def __init__(self,
in_planes,
out_planes,
scale,
is_aux=False,
norm_layer=nn.BatchNorm2d,
real_time=0):
super(BiSeNetHead, self).__init__()
self.real_time = real_time
if self.real_time:
feature_num = 128
else:
feature_num = 256
if is_aux:
self.conv_3x3 = ConvBnRelu(in_planes,
feature_num,
3,
1,
1,
has_bn=True,
norm_layer=norm_layer,
has_relu=True,
has_bias=False)
else:
self.conv_3x3 = ConvBnRelu(in_planes,
64,
3,
1,
1,
has_bn=True,
norm_layer=norm_layer,
has_relu=True,
has_bias=False)
# self.dropout = nn.Dropout(0.1)
if is_aux:
self.conv_1x1 = nn.Conv2d(feature_num,
out_planes,
kernel_size=1,
stride=1,
padding=0)
else:
self.conv_1x1 = nn.Conv2d(64,
out_planes,
kernel_size=1,
stride=1,
padding=0)
self.scale = scale
def __init__(self, in_channels, out_channels, kernel_size=1, stride=1,
padding=0, dilation=1,
has_relu=True, norm_layer=nn.BatchNorm2d):
super(SeparableConvBnRelu, self).__init__()
self.conv1 = nn.Conv2d(in_channels, in_channels, kernel_size, stride,
padding, dilation, groups=in_channels,
bias=False)
self.point_wise_cbr = ConvBnRelu(in_channels, out_channels, 1, 1, 0,
has_bn=True, norm_layer=norm_layer,
has_relu=has_relu, has_bias=False)
def __init__(self,
in_planes,
out_planes,
scale,
is_aux=False,
norm_layer=nn.BatchNorm2d):
super(BiSeNetHead, self).__init__()
if is_aux:
self.conv_3x3 = ConvBnRelu(in_planes,
256,
3,
1,
1,
has_bn=True,
norm_layer=norm_layer,
has_relu=True,
has_bias=False)
else:
self.conv_3x3 = ConvBnRelu(in_planes,
256,
3,
1,
1,
has_bn=True,
norm_layer=norm_layer,
has_relu=True,
has_bias=False)
# self.dropout = nn.Dropout(0.1)
if is_aux:
self.conv_1x1 = nn.Conv2d(256,
out_planes,
kernel_size=1,
stride=1,
padding=0)
else:
self.conv_1x1 = nn.Conv2d(256,
out_planes,
kernel_size=1,
stride=1,
padding=0)
self.scale = scale
def __init__(self, in_planes, out_planes, norm_layer=nn.BatchNorm2d):
super(SpatialPath, self).__init__()
inner_channel = 64
self.conv_7x7 = ConvBnRelu(in_planes,
inner_channel,
7,
2,
3,
has_bn=True,
norm_layer=norm_layer,
has_relu=True,
has_bias=False)
self.conv_3x3_1 = ConvBnRelu(inner_channel,
inner_channel,
3,
2,
1,
has_bn=True,
norm_layer=norm_layer,
has_relu=True,
has_bias=False)
self.conv_3x3_2 = ConvBnRelu(inner_channel,
inner_channel,
3,
2,
1,
has_bn=True,
norm_layer=norm_layer,
has_relu=True,
has_bias=False)
self.conv_1x1 = ConvBnRelu(inner_channel,
out_planes,
1,
1,
0,
has_bn=True,
norm_layer=norm_layer,
has_relu=True,
has_bias=False)
def __init__(self, block, layers, channels, norm_layer=nn.BatchNorm2d):
super(Xception, self).__init__()
self.in_channels = 8
self.conv1 = ConvBnRelu(3, self.in_channels, 3, 2, 1,
has_bn=True, norm_layer=norm_layer,
has_relu=True, has_bias=False)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(block, norm_layer,
layers[0], channels[0], stride=2)
self.layer2 = self._make_layer(block, norm_layer,
layers[1], channels[1], stride=2)
self.layer3 = self._make_layer(block, norm_layer,
layers[2], channels[2], stride=2)
def __init__(self,
out_planes,
is_training,
criterion,
ohem_criterion,
pretrained_model=None,
norm_layer=nn.BatchNorm2d,
in_planes=3,
real_time=0):
super(BiSeNet, self).__init__()
self.real_time = real_time
self.context_path = resnet18(pretrained_model,
norm_layer=norm_layer,
deep_stem=False,
stem_width=64,
in_planes=in_planes)
self.business_layer = []
self.is_training = is_training
self.spatial_path = SpatialPath(in_planes, 128, norm_layer)
