def __init__(self, channel=32):
super(MyNet, self).__init__()
self.backbone = pvt_v2_b2() # [64, 128, 320, 512]
path = 'F:\pretrain\pvt_v2_b3.pth'
save_model = torch.load(path)
model_dict = self.backbone.state_dict()
state_dict = {
k: v
for k, v in save_model.items() if k in model_dict.keys()
}
model_dict.update(state_dict)
self.backbone.load_state_dict(model_dict)
self.Translayer1 = BasicConv2d(64, channel, 1)
self.Translayer2 = BasicConv2d(128, channel, 1)
self.Translayer3 = BasicConv2d(320, channel, 1)
self.Translayer4 = BasicConv2d(512, channel, 1)
self.upsample = nn.Upsample(scale_factor=2,
mode='bilinear',
align_corners=True)
self.down01 = nn.Upsample(scale_factor=1 / 2,
mode='bilinear',
align_corners=True)
self.down02 = nn.Upsample(scale_factor=1 / 4,
mode='bilinear',
align_corners=True)
self.down03 = nn.Upsample(scale_factor=1 / 8,
mode='bilinear',
align_corners=True)
self.down04 = nn.Upsample(scale_factor=1 / 16,
mode='bilinear',
align_corners=True)
self.upsample1 = nn.Upsample(scale_factor=32,
mode='bilinear',
align_corners=True)
self.upsample2 = nn.Upsample(scale_factor=16,
mode='bilinear',
align_corners=True)
self.upsample3 = nn.Upsample(scale_factor=8,
mode='bilinear',
align_corners=True)
self.out1_1 = BasicConv2d(channel * 3, channel, 1)
self.out1_2 = BasicConv2d(channel * 3, channel, 1)
self.out1_3 = BasicConv2d(channel * 3, channel, 1)
self.out1_4 = BasicConv2d(channel, channel, 1)
self.refineconv = BasicConv2d(3, 1, 1)
self.refine = RefineUNet(1, 1)
self.ca = ChannelAttention(channel)
self.sa = SpatialAttention()
self.outatte = nn.Conv2d(channel, channel, 1)
# Decoder
# self.decoder5 = DecoderBlock(in_channels=512, out_channels=512)
self.decoder4 = DecoderBlock(in_channels=channel, out_channels=channel)
self.decoder3 = DecoderBlock(in_channels=channel * 3,
out_channels=channel)
self.decoder2 = DecoderBlock(in_channels=channel * 3,
out_channels=channel)
self.decoder1 = nn.Sequential(BasicConv2d(channel * 3, channel, 1),
BasicConv2d(channel, channel, 1))
# self.decoder5 = DecoderBlock(in_channels=512, out_channels=512)
self.decoder2_4 = DecoderBlock(in_channels=channel,
out_channels=channel)
self.decoder2_3 = DecoderBlock(in_channels=channel * 2,
out_channels=channel)
self.decoder2_2 = DecoderBlock(in_channels=channel * 2,
out_channels=channel)
self.decoder2_1 = DecoderBlock(in_channels=channel * 2,
out_channels=channel)
# adaptive selection module
self.asm4 = ASM(channel, channel * 3)
self.asm3 = ASM(channel, channel * 3)
self.asm2 = ASM(channel, channel * 3)
self.asm1 = ASM(channel, channel * 3)
self.unetout1 = nn.Conv2d(channel, 1, 1)
self.unetout2 = nn.Conv2d(channel, 1, 1)
self.COM = COM(channel)
self.cobv1 = BasicConv2d(3 * channel, channel, 1)
self.cobv2 = BasicConv2d(3 * channel, channel, 1)
self.nocal = NonLocalBlock(channel)
self.selayer = SELayer(channel)
def __init__(self, channel=64):
super(MyNet, self).__init__()
self.backbone = pvt_v2_b2() # [64, 128, 320, 512]
path = 'F:\pretrain\pvt_v2_b3.pth'
