def _add_topdown_lateral(self, body_name, body_input, upper_output):
lateral_name = 'fpn_inner_' + body_name + '_lateral'
topdown_name = 'fpn_topdown_' + body_name
fan = body_input.shape[1]
if self.norm_type:
initializer = Xavier(fan_out=fan)
lateral = ConvNorm(
body_input,
self.num_chan,
1,
initializer=initializer,
norm_type=self.norm_type,
name=lateral_name,
bn_name=lateral_name)
else:
lateral = fluid.layers.conv2d(
body_input,
self.num_chan,
1,
param_attr=ParamAttr(
name=lateral_name + "_w", initializer=Xavier(fan_out=fan)),
bias_attr=ParamAttr(
name=lateral_name + "_b",
learning_rate=2.,
regularizer=L2Decay(0.)),
name=lateral_name)
shape = fluid.layers.shape(upper_output)
shape_hw = fluid.layers.slice(shape, axes=[0], starts=[2], ends=[4])
out_shape_ = shape_hw * 2
out_shape = fluid.layers.cast(out_shape_, dtype='int32')
out_shape.stop_gradient = True
topdown = fluid.layers.resize_nearest(
upper_output, scale=2., actual_shape=out_shape, name=topdown_name)
return lateral + topdown
def __call__(self, roi_feat, wb_scalar=1.0, name=''):
conv = roi_feat
fan = self.conv_dim * 3 * 3
initializer = MSRA(uniform=False, fan_in=fan)
for i in range(self.num_conv):
name = 'bbox_head_conv' + str(i)
conv = ConvNorm(conv,
self.conv_dim,
3,
act='relu',
initializer=initializer,
norm_type=self.norm_type,
freeze_norm=self.freeze_norm,
lr_scale=wb_scalar,
name=name,
norm_name=name)
fan = conv.shape[1] * conv.shape[2] * conv.shape[3]
head_heat = fluid.layers.fc(
input=conv,
size=self.mlp_dim,
act='relu',
name='fc6' + name,
param_attr=ParamAttr(name='fc6%s_w' % name,
initializer=Xavier(fan_out=fan),
learning_rate=wb_scalar),
bias_attr=ParamAttr(name='fc6%s_b' % name,
regularizer=L2Decay(0.),
learning_rate=wb_scalar * 2))
return head_heat
def _add_topdown_lateral(self, body_name, body_input, upper_output):
lateral_name = 'fpn_inner_' + body_name + '_lateral'
topdown_name = 'fpn_topdown_' + body_name
fan = body_input.shape[1]
if self.norm_type:
initializer = Xavier(fan_out=fan)
lateral = ConvNorm(body_input,
self.num_chan,
1,
initializer=initializer,
norm_type=self.norm_type,
freeze_norm=self.freeze_norm,
name=lateral_name,
norm_name=lateral_name)
else:
lateral = fluid.layers.conv2d(
body_input,
self.num_chan,
1,
param_attr=ParamAttr(name=lateral_name + "_w",
initializer=Xavier(fan_out=fan)),
bias_attr=ParamAttr(name=lateral_name + "_b",
learning_rate=2.,
regularizer=L2Decay(0.)),
name=lateral_name)
topdown = fluid.layers.resize_nearest(upper_output,
scale=2.,
name=topdown_name)
return lateral + topdown
def dense_aspp_block(self, input, num_filters1, num_filters2,
dilation_rate, dropout_prob, name):
conv = ConvNorm(input,
num_filters=num_filters1,
filter_size=1,
stride=1,
groups=1,
norm_decay=0.,
norm_type='gn',
norm_groups=self.norm_groups,
dilation=dilation_rate,
lr_scale=1,
freeze_norm=False,
act="relu",
norm_name=name + "_gn",
initializer=None,
bias_attr=False,
name=name + "_gn")
conv = fluid.layers.conv2d(
conv,
num_filters2,
filter_size=3,
padding=dilation_rate,
dilation=dilation_rate,
act="relu",
param_attr=ParamAttr(name=name + "_conv_w"),
