def __init__(self, in_channels, out_channels, norm_layer):
super(DANetHead, self).__init__()
inter_channels = in_channels // 4
self.conv5a = nn.Sequential(
nn.Conv2d(in_channels, inter_channels, 3, padding=1, bias=False),
norm_layer(inter_channels), nn.ReLU())
self.conv5c = nn.Sequential(
nn.Conv2d(in_channels, inter_channels, 3, padding=1, bias=False),
norm_layer(inter_channels), nn.ReLU())
self.sa = PAM_Module(inter_channels)
self.sc = CAM_Module(inter_channels)
self.conv51 = nn.Sequential(
nn.Conv2d(inter_channels, inter_channels, 3, padding=1,
bias=False), norm_layer(inter_channels), nn.ReLU())
self.conv52 = nn.Sequential(
nn.Conv2d(inter_channels, inter_channels, 3, padding=1,
bias=False), norm_layer(inter_channels), nn.ReLU())
self.conv6 = nn.Sequential(nn.Dropout2d(0.1, False),
nn.Conv2d(512, out_channels, 1))
self.conv7 = nn.Sequential(nn.Dropout2d(0.1, False),
nn.Conv2d(512, out_channels, 1))
self.conv8 = nn.Sequential(nn.Dropout2d(0.1, False),
nn.Conv2d(512, out_channels, 1))
def __init__(self, BAND_A, BAND_B, classes):
super(SAR_simple, self).__init__()
self.name = 'sar_simple_'
# sar
self.conv21 = nn.Sequential(
nn.Conv2d(BAND_B, 32, kernel_size=3, padding=0),
nn.BatchNorm2d(32, momentum=1, affine=True),
mish()
)
self.conv22 = nn.Sequential(
nn.Conv2d(32, 64, kernel_size=3, padding=1),
nn.BatchNorm2d(64, momentum=1, affine=True),
mish()
)
self.conv23 = nn.Sequential(
nn.Conv2d(64, 64, kernel_size=3, padding=1),
nn.BatchNorm2d(64, momentum=1, affine=True),
mish()
)
self.conv24 = nn.Sequential(
nn.Conv2d(64, 64, kernel_size=3, padding=1),
nn.BatchNorm2d(64, momentum=1, affine=True),
mish()
)
self.conv25 = nn.Sequential(
nn.Conv2d(64, 32, kernel_size=3, padding=1),
nn.BatchNorm2d(32, momentum=1, affine=True),
mish()
)
# fusion
self.batch_norm_optical = nn.Sequential(
nn.BatchNorm2d(32, momentum=1, affine=True),
mish(),
nn.Dropout(p=0.5)
)
self.batch_norm_sar = nn.Sequential(
nn.BatchNorm2d(32, momentum=1, affine=True),
mish(),
nn.Dropout(p=0.5)
)
self.global_pooling = nn.AdaptiveAvgPool2d(1)
self.fc = nn.Sequential(
nn.Linear(32, classes),
# nn.Softmax(dim=-1)
)
self.cam = CAM_Module(32)
self.pam = PAM_Module(32)
def __init__(self, input_channels=3, pretrained=True):
super(CAMUNet, self).__init__()
self.resnet = resnet50(pretrained=pretrained, input_channels=input_channels)
self.aspp = _ASPPModule(2048, 256)
self.conv = nn.Conv2d(256, 1, (1, 1), stride=1)
self.pdcmblock0 = PDCM(64)
self.pdcmblock1 = PDCM(256)
self.pdcmblock2 = PDCM(512)
self.pdcmblock3 = PDCM(1024)
self.pdcmblock4 = PDCM(256)
self.upsample_2_2 = nn.ConvTranspose2d(64, 64, 4, stride=2, bias=False)
self.upsample_2_3 = nn.ConvTranspose2d(64, 64, 4, stride=2, bias=False)
self.upsample_2_4 = nn.ConvTranspose2d(64, 64, 4, stride=2, bias=False)
self.upsample_2_5 = nn.ConvTranspose2d(32, 32, 4, stride=2, bias=False)
self.cam_conv1_1 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(128, 64, 3, stride=1, padding=1)),
('bn' , nn.BatchNorm2d(64)),
('relu', nn.PReLU(num_parameters=64))]))
self.cam1 = CAM_Module(64)
self.cam_conv1_2 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(64, 64, 3, stride=1, padding=1)),
('bn' , nn.BatchNorm2d(64)),
('relu', nn.PReLU(num_parameters=64))]))
self.cam_drop1 = nn.Dropout2d(0.1, False)
self.cam_conv1_3 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(64, 32, 1, stride=1)),
('bn' , nn.BatchNorm2d(32)),
('relu', nn.PReLU(num_parameters=32))]))
self.cam_down_c1 = nn.Conv2d(32, 1, 1, stride=1)
self.cam_conv2_1 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(320, 128, 3, stride=1, padding=1)),
('bn' , nn.BatchNorm2d(128)),
('relu', nn.PReLU(num_parameters=128))]))
self.cam2 = CAM_Module(128)
self.cam_conv2_2 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(128, 64, 3, stride=1, padding=1)),
('bn' , nn.BatchNorm2d(64)),
('relu', nn.PReLU(num_parameters=64))]))
self.cam_drop2 = nn.Dropout2d(0.1, False)
self.cam_conv2_3 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(64, 64, 1, stride=1)),
('bn' , nn.BatchNorm2d(64)),
('relu', nn.PReLU(num_parameters=64))]))
self.cam_down_c2 = nn.Conv2d(64, 1, 1, stride=1)
self.cam_conv3_1 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(576, 256, 3, stride=1, padding=1)),
('bn' , nn.BatchNorm2d(256)),
('relu', nn.PReLU(num_parameters=256))]))
self.cam3 = CAM_Module(256)
self.cam_conv3_2 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(256, 128, 3, stride=1, padding=1)),
('bn' , nn.BatchNorm2d(128)),
