def __init__(self, n_classes):
super(HighOrder, self).__init__()
self.backbone = resnet(101, 16)
self.kernelrep = Kernel_Representation()
self.featurefusion = Feature_Fusion()
self.low_2x = lowpath()
self.low_4x = lowpath()
self.low_8x = lowpath()
self.conv_low1 = nn.Conv2d(384, 48, kernel_size=1, stride=1, padding=0)
self.bn1 = nn.BatchNorm2d(48)
self.conv_low2 = nn.Conv2d(640, 48, kernel_size=1, stride=1, padding=0)
self.bn2 = nn.BatchNorm2d(48)
self.conv_low3 = nn.Conv2d(1152,
48,
kernel_size=1,
stride=1,
padding=0)
self.bn3 = nn.BatchNorm2d(48)
self.conv_cat = nn.Sequential(ConvBNReLU(400, 256),
ConvBNReLU(256, 256))
self.conv_out = nn.Conv2d(256, n_classes, kernel_size=1, bias=False)
self.relu = nn.ReLU(inplace=True)
self.init_weight()
def __init__(self, classes):
super(PANet, self).__init__()
self.backbone = resnet(50, 16)
self.PAModule = PAmodule(classes)
self.DIGModule1 = DIGModule(3, 0, 1024)
self.DIGModule2 = DIGModule(2, 1, 512)
self.DIGModule3 = DIGModule(1, 1, 256)
# self.DIGModule4 = DIGModule(0, 1, 128)
self.feat1_conv = nn.Conv2d(256,
64,
kernel_size=1,
stride=1,
padding=0)
self.feat2_conv = nn.Conv2d(256,
64,
kernel_size=1,
stride=1,
padding=0)
self.feat3_conv = nn.Conv2d(256,
64,
kernel_size=1,
stride=1,
padding=0)
self.conv_cat = nn.Sequential(ConvBNReLU(448, 256),
ConvBNReLU(256, 256))
self.conv_out = nn.Conv2d(256, classes, kernel_size=1, bias=False)
self.init_weight()
def __init__(self):
super(Origin_Res, self).__init__()
self.resnet = resnet(101, 16)
# self.resnet = resnet101(pretrained=True)
self.conv1 = ConvBNReLU(2048, 256, 3, 1, 1, 1)
self.bn1 = nn.BatchNorm2d(256)
self.conv2 = ConvBNReLU(256, 19, 1, 1, 1, 1)
def __init__(self, classes):
super(PANet, self).__init__()
self.backbone = resnet(101, 16)
self.PAModule = PAmodule()
self.DIGModule1 = DIGModule(3, 0, 1024)
self.DIGModule2 = DIGModule(2, 1, 512)
self.DIGModule3 = DIGModule(1, 1, 256)
# self.DIGModule4 = DIGModule(0, 1, 128)
self.conv_out = nn.Conv2d(256, classes, kernel_size=1, bias=False)
def __init__(self):
super(Origin_Res, self).__init__()
self.resnet = resnet(101, 16)
self.low_2x = low_path_2x()
self.low_4x = low_path_4x()
self.low_8x = low_path_8x()
self.conv_8x = ConvBNReLU(3072, 256, 3, 1, 1, 1)
self.conv_4x = ConvBNReLU(1024, 256, 3, 1, 1, 1)
self.conv_2x = ConvBNReLU(512, 256, 3, 1, 1, 1)
self.conv_all = nn.Sequential(ConvBNReLU(768, 256, 3, 1, 1, 1),
ConvBNReLU(256, 256, 3, 1, 1, 1))
self.conv_out = nn.Conv2d(256, 19, kernel_size=1, bias=False)
def __init__(self, n_classes):
super(HighOrder, self).__init__()
self.backbone = resnet(101, 16)
self.featurefusion = Feature_Fusion()
self.conv_low = nn.Conv2d(256, 48, kernel_size=1, stride=1, padding=0)
self.bn1 = nn.BatchNorm2d(48)
self.conv_cat = nn.Sequential(ConvBNReLU(304, 256),
ConvBNReLU(256, 256))
self.conv_out = nn.Conv2d(256, n_classes, kernel_size=1, bias=False)
self.relu = nn.ReLU(inplace=True)
self.init_weight()
def __init__(self, classes):
super(PANet, self).__init__()
self.backbone = resnet(101, 16)
self.PAModule = PAmodule()
# self.DIGModule1 = DIGModule(3, 0, 1024)
# self.DIGModule2 = DIGModule(2, 1, 512)
# self.DIGModule3 = DIGModule(1, 1, 256)
# self.DIGModule4 = DIGModule(0, 1, 128)
self.conv_low = nn.Conv2d(256, 48, kernel_size=1, stride=1, padding=0)
self.bn1 = nn.BatchNorm2d(48)
self.conv_cat = nn.Sequential(ConvBNReLU(304, 256),
ConvBNReLU(256, 256))
self.conv_out = nn.Conv2d(256, classes, kernel_size=1, bias=False)
self.relu = nn.ReLU(inplace=True)
self.init_weight()
def __init__(self, classes):
super(PANet, self).__init__()
self.backbone = resnet(101, 16)
self.x4_low_conv = nn.Conv2d(2048,
256,
kernel_size=1,
stride=1,
padding=0)
self.PAModule = PAmodule()
self.DIGModule1 = DIGModule(3, 0, 1024)
self.DIGModule2 = DIGModule(2, 1, 512)
self.DIGModule3 = DIGModule(1, 1, 256)
# self.DIGModule4 = DIGModule(0, 1, 128)
self.low_2x = lowpath()
self.low_4x = lowpath()
self.low_8x = lowpath()
self.conv_low1 = nn.Conv2d(384, 48, kernel_size=1, stride=1, padding=0)
self.bn1 = nn.BatchNorm2d(48)
self.conv_low2 = nn.Conv2d(640, 48, kernel_size=1, stride=1, padding=0)
self.bn2 = nn.BatchNorm2d(48)
self.conv_low3 = nn.Conv2d(1152,
48,
kernel_size=1,
stride=1,
padding=0)
self.bn3 = nn.BatchNorm2d(48)
self.conv_cat = nn.Sequential(ConvBNReLU(400, 256),
ConvBNReLU(256, 256))
self.conv_out = nn.Conv2d(256, classes, kernel_size=1, bias=False)
self.relu = nn.ReLU(inplace=True)
self.init_weight()
def __init__(self, *args, **kwargs):
super(Deeplab_v3plus, self).__init__()
self.backbone = resnet(101, 16)
self.aspp = ASPP(in_chan=2048, out_chan=256, with_gp=False)
self.decoder = Decoder(config_CS.classes, low_chan=256)