save_model = torch.load(path)
model_dict = self.backbone.state_dict()
state_dict = {
k: v
for k, v in save_model.items() if k in model_dict.keys()
}
model_dict.update(state_dict)
self.backbone.load_state_dict(model_dict)
self.Translayer2_0 = BasicConv2d(64, channel, 1)
self.Translayer2_1 = BasicConv2d(128, channel, 1)
self.Translayer3_1 = BasicConv2d(320, channel, 1)
self.Translayer4_1 = BasicConv2d(512, channel, 1)
# self.CFM = CFM(channel)
#
# self.SAM = SAM()
self.upsample = nn.Upsample(scale_factor=2,
mode='bilinear',
align_corners=True)
self.down05 = nn.Upsample(scale_factor=0.5,
mode='bilinear',
align_corners=True)
self.out_SAM = nn.Conv2d(channel, 1, 1)
self.out_CFM = nn.Conv2d(channel, 1, 1)
self.rf1 = RF(512 + 320 + 128, channel)
self.rf2 = RF(512 + 320 + 128, channel)
self.rf3 = RF(512 + 320 + 128, channel)
self.att1 = BCA(channel, channel, channel)
self.att2 = BCA(channel, channel, channel)
self.att3 = BCA(channel, channel, channel)
self.out_SAM = nn.Conv2d(channel, 1, 1)
self.out_CFM = nn.Conv2d(channel, 1, 1)
self.down01 = nn.Upsample(scale_factor=1 / 8,
mode='bilinear',
align_corners=True)
self.down02 = nn.Upsample(scale_factor=1 / 4,
mode='bilinear',
align_corners=True)
self.down03 = nn.Upsample(scale_factor=1 / 2,
mode='bilinear',
align_corners=True)
self.upsample1 = nn.Upsample(scale_factor=32,
mode='bilinear',
align_corners=True)
self.upsample2 = nn.Upsample(scale_factor=16,
mode='bilinear',
align_corners=True)
self.upsample3 = nn.Upsample(scale_factor=8,
mode='bilinear',
align_corners=True)
self.out1 = BasicConv2d(channel, 1, 1)
self.out2 = BasicConv2d(channel, 1, 1)
self.out3 = BasicConv2d(channel, 1, 1)
self.out4 = BasicConv2d(channel, 1, 1)
self.refineconv = BasicConv2d(3, 1, 1)
self.refine = RefineUNet(1, 1)
self.ca = ChannelAttention(channel)
self.sa = SpatialAttention()
self.outatte = nn.Conv2d(channel, channel, 1)
self.conv_upsample4 = BasicConv2d(channel, channel, 3, padding=1)
self.conv_upsample5 = BasicConv2d(2 * channel,
2 * channel,
3,
padding=1)
def __init__(self, channel=64):
super(MyNet, self).__init__()
self.backbone = pvt_v2_b2() # [64, 128, 320, 512]
path = 'F:\pretrain\pvt_v2_b3.pth'
save_model = torch.load(path)
model_dict = self.backbone.state_dict()
state_dict = {
k: v
for k, v in save_model.items() if k in model_dict.keys()
}
model_dict.update(state_dict)
self.backbone.load_state_dict(model_dict)
self.Translayer1 = BasicConv2d(64, channel, 1)
self.Translayer2 = BasicConv2d(128, channel, 1)
self.Translayer3 = BasicConv2d(320, channel, 1)
self.Translayer4 = BasicConv2d(512, channel, 1)
self.upsample = nn.Upsample(scale_factor=2,
mode='bilinear',
align_corners=True)
self.down05 = nn.Upsample(scale_factor=0.5,
mode='bilinear',
align_corners=True)
self.down01 = nn.Upsample(scale_factor=1 / 2,
mode='bilinear',
align_corners=True)
self.down02 = nn.Upsample(scale_factor=1 / 4,
mode='bilinear',
align_corners=True)
self.down03 = nn.Upsample(scale_factor=1 / 8,
mode='bilinear',
align_corners=True)
self.down04 = nn.Upsample(scale_factor=1 / 16,