('relu', nn.PReLU(num_parameters=128))]))
self.cam_drop3 = nn.Dropout2d(0.1, False)
self.cam_conv3_3 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(128, 64, 1, stride=1)),
('bn' , nn.BatchNorm2d(64)),
('relu', nn.PReLU(num_parameters=64))]))
self.cam_down_c3 = nn.Conv2d(64, 1, 1, stride=1)
self.cam_conv4_1 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(1056, 512, 3, stride=1, padding=1)),
('bn' , nn.BatchNorm2d(512)),
('relu', nn.PReLU(num_parameters=512))]))
self.cam4 = CAM_Module(512)
self.cam_conv4_2 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(512, 256, 3, stride=1, padding=1)),
('bn' , nn.BatchNorm2d(256)),
('relu', nn.PReLU(num_parameters=256))]))
self.cam_drop4 = nn.Dropout2d(0.1, False)
self.cam_conv4_3 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(256, 64, 1, stride=1)),
('bn' , nn.BatchNorm2d(64)),
('relu', nn.PReLU(num_parameters=64))]))
self.cam_down_c4 = nn.Conv2d(64, 1, 1, stride=1)
self.cam_conv5_1 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(256, 128, 3, stride=1, padding=1)),
('bn' , nn.BatchNorm2d(128)),
('relu', nn.PReLU(num_parameters=128))]))
self.cam5 = CAM_Module(128)
self.cam_conv5_2 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(128, 64, 3, stride=1, padding=1)),
('bn' , nn.BatchNorm2d(64)),
('relu', nn.PReLU(num_parameters=64))]))
self.cam_drop5 = nn.Dropout2d(0.1, False)
self.cam_conv5_3 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(64, 32, 1, stride=1)),
('bn' , nn.BatchNorm2d(32)),
('relu', nn.PReLU(num_parameters=32))]))
self.cam_down_c5 = nn.Conv2d(32, 1, 1, stride=1)
self.conv_cat_1 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(32, 32, 3, stride=1, padding=1)),
('bn' , nn.BatchNorm2d(32)),
('relu', nn.PReLU(num_parameters=32))]))
self.conv_cat_2 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(64, 16, 3, stride=1, padding=1)),
('bn' , nn.BatchNorm2d(16)),
('relu', nn.PReLU(num_parameters=16))]))
self.conv_cat_3 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(64, 16, 3, stride=1, padding=1)),
('bn' , nn.BatchNorm2d(16)),
('relu', nn.PReLU(num_parameters=16))]))
self.conv_cat_4 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(64, 16, 3, stride=1, padding=1)),
('bn' , nn.BatchNorm2d(16)),
('relu', nn.PReLU(num_parameters=16))]))
self.conv_cat_5 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(32, 16, 3, stride=1, padding=1)),
('bn' , nn.BatchNorm2d(16)),
('relu', nn.PReLU(num_parameters=16))]))
self.conv_p1_0 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(32, 32, 3, stride=1, padding=1)),
('relu', nn.PReLU(num_parameters=32))]))
self.conv_p1_1 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(32, 32, 3, stride=1, padding=1)),
('relu', nn.PReLU(num_parameters=32))]))
self.conv_p1_2 = nn.Conv2d(32, 1, 1, stride=1)
self.conv_p2_up_0 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(16, 16, 3, stride=1, padding=1)),
('relu', nn.PReLU(num_parameters=16))]))
self.conv_p2_up_1 = nn.Conv2d(16, 1, 1, stride=1)
self.conv_p3_up_0 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(16, 16, 3, stride=1, padding=1)),
('relu', nn.PReLU(num_parameters=16))]))
self.conv_p3_up_1 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(16, 16, 3, stride=1, padding=1)),
('relu', nn.PReLU(num_parameters=16))]))
self.conv_p3_up_2 = nn.Conv2d(16, 1, 1, stride=1)
self.conv_p4_up_0 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(16, 16, 3, stride=1, padding=1)),
('relu', nn.PReLU(num_parameters=16))]))
self.conv_p4_up_1 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(16, 16, 3, stride=1, padding=1)),
('relu', nn.PReLU(num_parameters=16))]))
self.conv_p4_up_2 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(16, 16, 3, stride=1, padding=1)),
('relu', nn.PReLU(num_parameters=16))]))
self.conv_p4_up_3 = nn.Conv2d(16, 1, 1, stride=1)
self.conv_p5_up_0 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(16, 16, 3, stride=1, padding=1)),
('relu', nn.PReLU(num_parameters=16))]))
self.conv_p5_up_1 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(16, 16, 3, stride=1, padding=1)),
('relu', nn.PReLU(num_parameters=16))]))
self.conv_p5_up_2 = nn.Sequential(OrderedDict([
('conv', nn.Conv2d(16, 16, 3, stride=1, padding=1)),
('relu', nn.PReLU(num_parameters=16))]))
self.conv_p5_up_3 = nn.Conv2d(16, 1, 1, stride=1)
self.fuse = nn.Conv2d(36, 1, 1, stride=1)
if pretrained:
for key in self.state_dict():
if 'resnet' not in key:
self.init_layer(key)