Code fragments of isicnet.py

• import torch
• import torch.nn as nn
• from torchvision.models import resnet
• class ISICNet(nn.Module):
• def __init__(self, n_feature=3, n_class=8, backbone='resnet18'):
• super(ISICNet, self).__init__()
• self.n_feature = n_feature
• self.n_class = n_class
• self.backbnone = backbone
• if self.backbnone == 'resnet18':
• base = resnet.resnet18(pretrained=True)
• self.resnet_expansion = 1
• self.in_block = nn.Sequential( nn.Conv2d(self.n_feature, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False), base.bn1, base.relu, base.maxpool)
• self.encoder1 = base.layer1
• self.encoder2 = base.layer2
• self.encoder3 = base.layer3
• self.encoder4 = base.layer4
• self.avgpool = base.avgpool
• self.flatten = nn.Flatten()
• self.fc = nn.Linear(512*self.resnet_expansion, self.n_class , bias=True)
• def forward(self, x):
• h = self.in_block(x)
• h = self.encoder1(h)
• h = self.encoder2(h)
• h = self.encoder3(h)
• h = self.encoder4(h)
• y = self.fc(self.flatten(self.avgpool(h)))
• return y
• if __name__ == "__main__":
• model = ISICNet().cuda()
• x = torch.rand(2, 3, 512, 512).cuda()
• y = model(x)
• print(y)
• print(y.data.max(1)[1])
• # t = torch.tensor([1.0, 0.0, 1.0])
• # y = torch.tensor([0.02, 0.05, 0.99])
• # bce = nn.BCELoss()
• # l = torch.log(torch.tensor(0.02))+ torch.log(torch.tensor(1-0.05)) + torch.log(torch.tensor(0.99))
• # print(bce(y, t))
• # print(l)

Graph