def __init__(self):
super(EmbeddingNetwork, self).__init__()
self.resnet = res18_oct_or_octa()
self.resnet.load_state_dict(torch.load('./res18_octa.pth'))
#180 0.8116 210 0.7729 240 0.8043 270:0.837 300:0.8442 360:0.8490 390:0.8297 420:0.8418 450:0.8297
#60 0.8176 120:0.8249 810:0.8285 720:0.9034
self.cls = self.resnet.cls
self.cls.load_state_dict(self.resnet.cls.state_dict())
self.conv1 = self.resnet.conv1
self.conv1.load_state_dict(self.resnet.conv1.state_dict())
self.bn1 = self.resnet.bn1
self.bn1.load_state_dict(self.resnet.bn1.state_dict())
self.relu = self.resnet.relu
self.maxpool = self.resnet.maxpool
self.layer1 = self.resnet.layer1
self.layer1.load_state_dict(self.resnet.layer1.state_dict())
self.layer2 = self.resnet.layer2
self.layer2.load_state_dict(self.resnet.layer2.state_dict())
self.layer3 = self.resnet.layer3
self.layer3.load_state_dict(self.resnet.layer3.state_dict())
self.layer4 = self.resnet.layer4
self.layer4.load_state_dict(self.resnet.layer4.state_dict())
self.avgpool = self.resnet.avgpool
def __init__(self):
super(EmbeddingNetwork, self).__init__()
self.resnet = res18_oct_or_octa()
self.resnet.load_state_dict(torch.load('./res18_oct.pth'))
self.cls = self.resnet.cls
self.cls.load_state_dict(self.resnet.cls.state_dict())
self.conv1 = self.resnet.conv1
self.conv1.load_state_dict(self.resnet.conv1.state_dict())
self.bn1 = self.resnet.bn1
self.bn1.load_state_dict(self.resnet.bn1.state_dict())
self.relu = self.resnet.relu
self.maxpool = self.resnet.maxpool
self.layer1 = self.resnet.layer1
self.layer1.load_state_dict(self.resnet.layer1.state_dict())
self.layer2 = self.resnet.layer2
self.layer2.load_state_dict(self.resnet.layer2.state_dict())
self.layer3 = self.resnet.layer3
self.layer3.load_state_dict(self.resnet.layer3.state_dict())
self.layer4 = self.resnet.layer4
self.layer4.load_state_dict(self.resnet.layer4.state_dict())
self.avgpool = self.resnet.avgpool
# x=torch.cat([x1,x2],1)
# x = self.conv1(x)
# x = self.bn1(x)
# x = self.relu(x)
# x = self.maxpool(x)
# layer1 = self.layer1(x) # (, 64L, 56L, 56L)
# layer2 = self.layer2(layer1) # (, 128L, 28L, 28L)
# layer3 = self.layer3(layer2) # (, 256L, 14L, 14L)
# layer4 = self.layer4(layer3) # (,512,7,7)
# x = self.avgpool(layer4) # (,512,1,1)
#
#
# x = x.view(x.size(0), -1)
# return x
classificationNetwork = res18_oct_or_octa().cuda()
#############################################
# Define the optimizer
criterion = nn.CrossEntropyLoss()
optimizer_embedding = optim.Adam([
{
'params': classificationNetwork.parameters()
},
],
lr=0.001)
embedding_lr_scheduler = lr_scheduler.StepLR(optimizer_embedding,
step_size=10,