mode='bilinear',
align_corners=True)
self.upsample1 = nn.Upsample(scale_factor=32,
mode='bilinear',
align_corners=True)
self.upsample2 = nn.Upsample(scale_factor=16,
mode='bilinear',
align_corners=True)
self.upsample3 = nn.Upsample(scale_factor=8,
mode='bilinear',
align_corners=True)
self.out1_1 = BasicConv2d(64, channel, 1)
self.out1_2 = BasicConv2d(128, channel, 1)
self.out1_3 = BasicConv2d(320, channel, 1)
self.out1_4 = BasicConv2d(512, channel, 1)
self.out2_1 = BasicConv2d(64, 1, 1)
self.out2_2 = BasicConv2d(128, 1, 1)
self.out2_3 = BasicConv2d(320, 1, 1)
self.out2_4 = BasicConv2d(512, 1, 1)
self.refineconv = BasicConv2d(3, 1, 1)
self.refine = RefineUNet(1, 1)
self.ca = ChannelAttention(channel)
self.sa = SpatialAttention()
self.outatte = nn.Conv2d(channel, channel, 1)
# Decoder
# self.decoder5 = DecoderBlock(in_channels=512, out_channels=512)
# self.decoder4 = DecoderBlock(in_channels=channel, out_channels=channel)
# self.decoder3 = DecoderBlock(in_channels=channel*2, out_channels=channel)
# self.decoder2 = DecoderBlock(in_channels=channel*2, out_channels=channel)
# self.decoder1 = DecoderBlock(in_channels=channel*2, out_channels=channel)
#
# self.decoder5 = DecoderBlock(in_channels=512, out_channels=512)
self.decoder2_4 = DecoderBlock(in_channels=channel, out_channels=320)
self.decoder2_3 = DecoderBlock(in_channels=channel * 2,
out_channels=128)
self.decoder2_2 = DecoderBlock(in_channels=channel * 2,
out_channels=64)
self.decoder2_1 = DecoderBlock(in_channels=channel * 2,
out_channels=64)
# adaptive selection module
self.asm4 = ASM(channel, 512)
self.asm3 = ASM(channel, channel * 3)
self.asm2 = ASM(channel, channel * 3)
self.asm1 = ASM(channel, channel * 3)
self.unetout1 = nn.Sequential(
BasicConv2d(64, 32, kernel_size=3, stride=1, padding=1),
nn.Conv2d(32, 1, 1))
self.unetout1 = nn.Sequential(
BasicConv2d(channel, channel, kernel_size=3, stride=1, padding=1),
nn.Conv2d(channel, 1, 1))
# Sideout
self.sideout4 = SideoutBlock(32, 1)
self.sideout3 = SideoutBlock(32, 1)
self.sideout2 = SideoutBlock(32, 1)
self.sideout1 = SideoutBlock(32, 1)
self.decoder1 = Decoder()
self.decoder2 = Decoder()
self.acfm3 = ACFM()
self.acfm2 = ACFM()
self.dgcm3 = DGCM()
self.dgcm2 = DGCM()
self.upconv3 = BasicConv2d(64,
64,
kernel_size=3,
stride=1,
padding=1,
relu=True)
self.classifier = nn.Conv2d(64, 1, 1)
self.upconv2 = BasicConv2d(64,
64,
kernel_size=3,
stride=1,
padding=1,
relu=True)
def __init__(self, channel=64):
super(MyNet, self).__init__()
self.backbone = pvt_v2_b2() # [64, 128, 320, 512]
self.resnet = res2net50_v1b_26w_4s(pretrained=True)
self.Translayer1 = BasicConv2d(256, channel, 1)
self.Translayer2 = BasicConv2d(512, channel, 1)
self.Translayer3 = BasicConv2d(1024, channel, 1)
self.Translayer4 = BasicConv2d(2048, channel, 1)
self.upsample = nn.Upsample(scale_factor=2,
mode='bilinear',
align_corners=True)
self.down05 = nn.Upsample(scale_factor=0.5,
mode='bilinear',
align_corners=True)
self.down01 = nn.Upsample(scale_factor=1 / 2,
