Resnet

发表于 | 更新于 | 分类于 | 阅读次数：

吴恩达深度学习第四课第二周作业

网络按照下图构建

1713275526205

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
残差块构建
from torch import nn

class Resdual(nn.Module):

    def __init__(self,inchannel, outchannel,stride=1):
        super().__init__()
        self.conv1 = nn.Conv2d(in_channels=inchannel, out_channels=outchannel,kernel_size=3,stride=stride,padding=1)
        self.batchnorm1 = nn.BatchNorm2d(outchannel)
        self.relu = nn.ReLU()

        self.conv2 = nn.Conv2d(in_channels=outchannel, out_channels=outchannel,kernel_size=3,padding=1)
        self.batchnorm2 = nn.BatchNorm2d(outchannel)

        if inchannel != outchannel:
            self.conv1x1 = nn.Conv2d(in_channels=inchannel,out_channels=outchannel,kernel_size=1,stride=stride)
        else:
            self.conv1x1 = None

    def forward(self,x):
        output1 = self.relu(self.batchnorm1(self.conv1(x)))
        output2 = self.batchnorm2(self.conv2(output1))

        if self.conv1x1:
            x = self.conv1x1(x)
        return self.relu(x + output2)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
残差网络构建
import torch
from torch import nn

from model.resdual import Resdual


class Resnet(nn.Module):
    def __init__(self):
        super().__init__()
        #conv1
        self.conv1 = nn.Sequential(
            nn.Conv2d(in_channels=3, out_channels=64, kernel_size=7, stride=2),
            nn.BatchNorm2d(64),
            nn.ReLU()
        )

        #conv2
        self.conv2 = nn.Sequential(
            nn.MaxPool2d(kernel_size=3, stride=2),
            Resdual(64,64),
            Resdual(64,64),
            Resdual(64,64)
        )

        #conv3
        self.conv3 = nn.Sequential(
            Resdual(64,128,stride=2),
            Resdual(128,128),
            Resdual(128, 128),
            Resdual(128, 128)
        )

        #conv4
        self.conv4 = nn.Sequential(
            Resdual(128,256,stride=2),
            Resdual(256,256),
            Resdual(256, 256),
            Resdual(256, 256),
            Resdual(256, 256),
            Resdual(256, 256)
        )

        #conv5
        self.conv5 = nn.Sequential(
            Resdual(256,512,stride=2),
            Resdual(512,512),
            Resdual(512, 512)
        )

        self.average_pool = nn.AdaptiveAvgPool2d(1)
        self.flatten = nn.Flatten()
        self.fc = nn.Linear(512,6)

    def forward(self,x):
        x = self.conv1(x)
        x = self.conv2(x)
        x = self.conv3(x)
        x = self.conv4(x)
        x = self.conv5(x)
        x = self.average_pool(x)
        x = self.flatten(x)
        x = self.fc(x)
        return x

# a = torch.ones((1000,3,64,64))
# model = Resnet()
# output = model(a)
# print(output.shape)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
自定义数据
import torch
from torch.utils.data import Dataset, DataLoader
import h5py
import numpy as np
import os

class Mydata(Dataset):
    def __init__(self,path):
        super().__init__()
        file = h5py.File(path,'r')
        if path == 'datasets/test_signs.h5':
            self.x = np.array(file['test_set_x'])
            self.y = np.array(file['test_set_y'])
        else:
            self.x = np.array(file['train_set_x'])
            self.y = np.array(file['train_set_y'])

        self.x = self.x.astype('float32')/255

        self.x = torch.from_numpy(self.x)
        self.y = torch.from_numpy(self.y)

        self.x = self.x.permute(0,3,1,2)


    def __getitem__(self, item):
        return self.x[item], self.y[item]

    def __len__(self):
        return self.x.shape[0]
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
主函数
import torch
from torch import nn
from torch.utils.data import DataLoader

from datasets import Mydata
from model.resnet import Resnet


def check(model, test_loader, len):
    acc = 0
    with torch.no_grad():
        for i, (imgs, labels ) in enumerate(test_loader):
            imgs = imgs.to(device)
            labels = labels.to(device)
            output = model(imgs)
            predict = output.argmax(1)
            acc += (predict == labels).sum()

    print("test数据集上acc为{}".format(acc/len))


if __name__ == '__main__':
    #加载数据
    train_path = "datasets/train_signs.h5"
    test_path = "datasets/test_signs.h5"
    train_data = Mydata(train_path)
    test_data = Mydata(test_path)

    train_loader = DataLoader(train_data,batch_size=64,shuffle=True)
    test_loader = DataLoader(test_data, batch_size=64, shuffle=True)

    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
    #加载模型
    model = Resnet()
    model.to(device)

    #定义loss以及优化器
    loss_func = nn.CrossEntropyLoss()
    loss_func.to(device)

    optim = torch.optim.Adam(model.parameters(),lr=0.01)

    epoch = 50
    for i in range(epoch):
        print("----------第{}轮次开始--------".format(i+1))
        loss_cnt = 0
        acc = 0
        for i, (imgs, labels) in enumerate(train_loader):
            imgs = imgs.to(device)
            labels = labels.to(device)

            #64 * 6
            output = model(imgs)
            loss = loss_func(output, labels)

            loss_cnt += loss
            acc += (output.argmax(1) == labels).sum()

            optim.zero_grad()
            loss.backward()
            optim.step()
        print("第{}轮次的loss为{}".format(i+1, loss_cnt))
        print("第{}轮次的acc为{}".format(i+1, acc/len(train_data)))

    check(model, test_loader, len(test_data))