bias_attr=ParamAttr(name=name + "_conv_b"),
)
if dropout_prob > 0:
conv = fluid.layers.dropout(conv, dropout_prob=dropout_prob)
return conv
def _lite_conv(self, x, out_c, act=None, name=None):
conv1 = ConvNorm(input=x,
num_filters=x.shape[1],
filter_size=5,
groups=x.shape[1],
norm_type='bn',
act='relu6',
initializer=Xavier(),
name=name + '.depthwise',
norm_name=name + '.depthwise.bn')
conv2 = ConvNorm(input=conv1,
num_filters=out_c,
filter_size=1,
norm_type='bn',
act=act,
initializer=Xavier(),
name=name + '.pointwise_linear',
norm_name=name + '.pointwise_linear.bn')
conv3 = ConvNorm(input=conv2,
num_filters=out_c,
filter_size=1,
norm_type='bn',
act='relu6',
initializer=Xavier(),
name=name + '.pointwise',
norm_name=name + '.pointwise.bn')
conv4 = ConvNorm(input=conv3,
num_filters=out_c,
filter_size=5,
groups=out_c,
norm_type='bn',
act=act,
initializer=Xavier(),
name=name + '.depthwise_linear',
norm_name=name + '.depthwise_linear.bn')
return conv4
def _deconv_upsample(self, x, out_c, name=None):
conv1 = ConvNorm(input=x,
num_filters=out_c,
filter_size=1,
norm_type='bn',
act='relu6',
name=name + '.pointwise',
initializer=Xavier(),
norm_name=name + '.pointwise.bn')
conv2 = fluid.layers.conv2d_transpose(
input=conv1,
num_filters=out_c,
filter_size=4,
padding=1,
stride=2,
groups=out_c,
param_attr=ParamAttr(name=name + '.deconv.weights',
initializer=Xavier()),
bias_attr=False)
bn = fluid.layers.batch_norm(
input=conv2,
act='relu6',
param_attr=ParamAttr(name=name + '.deconv.bn.scale',
regularizer=L2Decay(0.)),
bias_attr=ParamAttr(name=name + '.deconv.bn.offset',
regularizer=L2Decay(0.)),
moving_mean_name=name + '.deconv.bn.mean',
moving_variance_name=name + '.deconv.bn.variance')
conv3 = ConvNorm(input=bn,
num_filters=out_c,
filter_size=1,
norm_type='bn',
act='relu6',
name=name + '.normal',
initializer=Xavier(),
norm_name=name + '.normal.bn')
return conv3
def _mask_conv_head(self, roi_feat, num_convs, norm_type):
if norm_type == 'gn':
for i in range(num_convs):
layer_name = "mask_inter_feat_" + str(i + 1)
fan = self.conv_dim * 3 * 3
initializer = MSRA(uniform=False, fan_in=fan)
roi_feat = ConvNorm(roi_feat,
self.conv_dim,
3,
act='relu',
dilation=self.dilation,
initializer=initializer,
norm_type=self.norm_type,
name=layer_name,
norm_name=layer_name)
else:
for i in range(num_convs):
layer_name = "mask_inter_feat_" + str(i + 1)
fan = self.conv_dim * 3 * 3
initializer = MSRA(uniform=False, fan_in=fan)
roi_feat = fluid.layers.conv2d(
input=roi_feat,
num_filters=self.conv_dim,
filter_size=3,
padding=1 * self.dilation,
act='relu',
stride=1,
dilation=self.dilation,
name=layer_name,
param_attr=ParamAttr(name=layer_name + '_w',
initializer=initializer),
bias_attr=ParamAttr(name=layer_name + '_b',
learning_rate=2.,
regularizer=L2Decay(0.)))
fan = roi_feat.shape[1] * 2 * 2
feat = fluid.layers.conv2d_transpose(
input=roi_feat,
num_filters=self.conv_dim,
filter_size=2,
stride=2,
act='relu',
param_attr=ParamAttr(name='conv5_mask_w',
initializer=MSRA(uniform=False, fan_in=fan)),
bias_attr=ParamAttr(name='conv5_mask_b',
learning_rate=2.,
regularizer=L2Decay(0.)))