mode='bilinear',
align_corners=True)
self.down02 = nn.Upsample(scale_factor=1 / 4,
mode='bilinear',
align_corners=True)
self.down03 = nn.Upsample(scale_factor=1 / 8,
mode='bilinear',
align_corners=True)
self.down04 = nn.Upsample(scale_factor=1 / 16,
mode='bilinear',
align_corners=True)
self.upsample1 = nn.Upsample(scale_factor=32,
mode='bilinear',
align_corners=True)
self.upsample2 = nn.Upsample(scale_factor=16,
mode='bilinear',
align_corners=True)
self.upsample3 = nn.Upsample(scale_factor=8,
mode='bilinear',
align_corners=True)
self.out1_1 = BasicConv2d(channel * 3, channel, 1)
self.out1_2 = BasicConv2d(channel * 3, channel, 1)
self.out1_3 = BasicConv2d(channel * 3, channel, 1)
self.out1_4 = BasicConv2d(channel * 3, channel, 1)
self.out2_1 = BasicConv2d(256, 1, 1)
self.out2_2 = BasicConv2d(512, 1, 1)
self.out2_3 = BasicConv2d(1024, 1, 1)
self.out2_4 = BasicConv2d(512, 1, 1)
self.refineconv = BasicConv2d(3, 1, 1)
self.refine = RefineUNet(1, 1)
self.ca = ChannelAttention(channel)
self.sa = SpatialAttention()
self.outatte = nn.Conv2d(channel, channel, 1)
# Decoder
# self.decoder5 = DecoderBlock(in_channels=512, out_channels=512)
self.decoder4 = DecoderBlock(in_channels=channel * 3,
out_channels=channel)
self.decoder3 = DecoderBlock(in_channels=channel * 3,
out_channels=channel)
self.decoder2 = DecoderBlock(in_channels=channel * 3,
out_channels=channel)
self.decoder1 = nn.Sequential(BasicConv2d(channel * 2, channel, 1),
BasicConv2d(channel, channel, 1))
# self.decoder5 = DecoderBlock(in_channels=512, out_channels=512)
self.decoder2_4 = DecoderBlock(in_channels=channel,
out_channels=channel)
self.decoder2_3 = DecoderBlock(in_channels=channel * 2,
out_channels=channel)
self.decoder2_2 = DecoderBlock(in_channels=channel * 2,
out_channels=channel)
self.decoder2_1 = DecoderBlock(in_channels=channel * 2,
out_channels=channel)
# adaptive selection module
self.asm4 = ASM(channel, channel * 3)
self.asm3 = ASM(channel, channel * 3)
self.asm2 = ASM(channel, channel * 3)
self.asm1 = ASM(channel, channel * 3)
self.unetout1 = nn.Conv2d(channel, 1, 1)
self.unetout2 = nn.Conv2d(channel, 1, 1)
self.COM = COM(channel)
self.cobv1 = BasicConv2d(3 * channel, channel, 1)
self.cobv2 = BasicConv2d(3 * channel, channel, 1)
self.nocal = NonLocalBlock(channel)
def __init__(self, channel=64):
super(MyNet, self).__init__()
self.resnet = res2net50_v1b_26w_4s(pretrained=True)
self.Translayer1= BasicConv2d(64, channel, 1)
self.Translayer2 = BasicConv2d(128, channel, 1)
self.Translayer3 = BasicConv2d(320, channel, 1)
self.Translayer4 = BasicConv2d(512, channel, 1)
self.Translayerup1 = BasicConv2d(256 ,channel, 1)
self.Translayerup2 = BasicConv2d(512, channel, 1)
self.upsample = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=True)
self.downsample = nn.Upsample(scale_factor=1/4, mode='bilinear', align_corners=True)
self.down01 = nn.Upsample(scale_factor=1 / 2, mode='bilinear', align_corners=True)
self.down02 = nn.Upsample(scale_factor=1/4, mode='bilinear', align_corners=True)
