: LayerOutput
:type input: LayerOutput
:param filter_size: see img_conv_layer for details.
:type filter_size: int
:param num_filters: see img_conv_layer for details.
...
...
@@ -435,6 +435,85 @@ def img_conv_group(input,
input
=
tmp
,
stride
=
pool_stride
,
pool_size
=
pool_size
,
pool_type
=
pool_type
)
@
wrap_name_default
(
"separable_conv"
)
def
img_separable_conv
(
input
,
num_channels
,
num_out_channels
,
filter_size
,
stride
=
1
,
padding
=
0
,
depth_multiplier
=
1
,
act
=
None
,
bias_attr
=
None
,
param_attr
=
None
,
shared_bias
=
True
,
layer_type
=
'exconv'
,
name
=
None
):
"""
Separable Convolution.
The separable convolution module is consisted of a depthwise convolution
that acts separately on input channels, followed by a pointwise convolution
with 1*1 kernels that mixes channels. It is used for Xception:
https://arxiv.org/pdf/1610.02357.pdf
:param input: input layer.
:type input: LayerOutput
:param num_channels: the number of input channels.
:type num_channels: int
:param num_out_channels: the number of output channels.
:type num_out_channels: int
:param filter_size: the filter size for the depthwise convolution.
:type filter_size: int|tuple
:param stride: the stride size for the depthwise convolution.
:type stride: int|tuple
:param padding: the padding size for the depthwise convolution.
:type padding: int|tuple
:param depth_multiplier: the number of filter for one channel in the
depthwize convolution.
:type depth_multiplier: int
:param act: the activation function for the output.
:type act: BaseActivation
:param bias_attr: see img_conv_layer for details.
:type bias_attr: ParameterAttribute
:param param_attr: see img_conv_layer for details.
:type param_attr: ParameterAttribute
:param shared_bias: see img_conv_layer for details.
:type shared_bias: bool
:param layer_type: see img_conv_layer for details.
:type layer_type: bool
:return: layer's output
:rtype: LayerOutput
"""
__depthwise_conv__
=
img_conv_layer
(
name
=
"%s_depthwise_conv"
%
name
,
input
=
input
,
num_channels
=
num_channels
,
num_filters
=
num_channels
*
depth_multiplier
,
groups
=
num_channels
,
filter_size
=
filter_size
,
stride
=
stride
,
padding
=
padding
,
act
=
LinearActivation
(),
bias_attr
=
bias_attr
,
param_attr
=
param_attr
,
shared_biases
=
shared_bias
,
layer_type
=
layer_type
)
__pointwise_conv__
=
img_conv_layer
(
name
=
"%s_pointwise_conv"
%
name
,
input
=
__depthwise_conv__
,
num_channels
=
num_channels
*
depth_multiplier
,
num_filters
=
num_out_channels
,
filter_size
=
1
,
stride
=
1
,
padding
=
0
,
act
=
act
,
bias_attr
=
bias_attr
,
param_attr
=
param_attr
,
shared_biases
=
shared_bias
)
return
__pointwise_conv__
def
small_vgg
(
input_image
,
num_channels
,
num_classes
):
def
__vgg__
(
ipt
,
num_filter
,
times
,
dropouts
,
num_channels_
=
None
):
return
img_conv_group
(
...
...
@@ -648,7 +727,7 @@ def lstmemory_unit(input,
lstm_bias_attr
=
None
,
lstm_layer_attr
=
None
):
"""
lstmemory_unit defines the caculation process of a LSTM unit during a
lstmemory_unit defines the caculation process of a LSTM unit during a
single time step. This function is not a recurrent layer, so it can not be
directly used to process sequence input. This function is always used in
recurrent_group (see layers.py for more details) to implement attention
...
...
@@ -869,7 +948,7 @@ def gru_unit(input,
gru_layer_attr
=
None
,
naive
=
False
):
"""
gru_unit defines the calculation process of a gated recurrent unit during a single
gru_unit defines the calculation process of a gated recurrent unit during a single
time step. This function is not a recurrent layer, so it can not be
directly used to process sequence input. This function is always used in
the recurrent_group (see layers.py for more details) to implement attention
...
...
@@ -1012,7 +1091,7 @@ def simple_gru(input,
simple_gru in network.py. The reason why there are so many interfaces is
that we have two ways to implement recurrent neural network. One way is to
use one complete layer to implement rnn (including simple rnn, gru and lstm)
with multiple time steps, such as recurrent_layer, lstmemory, grumemory. But
with multiple time steps, such as recurrent_layer, lstmemory, grumemory. But
the multiplication operation :math:`W x_t` is not computed in these layers.
See details in their interfaces in layers.py.
The other implementation is to use an recurrent group which can ensemble a
...
...
@@ -1116,7 +1195,7 @@ def simple_gru2(input,
:type act: BaseActivation
:param gate_act: gate activiation type of gru
:type gate_act: BaseActivation
:param gru_bias_attr: bias parameter attribute of gru layer,
:param gru_bias_attr: bias parameter attribute of gru layer,
False means no bias, None means default bias.
:type gru_bias_attr: ParameterAttribute|False|None
:param gru_param_attr: param parameter attribute of gru layer,
...
...
@@ -1189,7 +1268,7 @@ def bidirectional_gru(input,
:type size: int
:param return_seq: If set False, the last time step of output are
concatenated and returned.
If set True, the entire output sequences in forward
If set True, the entire output sequences in forward
and backward directions are concatenated and returned.
:type return_seq: bool
:return: LayerOutput object.
...
...
@@ -1278,7 +1357,7 @@ def bidirectional_lstm(input,
:type size: int
:param return_seq: If set False, the last time step of output are
concatenated and returned.
If set True, the entire output sequences in forward
If set True, the entire output sequences in forward
and backward directions are concatenated and returned.
:type return_seq: bool
:return: LayerOutput object.
...
...
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