# print(feat)
return feat
def _mask_conv_head(self,
roi_feat,
num_convs,
norm_type,
wb_scalar=1.0,
name=''):
if norm_type == 'gn':
for i in range(num_convs):
layer_name = "mask_inter_feat_" + str(i + 1)
if not self.share_mask_conv:
layer_name += name
fan = self.conv_dim * 3 * 3
initializer = MSRA(uniform=False, fan_in=fan)
roi_feat = ConvNorm(roi_feat,
self.conv_dim,
3,
act='relu',
dilation=self.dilation,
initializer=initializer,
norm_type=self.norm_type,
name=layer_name,
norm_name=layer_name)
else:
for i in range(num_convs):
layer_name = "mask_inter_feat_" + str(i + 1)
if not self.share_mask_conv:
layer_name += name
fan = self.conv_dim * 3 * 3
initializer = MSRA(uniform=False, fan_in=fan)
roi_feat = fluid.layers.conv2d(
input=roi_feat,
num_filters=self.conv_dim,
filter_size=3,
padding=1 * self.dilation,
act='relu',
stride=1,
dilation=self.dilation,
name=layer_name,
param_attr=ParamAttr(name=layer_name + '_w',
initializer=initializer),
bias_attr=ParamAttr(name=layer_name + '_b',
learning_rate=wb_scalar *
self.lr_ratio,
regularizer=L2Decay(0.)))
return roi_feat
def get_output(self, body_dict):
"""
Add FPN onto backbone.
Args:
body_dict(OrderedDict): Dictionary of variables and each element is the
output of backbone.
Return:
fpn_dict(OrderedDict): A dictionary represents the output of FPN with
their name.
spatial_scale(list): A list of multiplicative spatial scale factor.
"""
body_name_list = list(body_dict.keys())[::-1]
num_backbone_stages = len(body_name_list)
self.fpn_inner_output = [[] for _ in range(num_backbone_stages)]
fpn_inner_name = 'fpn_inner_' + body_name_list[0]
body_input = body_dict[body_name_list[0]]
fan = body_input.shape[1]
if self.norm_type:
initializer = Xavier(fan_out=fan)
self.fpn_inner_output[0] = ConvNorm(
body_input,
self.num_chan,
1,
initializer=initializer,
norm_type=self.norm_type,
name=fpn_inner_name,
bn_name=fpn_inner_name)
else:
self.fpn_inner_output[0] = fluid.layers.conv2d(
body_input,
self.num_chan,
1,
param_attr=ParamAttr(
name=fpn_inner_name + "_w",
initializer=Xavier(fan_out=fan)),
bias_attr=ParamAttr(
name=fpn_inner_name + "_b",
learning_rate=2.,
regularizer=L2Decay(0.)),
name=fpn_inner_name)
for i in range(1, num_backbone_stages):
body_name = body_name_list[i]
body_input = body_dict[body_name]
top_output = self.fpn_inner_output[i - 1]
fpn_inner_single = self._add_topdown_lateral(body_name, body_input,
top_output)
self.fpn_inner_output[i] = fpn_inner_single
fpn_dict = {}
fpn_name_list = []
for i in range(num_backbone_stages):
fpn_name = 'fpn_' + body_name_list[i]
fan = self.fpn_inner_output[i].shape[1] * 3 * 3
if self.norm_type:
initializer = Xavier(fan_out=fan)
fpn_output = ConvNorm(
self.fpn_inner_output[i],
self.num_chan,
3,
initializer=initializer,
norm_type=self.norm_type,
name=fpn_name,
bn_name=fpn_name)
else:
fpn_output = fluid.layers.conv2d(
self.fpn_inner_output[i],
self.num_chan,
filter_size=3,
padding=1,
param_attr=ParamAttr(
name=fpn_name + "_w", initializer=Xavier(fan_out=fan)),
bias_attr=ParamAttr(
name=fpn_name + "_b",
learning_rate=2.,
regularizer=L2Decay(0.)),
name=fpn_name)
fpn_dict[fpn_name] = fpn_output
fpn_name_list.append(fpn_name)
if not self.has_extra_convs and self.max_level - self.min_level == len(
self.spatial_scale):
body_top_name = fpn_name_list[0]
body_top_extension = fluid.layers.pool2d(
fpn_dict[body_top_name],
1,
'max',
pool_stride=2,
name=body_top_name + '_subsampled_2x')
fpn_dict[body_top_name + '_subsampled_2x'] = body_top_extension
fpn_name_list.insert(0, body_top_name + '_subsampled_2x')
self.spatial_scale.insert(0, self.spatial_scale[0] * 0.5)
# Coarser FPN levels introduced for RetinaNet
highest_backbone_level = self.min_level + len(self.spatial_scale) - 1
if self.has_extra_convs and self.max_level > highest_backbone_level:
fpn_blob = body_dict[body_name_list[0]]
for i in range(highest_backbone_level + 1, self.max_level + 1):
fpn_blob_in = fpn_blob
fpn_name = 'fpn_' + str(i)
if i > highest_backbone_level + 1:
fpn_blob_in = fluid.layers.relu(fpn_blob)
fan = fpn_blob_in.shape[1] * 3 * 3
fpn_blob = fluid.layers.conv2d(
input=fpn_blob_in,
num_filters=self.num_chan,
filter_size=3,
stride=2,
padding=1,
param_attr=ParamAttr(
name=fpn_name + "_w", initializer=Xavier(fan_out=fan)),
bias_attr=ParamAttr(
name=fpn_name + "_b",
learning_rate=2.,
regularizer=L2Decay(0.)),
name=fpn_name)
fpn_dict[fpn_name] = fpn_blob
fpn_name_list.insert(0, fpn_name)
self.spatial_scale.insert(0, self.spatial_scale[0] * 0.5)
res_dict = OrderedDict([(k, fpn_dict[k]) for k in fpn_name_list])
return res_dict, self.spatial_scale
def dense_aspp(self, input, name=None):
dropout0 = 0.1
d_feature0 = 512
d_feature1 = 256
aspp3 = self.dense_aspp_block(input,
num_filters1=d_feature0,
num_filters2=d_feature1,
dropout_prob=dropout0,
name=name + '_aspp3',
dilation_rate=3)
conv = fluid.layers.concat([aspp3, input], axis=1)
aspp6 = self.dense_aspp_block(conv,
num_filters1=d_feature0,
num_filters2=d_feature1,
dropout_prob=dropout0,
name=name + '_aspp6',
dilation_rate=6)
conv = fluid.layers.concat([aspp6, conv], axis=1)
aspp12 = self.dense_aspp_block(conv,
num_filters1=d_feature0,
num_filters2=d_feature1,
dropout_prob=dropout0,
name=name + '_aspp12',
dilation_rate=12)
conv = fluid.layers.concat([aspp12, conv], axis=1)
aspp18 = self.dense_aspp_block(conv,
num_filters1=d_feature0,
num_filters2=d_feature1,
dropout_prob=dropout0,
name=name + '_aspp18',
dilation_rate=18)
conv = fluid.layers.concat([aspp18, conv], axis=1)
aspp24 = self.dense_aspp_block(conv,
num_filters1=d_feature0,
num_filters2=d_feature1,
dropout_prob=dropout0,
name=name + '_aspp24',
dilation_rate=24)
conv = fluid.layers.concat([aspp3, aspp6, aspp12, aspp18, aspp24],