self.down03 = nn.Upsample(scale_factor=1/8, mode='bilinear', align_corners=True)
self.down04 = nn.Upsample(scale_factor=1/16, mode='bilinear', align_corners=True)
self.upsample1 = nn.Upsample(scale_factor=32, mode='bilinear', align_corners=True)
self.upsample2 = nn.Upsample(scale_factor=16, mode='bilinear', align_corners=True)
self.upsample3 = nn.Upsample(scale_factor=8, mode='bilinear', align_corners=True)
self.out1_1 = BasicConv2d(channel*2, channel, 1)
self.out1_2 = BasicConv2d(channel*2, channel, 1)
self.out1_3 = BasicConv2d(channel*2, channel, 1)
self.out1_4 = BasicConv2d(channel, channel, 1)
self.refineconv = BasicConv2d(3, 1, 1)
self.refine = RefineUNet(1,1)
self.ca1 = ChannelAttention(channel)
self.sa4= SpatialAttention()
self.outatte = nn.Conv2d(channel, channel, 1)
# self.decoder5 = DecoderBlock(in_channels=512, out_channels=512)
self.decoder4 = DecoderBlock(in_channels=channel, out_channels=channel)
self.decoder3 = DecoderBlock(in_channels=channel*2, out_channels=channel)
self.decoder2 = DecoderBlock(in_channels=channel*2, out_channels=channel)
self.decoder1 = nn.Sequential(BasicConv2d(channel*2, channel,1),
BasicConv2d(channel, channel,1))
# self.decoder5 = DecoderBlock(in_channels=512, out_channels=512)
self.decoder2_4 = DecoderBlock(in_channels=channel, out_channels=channel)
self.decoder2_3 = DecoderBlock(in_channels=channel*2, out_channels=channel)
self.decoder2_2 = DecoderBlock(in_channels=channel*2, out_channels=channel)
self.decoder2_1 = DecoderBlock(in_channels=channel*2, out_channels=channel)
self.unetout1 = nn.Conv2d(channel, 1, 1)
self.unetout2 = nn.Conv2d(channel, 1, 1)
self.detailout = nn.Conv2d(channel*2, 1, 1)
self.com =COM(channel)
self.cobv1 =BasicConv2d(3*channel,channel,1)
self.cobv2 =BasicConv2d(3*channel,channel,1)
self.nocal = NonLocalBlock(channel)
self.selayer = SELayer(channel)
self.upconv1 =BasicConv2d(channel,channel,1)
self.noncal = BCA(channel, channel, 16)
self.duatt =_DAHead(channel)
self.ca = ChannelAttention(channel)
self.sa = SpatialAttention()
self.conv = nn.Sequential(
BasicConv2d(channel*2,channel*2,1),
BasicConv2d(channel*2,channel,1)
)
self.edgeconv =BasicConv2d(channel,channel,1)
self.downconv =BasicConv2d(channel,channel,1)
self.rfb2_1 = RFB_modified(512, channel)
self.rfb1_1 = RFB_modified(256, channel)
self.rfb3_1 = RFB_modified(1024, channel)
self.rfb4_1 = RFB_modified(2048, channel)
def __init__(self, channel=32):
super(MyNet, self).__init__()
self.backbone = pvt_v2_b2() # [64, 128, 320, 512]
path = 'F:\pretrain\pvt_v2_b3.pth'
save_model = torch.load(path)
model_dict = self.backbone.state_dict()
state_dict = {
k: v
for k, v in save_model.items() if k in model_dict.keys()
}
model_dict.update(state_dict)
self.backbone.load_state_dict(model_dict)
self.Translayer1 = BasicConv2d(64, channel, 1)
self.Translayer2 = BasicConv2d(128, channel, 1)
self.Translayer3 = BasicConv2d(320, channel, 1)
self.Translayer4 = BasicConv2d(512, channel, 1)
self.Translayer5 = BasicConv2d(64, channel, 1)
