def constructor(self,
inChannels,
growRate,
sh_groups,
conv_groups,
kSize=3):
"""Create Block_Base Block.
:param inChannels: channel number of input
:type inChannels: int
:param growRate: growth rate of block
:type growRate: int
:param sh_groups: group number of shuffle operation
:type sh_groups: int
:param conv_groups: group number of convolution operation
:type conv_groups: int
:param kSize: kernel size of convolution operation
:type kSize: int
"""
self.channel_shuffle = ChannelShuffle(sh_groups)
self.conv = op.Conv2d(in_channels=inChannels,
out_channels=growRate,
kernel_size=kSize,
stride=1,
padding=(kSize - 1) // 2,
groups=conv_groups)
self.relu = op.ReLU()
def __init__(self,
inchannel,
outchannel,
expansion,
groups,
base_width,
stride=1):
"""Create BottleConv layer.
:param inchannel: input channel.
:type inchannel: int
:param outchannel: output channel.
:type outchannel: int
:param expansion: expansion
:type expansion: int
:param stride: the number to jump, default 1
:type stride: int
"""
super(BottleConv, self).__init__(inchannel,
outchannel,
groups,
base_width,
stride=1)
outchannel = int(outchannel * (base_width / 64.)) * groups
self.conv1 = conv1X1(inchannel=inchannel, outchannel=outchannel)
self.batch1 = op.BatchNorm2d(num_features=outchannel)
self.conv2 = conv3x3(inchannel=outchannel,
outchannel=outchannel,
groups=groups,
stride=stride)
self.batch2 = op.BatchNorm2d(num_features=outchannel)
self.conv3 = conv1X1(inchannel=outchannel,
outchannel=outchannel * expansion)
self.batch3 = op.BatchNorm2d(num_features=outchannel * expansion)
self.relu = op.ReLU()
def __init__(self,
inchannel,
outchannel,
groups=1,
base_width=64,
stride=1,
inner_plane=None):
"""Create BasicBlock layers.
:param inchannel: input channel.
:type inchannel: int
:param outchannel: output channel.
:type outchannel: int
:param stride: the number to jump, default 1
:type stride: int
"""
super(BasicBlock, self).__init__(inchannel, outchannel, groups,
base_width, stride, inner_plane)
base_conv = BasicConv(inchannel=inchannel,
outchannel=outchannel,
stride=stride,
groups=groups,
base_width=base_width,
inner_plane=inner_plane)
shortcut = ShortCut(inchannel=inchannel,
outchannel=outchannel,
expansion=self.expansion,
stride=stride)
self.block = Add(base_conv, shortcut)
self.relu = op.ReLU()
def constructor(self, in_channels=256, feat_channels=256, num_classes=2):
"""Create rpn Search Space."""
anchor_scales = [8, 16, 32]
anchor_ratios = [0.5, 1.0, 2.0]
num_anchors = len(anchor_ratios) * len(anchor_scales)
if feat_channels > 0:
conv = op.Conv2d(in_channels=in_channels,
out_channels=feat_channels,
kernel_size=3,
padding=1)
relu = op.ReLU(inplace=True)
rpn_cls_conv = op.Conv2d(in_channels=feat_channels,
out_channels=num_anchors * num_classes,
kernel_size=1)
rpn_reg_conv = op.Conv2d(in_channels=feat_channels,
out_channels=num_anchors * 4,
kernel_size=1)
rpn_cls = Sequential(conv, relu, rpn_cls_conv)
rpn_reg = Sequential(conv, relu, rpn_reg_conv)
else:
rpn_cls = op.Conv2d(in_channels=in_channels,
out_channels=num_anchors * num_classes,
kernel_size=1)
rpn_reg = op.Conv2d(in_channels=in_channels,
out_channels=num_anchors * 4,
kernel_size=1)
self.rpn = Tuple(Map(rpn_cls), Map(rpn_reg))
def __init__(self, init_plane):
"""Create SmallInputInitialBlock layer.
:param init_plane: input channel.
:type init_plane: int
"""
super(SmallInputInitialBlock, self).__init__(init_plane)
self.conv = conv3x3(inchannel=3, outchannel=init_plane, stride=1)
self.batch = op.BatchNorm2d(num_features=init_plane)
self.relu = op.ReLU()
def __init__(self,
inchannel,
outchannel,
groups=1,
base_width=64,
stride=1,
inner_plane=None):
"""Create BasicConv layer.
:param inchannel: input channel.
:type inchannel: int
:param outchannel: output channel.
:type outchannel: int
:param stride: the number to jump, default 1
:type stride: int
"""
super(BasicConv, self).__init__(inchannel, outchannel, groups,
base_width, stride, inner_plane)
if inner_plane is None:
if groups != 1 or base_width != 64:
raise ValueError(
"BasicBlock only supports groups=1 and base_width=64")
self.conv = conv3x3(inchannel=inchannel,
outchannel=outchannel,
stride=stride)
self.batch = op.BatchNorm2d(num_features=outchannel)
self.relu = op.ReLU(inplace=True)
self.conv2 = conv3x3(inchannel=outchannel, outchannel=outchannel)
self.batch2 = op.BatchNorm2d(num_features=outchannel)
else:
self.conv = conv3x3(inchannel=inchannel,
outchannel=inner_plane,
stride=stride)
self.batch = op.BatchNorm2d(num_features=inner_plane)
self.relu = op.ReLU(inplace=True)
self.conv2 = conv3x3(inchannel=inner_plane, outchannel=outchannel)
self.batch2 = op.BatchNorm2d(num_features=outchannel)
def __init__(self, init_plane):
"""Create InitialBlock layer.
:param init_plane: input channel.
:type init_plane: int
"""
super(InitialBlock, self).__init__(init_plane)
self.conv = op.Conv2d(in_channels=3,
out_channels=init_plane,
kernel_size=7,
stride=2,
padding=3,
bias=False)
self.batch = op.BatchNorm2d(num_features=init_plane)
self.relu = op.ReLU()
self.maxpool2d = op.MaxPool2d(kernel_size=3, stride=2, padding=1)
def __init__(self, inchannel, outchannel, groups, base_width, stride=1):
"""Create BottleneckBlock layers.
:param inchannel: input channel.
:type inchannel: int
:param outchannel: output channel.
:type outchannel: int
:param stride: the number to jump, default 1
:type stride: int
"""
super(BottleneckBlock, self).__init__(inchannel, outchannel, groups,
base_width, stride)
bottle_conv = BottleConv(inchannel=inchannel,
outchannel=outchannel,
expansion=self.expansion,
stride=stride,
groups=groups,
base_width=base_width)
shortcut = ShortCut(inchannel=inchannel,
outchannel=outchannel,
expansion=self.expansion,
stride=stride)
self.block = Add(bottle_conv, shortcut)
self.relu = op.ReLU()