axis=1)
conv = ConvNorm(conv,
num_filters=self.num_chan,
filter_size=1,
stride=1,
groups=1,
norm_decay=0.,
norm_type='gn',
norm_groups=self.norm_groups,
dilation=1,
lr_scale=1,
freeze_norm=False,
act="relu",
norm_name=name + "_dense_aspp_reduce_gn",
initializer=None,
bias_attr=False,
name=name + "_dense_aspp_reduce_gn")
return conv
def _fcos_head(self, features, fpn_stride, fpn_scale, is_training=False):
"""
Args:
features (Variables): feature map from FPN
fpn_stride (int): the stride of current feature map
is_training (bool): whether is train or test mode
"""
subnet_blob_cls = features
subnet_blob_reg = features
in_channles = features.shape[1]
for lvl in range(0, self.num_convs):
conv_cls_name = 'fcos_head_cls_tower_conv_{}'.format(lvl)
subnet_blob_cls = ConvNorm(
input=subnet_blob_cls,
num_filters=in_channles,
filter_size=3,
stride=1,
norm_type=self.norm_type,
act='relu',
initializer=Normal(
loc=0., scale=0.01),
bias_attr=True,
norm_name=conv_cls_name + "_norm",
name=conv_cls_name)
conv_reg_name = 'fcos_head_reg_tower_conv_{}'.format(lvl)
subnet_blob_reg = ConvNorm(
input=subnet_blob_reg,
num_filters=in_channles,
filter_size=3,
stride=1,
norm_type=self.norm_type,
act='relu',
initializer=Normal(
loc=0., scale=0.01),
bias_attr=True,
norm_name=conv_reg_name + "_norm",
name=conv_reg_name)
conv_cls_name = "fcos_head_cls"
bias_init_value = -math.log((1 - self.prior_prob) / self.prior_prob)
cls_logits = fluid.layers.conv2d(
input=subnet_blob_cls,
num_filters=self.num_classes,
filter_size=3,
stride=1,
padding=1,
param_attr=ParamAttr(
name=conv_cls_name + "_weights",
initializer=Normal(
loc=0., scale=0.01)),
bias_attr=ParamAttr(
name=conv_cls_name + "_bias",
initializer=Constant(value=bias_init_value)),
name=conv_cls_name)
conv_reg_name = "fcos_head_reg"
bbox_reg = fluid.layers.conv2d(
input=subnet_blob_reg,
num_filters=4,
filter_size=3,
stride=1,
padding=1,
param_attr=ParamAttr(
name=conv_reg_name + "_weights",
initializer=Normal(
loc=0., scale=0.01)),
bias_attr=ParamAttr(
name=conv_reg_name + "_bias", initializer=Constant(value=0)),
name=conv_reg_name)
bbox_reg = bbox_reg * fpn_scale
if self.norm_reg_targets:
bbox_reg = fluid.layers.relu(bbox_reg)
if not is_training:
bbox_reg = bbox_reg * fpn_stride
else:
bbox_reg = fluid.layers.exp(bbox_reg)
conv_centerness_name = "fcos_head_centerness"
if self.centerness_on_reg:
subnet_blob_ctn = subnet_blob_reg
else:
subnet_blob_ctn = subnet_blob_cls
centerness = fluid.layers.conv2d(
input=subnet_blob_ctn,
num_filters=1,
filter_size=3,
stride=1,
padding=1,
param_attr=ParamAttr(
name=conv_centerness_name + "_weights",
initializer=Normal(
loc=0., scale=0.01)),
bias_attr=ParamAttr(
name=conv_centerness_name + "_bias",
initializer=Constant(value=0)),
name=conv_centerness_name)
return cls_logits, bbox_reg, centerness