self.Translayer6 = BasicConv2d(128, channel, 1)
self.Translayer7 = BasicConv2d(320, channel, 1)
self.Translayer8 = BasicConv2d(512, channel, 1)
self.upsample = nn.Upsample(scale_factor=2,
mode='bilinear',
align_corners=True)
self.downsample = nn.Upsample(scale_factor=1 / 4,
mode='bilinear',
align_corners=True)
self.down01 = nn.Upsample(scale_factor=1 / 2,
mode='bilinear',
align_corners=True)
self.down02 = nn.Upsample(scale_factor=1 / 4,
mode='bilinear',
align_corners=True)
self.down03 = nn.Upsample(scale_factor=1 / 8,
mode='bilinear',
align_corners=True)
self.down04 = nn.Upsample(scale_factor=1 / 16,
mode='bilinear',
align_corners=True)
self.upsample1 = nn.Upsample(scale_factor=32,
mode='bilinear',
align_corners=True)
self.upsample2 = nn.Upsample(scale_factor=16,
mode='bilinear',
align_corners=True)
self.upsample3 = nn.Upsample(scale_factor=8,
mode='bilinear',
align_corners=True)
self.out1_1 = BasicConv2d(channel * 2, channel, 1)
self.out1_2 = BasicConv2d(channel * 2, channel, 1)
self.out1_3 = BasicConv2d(channel * 2, channel, 1)
self.out1_4 = BasicConv2d(channel, channel, 1)
self.refineconv = BasicConv2d(3, 1, 1)
self.refine = RefineUNet(1, 1)
self.ca1 = ChannelAttention(channel)
self.sa4 = SpatialAttention()
self.outatte = nn.Conv2d(channel, channel, 1)
# self.decoder5 = DecoderBlock(in_channels=512, out_channels=512)
self.decoder4 = DecoderBlock(in_channels=channel, out_channels=channel)
self.decoder3 = DecoderBlock(in_channels=channel * 2,
out_channels=channel)
self.decoder2 = DecoderBlock(in_channels=channel * 2,
out_channels=channel)
self.decoder1 = nn.Sequential(BasicConv2d(channel * 2, channel, 1),
BasicConv2d(channel, channel, 1))
self.decoder5 = DecoderBlock(in_channels=channel * 2,
out_channels=channel)
self.decoder6 = DecoderBlock(in_channels=channel * 2,
out_channels=channel)
self.decoder7 = DecoderBlock(in_channels=channel * 2,
out_channels=channel)
self.decoder8 = nn.Sequential(BasicConv2d(channel * 2, channel, 1),
BasicConv2d(channel, channel, 1))
self.encoder2 = EncooderBlock(in_channels=channel * 2,
out_channels=channel)
self.encoder3 = EncooderBlock(in_channels=channel * 2,
out_channels=channel)
self.encoder4 = EncooderBlock(in_channels=channel * 2,
out_channels=channel)
self.unetout1 = nn.Conv2d(channel, 1, 1)
self.unetout2 = nn.Conv2d(channel, 1, 1)
self.detailout = nn.Conv2d(channel * 2, 1, 1)
self.cobv1 = BasicConv2d(3 * channel, channel, 1)
self.cobv2 = BasicConv2d(3 * channel, channel, 1)
self.selayer = SELayer(channel)
self.upconv1 = BasicConv2d(channel, channel, 1)
self.noncal = BCA(channel, channel, 16)
self.duatt = _DAHead(channel)
self.ca = ChannelAttention(channel * 2)
self.sa = SpatialAttention()
self.conv = nn.Sequential(BasicConv2d(channel * 3, channel * 2, 1),
BasicConv2d(channel * 2, channel * 2, 1),
BasicConv2d(channel * 2, channel, 1))
self.edgeconv = BasicConv2d(channel, channel, 1)
self.downconv = BasicConv2d(channel * 2, channel, 1)
self.catt1 = CasAtt(channel)
self.catt2 = CasAtt(channel)