def small_vgg(input_image, num_channels, num_classes):
def __vgg__(ipt, num_filter, times, dropouts, num_channels_=None):
return img_conv_group(input=ipt,
num_channels=num_channels_,
pool_size=2,
pool_stride=2,
conv_num_filter=[num_filter] * times,
conv_filter_size=3,
conv_act=ReluActivation(),
conv_with_batchnorm=True,
conv_batchnorm_drop_rate=dropouts,
pool_type=MaxPooling())
tmp = __vgg__(input_image, 64, 2, [0.3, 0], num_channels)
tmp = __vgg__(tmp, 128, 2, [0.4, 0])
tmp = __vgg__(tmp, 256, 3, [0.4, 0.4, 0])
tmp = __vgg__(tmp, 512, 3, [0.4, 0.4, 0])
tmp = img_pool_layer(input=tmp,
stride=2,
pool_size=2,
pool_type=MaxPooling())
tmp = dropout_layer(input=tmp, dropout_rate=0.5)
tmp = fc_layer(input=tmp,
size=512,
layer_attr=ExtraAttr(drop_rate=0.5),
act=LinearActivation())
tmp = batch_norm_layer(input=tmp, act=ReluActivation())
return fc_layer(input=tmp, size=num_classes, act=SoftmaxActivation())
def dropout_layer(input, dropout_rate, name=None):
"""
@TODO(yuyang18): Add comments.
:param name:
:param input:
:param dropout_rate:
:return:
"""
return addto_layer(name=name, input=input, act=LinearActivation(),
bias_attr=False,
layer_attr=ExtraAttr(drop_rate=dropout_rate))
def sequence_conv_pool(input,
context_len,
hidden_size,
name=None,
context_start=None,
pool_type=None,
context_proj_layer_name=None,
context_proj_param_attr=False,
fc_layer_name=None,
fc_param_attr=None,
fc_bias_attr=None,
fc_act=None,
pool_bias_attr=None,
fc_attr=None,
context_attr=None,
pool_attr=None):
"""
Text convolution pooling layers helper.
Text input => Context Projection => FC Layer => Pooling => Output.
:param name: name of output layer(pooling layer name)
:type name: basestring
:param input: name of input layer
:type input: LayerOutput
:param context_len: context projection length. See
context_projection's document.
:type context_len: int
:param hidden_size: FC Layer size.
:type hidden_size: int
:param context_start: context projection length. See
context_projection's context_start.
:type context_start: int or None
:param pool_type: pooling layer type. See pooling_layer's document.
:type pool_type: BasePoolingType.
:param context_proj_layer_name: context projection layer name.
None if user don't care.
:type context_proj_layer_name: basestring
:param context_proj_param_attr: context projection parameter attribute.
None if user don't care.
:type context_proj_param_attr: ParameterAttribute or None.
:param fc_layer_name: fc layer name. None if user don't care.
:type fc_layer_name: basestring
:param fc_param_attr: fc layer parameter attribute. None if user don't care.
:type fc_param_attr: ParameterAttribute or None
:param fc_bias_attr: fc bias parameter attribute. False if no bias,
None if user don't care.
:type fc_bias_attr: ParameterAttribute or None
:param fc_act: fc layer activation type. None means tanh
:type fc_act: BaseActivation
:param pool_bias_attr: pooling layer bias attr. None if don't care.
False if no bias.
:type pool_bias_attr: ParameterAttribute or None.
:param fc_attr: fc layer extra attribute.
:type fc_attr: ExtraLayerAttribute
:param context_attr: context projection layer extra attribute.
:type context_attr: ExtraLayerAttribute
:param pool_attr: pooling layer extra attribute.
:type pool_attr: ExtraLayerAttribute
:return: output layer name.
:rtype: LayerOutput
"""
# Set Default Value to param
context_proj_layer_name = "%s_conv_proj" % name \
if context_proj_layer_name is None else context_proj_layer_name
with mixed_layer(name=context_proj_layer_name,
size=input.size * context_len,
act=LinearActivation(),
layer_attr=context_attr) as m:
m += context_projection(input,
context_len=context_len,
context_start=context_start,
padding_attr=context_proj_param_attr)
fc_layer_name = "%s_conv_fc" % name \
if fc_layer_name is None else fc_layer_name
fl = fc_layer(name=fc_layer_name,
input=m,
size=hidden_size,
act=fc_act,
layer_attr=fc_attr,
param_attr=fc_param_attr,
bias_attr=fc_bias_attr)
return pooling_layer(name=name,
input=fl,
pooling_type=pool_type,
bias_attr=pool_bias_attr,
layer_attr=pool_attr)
def img_conv_group(input,
conv_num_filter,
pool_size,
num_channels=None,
conv_padding=1,
conv_filter_size=3,
conv_act=None,
conv_with_batchnorm=False,
conv_batchnorm_drop_rate=0,
pool_stride=1,
pool_type=None):
"""
Image Convolution Group, Used for vgg net.
