def __init__(self, num_classes, context_path):
super().__init__()
# build spatial path
self.saptial_path = Spatial_path()
# build context path
self.context_path = build_contextpath(name=context_path)
# build attention refinement module for resnet 101
if context_path == 'resnet101':
self.attention_refinement_module1 = AttentionRefinementModule(1024, 1024)
self.attention_refinement_module2 = AttentionRefinementModule(2048, 2048)
# supervision block
self.supervision1 = nn.Conv2d(in_channels=1024, out_channels=num_classes, kernel_size=1)
self.supervision2 = nn.Conv2d(in_channels=2048, out_channels=num_classes, kernel_size=1)
# build feature fusion module
self.feature_fusion_module = FeatureFusionModule(num_classes, 3328)
elif context_path == 'resnet18':
# build attention refinement module for resnet 18
self.attention_refinement_module1 = AttentionRefinementModule(256, 256)
self.attention_refinement_module2 = AttentionRefinementModule(512, 512)
# supervision block
self.supervision1 = nn.Conv2d(in_channels=256, out_channels=num_classes, kernel_size=1)
self.supervision2 = nn.Conv2d(in_channels=512, out_channels=num_classes, kernel_size=1)
# build feature fusion module
self.feature_fusion_module = FeatureFusionModule(num_classes, 1024)
else:
print('Error: unspport context_path network \n')
# build final convolution
self.conv = nn.Conv2d(in_channels=num_classes, out_channels=num_classes, kernel_size=1)
def __init__(self, num_classes, num_char, context_path):
super().__init__()
# build spatial path
self.saptial_path = Spatial_path()
# build context path
self.context_path = build_contextpath(name=context_path)
# build attention refinement module
if context_path == 'resnet101':
self.attention_refinement_module1 = AttentionRefinementModule(
1024, 1024)
self.attention_refinement_module2 = AttentionRefinementModule(
2048, 2048)
self.feature_fusion_module1 = FeatureFusionModule(
num_classes, 3328)
self.feature_fusion_module2 = FeatureFusionModule(num_char, 3328)
else:
self.attention_refinement_module1 = AttentionRefinementModule(
256, 256)
self.attention_refinement_module2 = AttentionRefinementModule(
512, 512)
self.feature_fusion_module1 = FeatureFusionModule(
num_classes, 1024)
self.feature_fusion_module2 = FeatureFusionModule(num_char, 1024)
# build final convolution
self.conv_seg = nn.Conv2d(in_channels=num_classes,
out_channels=num_classes,
kernel_size=1)
self.conv_pos = nn.Conv2d(in_channels=num_char,
out_channels=num_char,
kernel_size=1)
def __init__(self, num_classes, context_path):
super().__init__()
# build spatial path
self.saptial_path = Spatial_path()
# build context path
self.context_path = build_contextpath(name=context_path)
# build attention refinement module
self.attention_refinement_module1 = AttentionRefinementModule(1024, 1024)
self.attention_refinement_module2 = AttentionRefinementModule(2048, 2048)
# build feature fusion module
self.feature_fusion_module = FeatureFusionModule(num_classes)
# build final convolution
self.conv = nn.Conv2d(in_channels=num_classes, out_channels=num_classes, kernel_size=1)
# supervision block
self.supervision1 = nn.Conv2d(in_channels=1024, out_channels=num_classes, kernel_size=1)
self.supervision2 = nn.Conv2d(in_channels=2048, out_channels=num_classes, kernel_size=1)
def __init__(self, num_classes, context_path, in_channel=3):
super().__init__()
# build context path
self.context_path = build_contextpath(name=context_path, in_channel=in_channel)
# build attention refinement module
if context_path in ['resnet18', 'resnet34']:
self.attention_refinement_module0 = AttentionRefinementModule(128, 128)
self.attention_refinement_module1 = AttentionRefinementModule(256, 256)