conv_channel = 128
self.global_context = nn.Sequential(
nn.AdaptiveAvgPool2d(1),
ConvBnRelu(512,
conv_channel,
1,
1,
0,
has_bn=True,
has_relu=True,
has_bias=False,
norm_layer=norm_layer))
# stage = [512, 256, 128, 64]
arms = [
AttentionRefinement(512, conv_channel, norm_layer),
AttentionRefinement(256, conv_channel, norm_layer)
]
refines = [
ConvBnRelu(conv_channel,
conv_channel,
3,
1,
1,
has_bn=True,
norm_layer=norm_layer,
has_relu=True,
has_bias=False),
ConvBnRelu(conv_channel,
conv_channel,
3,
1,
1,
has_bn=True,
norm_layer=norm_layer,
has_relu=True,
has_bias=False)
]
if self.real_time:
if is_training:
heads = [
BiSeNetHead(conv_channel,
out_planes,
2,
True,
norm_layer,
real_time=real_time),
BiSeNetHead(conv_channel,
out_planes,
1,
True,
norm_layer,
real_time=real_time),
BiSeNetHead(conv_channel * 2,
out_planes,
1,
False,
norm_layer,
real_time=real_time)
]
else:
heads = [
None, None,
BiSeNetHead(conv_channel * 2,
out_planes,
1,
False,
norm_layer,
real_time=real_time)
]
else:
heads = [
BiSeNetHead(conv_channel,
out_planes,
16,
True,
norm_layer,
real_time=real_time),
BiSeNetHead(conv_channel,
out_planes,
8,
True,
norm_layer,
real_time=real_time),
BiSeNetHead(conv_channel * 2,
out_planes,
1,
False,
norm_layer,
real_time=real_time)
]
self.ffm = FeatureFusion(conv_channel * 2, conv_channel * 2, 1,
norm_layer)
self.arms = nn.ModuleList(arms)
self.refines = nn.ModuleList(refines)
self.heads = nn.ModuleList(heads)
self.business_layer.append(self.spatial_path)
self.business_layer.append(self.global_context)
self.business_layer.append(self.arms)
self.business_layer.append(self.refines)
self.business_layer.append(self.heads)
self.business_layer.append(self.ffm)
if is_training:
self.criterion = criterion
self.ohem_criterion = ohem_criterion
def __init__(self,
out_planes,
is_training,
criterion,
pretrained_model=None,
norm_layer=nn.BatchNorm2d):
super(BiSeNet, self).__init__()
self.context_path = resnet101(pretrained_model,
norm_layer=norm_layer,
bn_eps=config.bn_eps,
bn_momentum=config.bn_momentum,
deep_stem=True,
stem_width=64)
self.business_layer = []
self.is_training = is_training
self.spatial_path = SpatialPath(3, 128, norm_layer)
conv_channel = 128
self.global_context = nn.Sequential(
nn.AdaptiveAvgPool2d(1),
ConvBnRelu(2048,
conv_channel,
1,
1,
0,
has_bn=True,
has_relu=True,
has_bias=False,
norm_layer=norm_layer))
# stage = [512, 256, 128, 64]
arms = [
AttentionRefinement(2048, conv_channel, norm_layer),
AttentionRefinement(1024, conv_channel, norm_layer)
]
refines = [
ConvBnRelu(conv_channel,
conv_channel,
3,
1,
1,
has_bn=True,
norm_layer=norm_layer,
has_relu=True,
has_bias=False),
ConvBnRelu(conv_channel,
conv_channel,
3,
1,
1,
has_bn=True,
norm_layer=norm_layer,
has_relu=True,
has_bias=False)
]
heads = [
BiSeNetHead(conv_channel, out_planes, 16, True, norm_layer),
BiSeNetHead(conv_channel, out_planes, 8, True, norm_layer),
BiSeNetHead(conv_channel * 2, out_planes, 8, False, norm_layer)
]
self.ffm = FeatureFusion(conv_channel * 2, conv_channel * 2, 1,
norm_layer)
self.arms = nn.ModuleList(arms)
self.refines = nn.ModuleList(refines)
self.heads = nn.ModuleList(heads)
self.business_layer.append(self.spatial_path)
self.business_layer.append(self.global_context)
self.business_layer.append(self.arms)
self.business_layer.append(self.refines)
self.business_layer.append(self.heads)
self.business_layer.append(self.ffm)
if is_training:
self.criterion = criterion