TODO(yuyang18): Complete docs
:param conv_batchnorm_drop_rate:
:param input:
:param conv_num_filter:
:param pool_size:
:param num_channels:
:param conv_padding:
:param conv_filter_size:
:param conv_act:
:param conv_with_batchnorm:
:param pool_stride:
:param pool_type:
:return:
"""
tmp = input
# Type checks
assert isinstance(tmp, LayerOutput)
assert isinstance(conv_num_filter, list) or isinstance(
conv_num_filter, tuple)
for each_num_filter in conv_num_filter:
assert isinstance(each_num_filter, int)
assert isinstance(pool_size, int)
def __extend_list__(obj):
if not hasattr(obj, '__len__'):
return [obj] * len(conv_num_filter)
else:
return obj
conv_padding = __extend_list__(conv_padding)
conv_filter_size = __extend_list__(conv_filter_size)
conv_act = __extend_list__(conv_act)
conv_with_batchnorm = __extend_list__(conv_with_batchnorm)
conv_batchnorm_drop_rate = __extend_list__(conv_batchnorm_drop_rate)
for i in xrange(len(conv_num_filter)):
extra_kwargs = dict()
if num_channels is not None:
extra_kwargs['num_channels'] = num_channels
num_channels = None
if conv_with_batchnorm[i]:
extra_kwargs['act'] = LinearActivation()
else:
extra_kwargs['act'] = conv_act[i]
tmp = img_conv_layer(input=tmp,
padding=conv_padding[i],
filter_size=conv_filter_size[i],
num_filters=conv_num_filter[i],
**extra_kwargs)
# logger.debug("tmp.num_filters = %d" % tmp.num_filters)
if conv_with_batchnorm[i]:
dropout = conv_batchnorm_drop_rate[i]
if dropout == 0 or abs(dropout) < 1e-5: # dropout not set
tmp = batch_norm_layer(input=tmp, act=conv_act[i])
else:
tmp = batch_norm_layer(input=tmp,
act=conv_act[i],
layer_attr=ExtraAttr(drop_rate=dropout))
return img_pool_layer(input=tmp,
stride=pool_stride,
pool_size=pool_size,
pool_type=pool_type)
def img_conv_bn_pool(input,
filter_size,
num_filters,
pool_size,
name=None,
pool_type=None,
act=None,
groups=1,
conv_stride=1,
conv_padding=0,
conv_bias_attr=None,
num_channel=None,
conv_param_attr=None,
shared_bias=True,
conv_layer_attr=None,
bn_param_attr=None,
bn_bias_attr=None,
bn_layer_attr=None,
pool_stride=1,
pool_padding=0,
pool_layer_attr=None):
"""
Convolution, batch normalization, pooling group.
:param name: group name
:type name: basestring
:param input: layer's input
:type input: LayerOutput
:param filter_size: see img_conv_layer's document
:type filter_size: int
:param num_filters: see img_conv_layer's document
:type num_filters: int
:param pool_size: see img_pool_layer's document.
:type pool_size: int
:param pool_type: see img_pool_layer's document.
:type pool_type: BasePoolingType
:param act: see batch_norm_layer's document.
:type act: BaseActivation
:param groups: see img_conv_layer's document
:type groups: int
:param conv_stride: see img_conv_layer's document.
:type conv_stride: int
:param conv_padding: see img_conv_layer's document.
:type conv_padding: int
:param conv_bias_attr: see img_conv_layer's document.
:type conv_bias_attr: ParameterAttribute
:param num_channel: see img_conv_layer's document.
:type num_channel: int
:param conv_param_attr: see img_conv_layer's document.
:type conv_param_attr: ParameterAttribute
:param shared_bias: see img_conv_layer's document.
:type shared_bias: bool
:param conv_layer_attr: see img_conv_layer's document.
:type conv_layer_attr: ExtraLayerOutput
:param bn_param_attr: see batch_norm_layer's document.
:type bn_param_attr: ParameterAttribute.
:param bn_bias_attr: see batch_norm_layer's document.
:param bn_layer_attr: ParameterAttribute.
:param pool_stride: see img_pool_layer's document.
:type pool_stride: int
:param pool_padding: see img_pool_layer's document.
:type pool_padding: int
:param pool_layer_attr: see img_pool_layer's document.
:type pool_layer_attr: ExtraLayerAttribute
:return: Layer groups output
:rtype: LayerOutput
"""
__conv__ = img_conv_layer(name="%s_conv" % name,
input=input,
filter_size=filter_size,
num_filters=num_filters,
num_channels=num_channel,
act=LinearActivation(),
groups=groups,
stride=conv_stride,
padding=conv_padding,
bias_attr=conv_bias_attr,
param_attr=conv_param_attr,
shared_biases=shared_bias,
layer_attr=conv_layer_attr)
__bn__ = batch_norm_layer(name="%s_bn" % name,
input=__conv__,
act=act,
bias_attr=bn_bias_attr,
param_attr=bn_param_attr,
layer_attr=bn_layer_attr)
return img_pool_layer(name="%s_pool" % name,
input=__bn__,
pool_type=pool_type,
pool_size=pool_size,
stride=pool_stride,
padding=pool_padding,
layer_attr=pool_layer_attr)