self.attention_refinement_module2 = AttentionRefinementModule(512, 512)
n_ch = 512
elif context_path in ['xception']:
self.attention_refinement_module0 = AttentionRefinementModule(256, 256)
self.attention_refinement_module1 = AttentionRefinementModule(728, 728)
self.attention_refinement_module2 = AttentionRefinementModule(1024, 1024)
n_ch = 1024
else:
self.attention_refinement_module0 = AttentionRefinementModule(512, 512)
self.attention_refinement_module1 = AttentionRefinementModule(1024, 1024)
self.attention_refinement_module2 = AttentionRefinementModule(2048, 2048)
n_ch = 2048
# build feature fusion module
self.feature_fusion_module = FeatureFusionModule(num_classes, context_path)
# build final convolution
self.conv = nn.Conv2d(in_channels=num_classes, out_channels=num_classes, kernel_size=1)
self.cls = nn.Linear(in_features=n_ch, out_features=num_classes-1)
# supervision block
if context_path in ['resnet18', 'resnet34']:
self.supervision1 = nn.Conv2d(in_channels=256, out_channels=num_classes, kernel_size=1)
self.supervision2 = nn.Conv2d(in_channels=512, out_channels=num_classes, kernel_size=1)
elif context_path in ['xception']:
self.supervision1 = nn.Conv2d(in_channels=728, out_channels=num_classes, kernel_size=1)
self.supervision2 = nn.Conv2d(in_channels=1024, out_channels=num_classes, kernel_size=1)
else:
self.supervision1 = nn.Conv2d(in_channels=1024, out_channels=num_classes, kernel_size=1)
self.supervision2 = nn.Conv2d(in_channels=2048, out_channels=num_classes, kernel_size=1)
def __init__(self, num_classes, num_char, context_path):
super().__init__()
# build spatial path
self.saptial_path = Spatial_path()
# build context path
self.context_path = build_contextpath(name=context_path)
# build attention refinement module
if context_path == 'resnet101':
self.attention_refinement_module1 = AttentionRefinementModule(
1024, 1024)
self.attention_refinement_module2 = AttentionRefinementModule(
2048, 2048)
self.feature_fusion_module1 = FeatureFusionModule(
num_classes, 3328)
self.feature_fusion_module2 = FeatureFusionModule(num_char, 3328)
else:
self.attention_refinement_module1 = AttentionRefinementModule(
256, 256)
self.attention_refinement_module2 = AttentionRefinementModule(
512, 512)
self.feature_fusion_module1 = FeatureFusionModule(
num_classes, 1024)
self.feature_fusion_module2 = FeatureFusionModule(num_char, 1024)
# build final convolution
self.conv_seg = nn.Conv2d(in_channels=num_classes,
out_channels=num_classes,
kernel_size=1)
self.conv_pos = nn.Conv2d(in_channels=num_char,
out_channels=num_char,
kernel_size=1)
self.convblock = nn.Sequential(
ConvBlock(in_channels=num_classes - 1,
out_channels=1000,
kernel_size=5,
stride=4,
padding=2),
nn.AdaptiveAvgPool2d((1, 1)),
)
# self.convblock1 = nn.Sequential(
# ConvBlock(in_channels=num_classes-1, out_channels=256, kernel_size=5, stride=4, padding=2),
# ConvBlock(in_channels=256, out_channels=1000, kernel_size=5, stride=4, padding=2),
# nn.AdaptiveAvgPool2d((1, 1)),
# )
# self.convblock2 = nn.Sequential(
# ConvBlock(in_channels=num_classes-42, out_channels=1000, kernel_size=5, stride=4, padding=2),
# nn.AdaptiveAvgPool2d((1, 1)),
# )
# self.convblock3 = nn.Sequential(
# ConvBlock(in_channels=num_classes-32, out_channels=1000, kernel_size=5, stride=4, padding=2),
# nn.AdaptiveAvgPool2d((1, 1)),
# )
self.classifier1 = nn.Linear(1000, provNum)
self.classifier2 = nn.Linear(1000, alphaNum)
self.classifier3 = nn.Linear(1000, adNum)
self.classifier4 = nn.Linear(1000, adNum)
self.classifier5 = nn.Linear(1000, adNum)
self.classifier6 = nn.Linear(1000, adNum)
self.classifier7 = nn.Linear(1000, adNum)