迁移学习第一篇 - 赵志斌UDTL backbone代码

Label-consistent UDTL(unsupervised deep transfer learning)
本文内容：赵志斌UDTL开源代码的backbone实现

代码github链接：https://github.com/HappyBoy-cmd/UDTL

backbone主干网络模型组成

网络模型

网络模型参数

模型pytorch代码

#!/usr/bin/python
# -*- coding:utf-8 -*-
from torch import nn
import warnings

# ----------------------------inputsize >=28-------------------------------------------------------------------------

class CNN(nn.Module):
    def __init__(self, pretrained=False, in_channel=1, out_channel=10):
        super(CNN, self).__init__()
        if pretrained == True:
            warnings.warn("Pretrained model is not available")

        self.layer1 = nn.Sequential(
            nn.Conv1d(in_channel, 16, kernel_size=15),  # 16, 26 ,26
            nn.BatchNorm1d(16),
            nn.ReLU(inplace=True))


        self.layer2 = nn.Sequential(
            nn.Conv1d(16, 32, kernel_size=3),  # 32, 24, 24
            nn.BatchNorm1d(32),
            nn.ReLU(inplace=True),
            nn.MaxPool1d(kernel_size=2, stride=2),
            )  # 32, 12,12     (24-2) /2 +1

        self.layer3 = nn.Sequential(
            nn.Conv1d(32, 64, kernel_size=3),  # 64,10,10
            nn.BatchNorm1d(64),
            nn.ReLU(inplace=True))

        self.layer4 = nn.Sequential(
            nn.Conv1d(64, 128, kernel_size=3),  # 128,8,8
            nn.BatchNorm1d(128),
            nn.ReLU(inplace=True),
            nn.AdaptiveMaxPool1d(4))  # 128, 4,4

        self.layer5 = nn.Sequential(
            nn.Linear(128 * 4, 256),
            nn.ReLU(inplace=True),
            nn.Dropout(),
            nn.Linear(256, 256),
            nn.ReLU(inplace=True),
            nn.Dropout(),
        )
        self.fc = nn.Linear(256, out_channel)

    def forward(self, x):  # x输入为channel为1，长度为1024
        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)
        x = x.view(x.size(0), -1)
        x = self.layer5(x)
        x = self.fc(x)

        return x

模型测试：

model = CNN() #加载模型
x = torch.randn([16, 1, 1024]) # batch_size = 16, channel = 1, len = 1024
print(x)

>>>结果输出
tensor([[[ 1.6765e+00, -5.0673e-01, -8.4661e-01,  ...,  1.3182e-03,
           3.6003e-01, -4.9098e-01]],

        [[ 1.8038e-01,  3.3888e+00, -6.2151e-01,  ...,  1.8527e-01,
           8.6888e-01, -3.9862e-01]],

        [[-5.1029e-02, -2.7900e+00,  8.7660e-01,  ..., -4.1347e-01,
          -5.2011e-01, -1.1997e-01]],

        ...,

        [[-1.3845e+00, -9.7379e-01, -1.1276e+00,  ...,  4.0277e-01,
          -2.1397e+00,  3.9430e-02]],

        [[ 4.9035e-01,  1.4926e-01, -2.9260e-01,  ...,  3.6087e-01,
           5.0232e-01,  1.1382e+00]],

        [[-1.3410e+00, -8.0721e-01, -2.4605e+00,  ..., -1.5650e+00,
          -5.9978e-01, -9.1267e-01]]])``model(x)

y = model(x)
print(y)

tensor([[ 0.0508, -0.2657,  0.0036, -0.8904,  0.4402, -0.2396, -0.0749, -0.8506,
          0.0321, -1.0699],
        [-0.5058,  0.0264,  1.8257, -1.3112,  0.7088, -0.1765,  0.1929,  0.4886,
          0.1973, -0.8568],
        [-0.0061,  0.3359,  0.3943, -0.4675, -1.1316,  0.1924,  1.7509,  0.2536,
         -0.6900, -0.9577],
        [ 0.3317, -0.5512,  0.2652,  0.1834, -0.2543, -0.2176,  0.4419,  0.4914,
          1.1083, -0.7718],
        ...
        [-0.1593, -0.8255, -0.2867, -1.0740, -0.8413,  0.2205,  0.5246, -0.0087,
          0.0191, -0.9652],
        [ 0.1454, -1.6750,  0.3506, -0.1599, -0.2638,  0.5976,  0.0670, -0.7719,
          0.0851,  0.0249]], grad_fn=<AddmmBackward0>)