PyTorch 猫狗识别深度学习 I 低准确度版本

数据源

Cat and Dog

下载的文件层级为：

PetImages(doc) - Cat(doc)/Dog(doc) - image(jpg)

需要注意狗的 11702 和猫的 666 这两张图片损坏了。在加载数据集的时候会出现警告。

处理图片

获取猫狗图片的 txt

1
2
3
4
5
6
7
8
9
10
def get_txt(path, file_name, label):
    # path：dataset的根目录
    # file_path：dataset文件夹
    file_path = os.path.join(path, file_name)
    data = os.listdir(file_path)
    # 'w' is mean write
    f = open(path + '/' + file_name + 'data.txt', 'w')
    for item in data:
        f.write(item + ' ' + str(label) + ' ' + file_name + '\n')
    f.close()

这里是将 PetImages 文件夹下的 Cat 或者 Dog 文件打开，遍历里面的照片，最后输出 txt 文件。

txt 的内容为（以猫为例）：

1
2
3
4
1.jpg 0 Cat
2.jpg 0 Cat
3.jpg 0 Cat
4.jpg 0 Cat

这里，Cat 的 label 为 0，Dog 的 label 为 1。

混合猫狗图片的 txt

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
def mix_txt(path, file1_name, file2_name):
    f = open(path + '/' + 'DataSet.txt', 'w')
    # 'r' is mean read
    f1 = open(path + '/' + file1_name + 'Data.txt', 'r')
    # readline/readlines
    data1 = f1.readlines()
    # PC 上生成的 txt 有 DStore 信息。合并 txt 时需要只读取第一行之后的内容
    # data1 = data1[1:]
    for item in data1:
        f.write(item)

    f2 = open(path + '/' + file2_name + 'Data.txt', 'r')
    data2 = f2.readlines()
    # data2 = data2[1:]
    for item in data2:
        f.write(item)

    f.close()
    f1.close()
    f2.close()

构建数据集

定义数据集类，通过 txt 索引将所有图片变成数据集

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
class ImageDataset(Dataset):
    def __init__(self, path, transform):
        self.path = path
        self.transform = transform
        self.txt_path = self.path + '/' + 'Dataset.txt'
        f = open(self.txt_path, 'r')
        data = f.readlines()
        images = []
        labels = []

        for item in data:
            # 去掉空格，并组成数组
            word = item.split()
            images.append(os.path.join(self.path, word[2], word[0]))
            labels.append(int(word[1]))

        # 张量化
        labels = torch.LongTensor(labels)
        self.images = images
        self.labels = labels

    def __getitem__(self, index):
        image = self.images[index]
        label = self.labels[index]
        image = Image.open(image).convert('RGB')
        image = self.transform(image)
        return image, label

    def __len__(self):
        return len(self.labels)

初始化数据集，并切割

1
2
3
4
5
6
7
8
def get_dataset(path, transform):
    tic = time.time()
    print('Loading Dataset...')
    dataset = ImageDataset(path, transform)
    train, test = train_test_split(dataset, test_size=0.2)
    toc = time.time()
    print('Loading Over, Cost Time: ' + str(toc - tic) + 's')
    return train, test

MODEL

定义模型

不负责的定义

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
class Model(nn.Module):
    # ValueError:optimizer got an empty parameter list
    # init写错（int），没有__，没有super都会报错
    def __init__(self):
        super(Model, self).__init__()
        # 卷积层
        # original 3 x 256 x 256
        # conv1 -> 8 x 256 x 256
        self.conv1 = nn.Conv2d(in_channels=3, out_channels=8, kernel_size=3, stride=1, padding=1)
        # pool1 -> 8 x 128 x 128
        self.pool1 = nn.MaxPool2d(2, 2)

        # conv2 -> 16 x 128 x 128
        self.conv2 = nn.Conv2d(in_channels=8, out_channels=16, kernel_size=3, stride=1, padding=1)
        # pool2 -> 16 x 64 x 64
        self.pool2 = nn.MaxPool2d(2, 2)

        # conv3 -> 32 x 64 x 64
        self.conv3 = nn.Conv2d(in_channels=16, out_channels=32, kernel_size=3, stride=1, padding=1)
        # pool3 -> 32 x 32 x 32
        self.pool3 = nn.MaxPool2d(2, 2)

        # 全连接层
        self.fc1 = nn.Linear(32 * 32 * 32, 256)
        self.fc2 = nn.Linear(256, 64)
        self.fc3 = nn.Linear(64, 2)

    def forward(self, x):
        x = self.conv1(x)
        x = nn.functional.relu(x)
        x = self.pool1(x)

        x = self.conv2(x)
        x = nn.functional.relu(x)
        x = self.pool2(x)

        x = self.conv3(x)
        x = nn.functional.relu(x)
        x = self.pool3(x)

        # x = x.view(32 * 32 * 32, -1)
        # mat1 and mat2 shapes cannot be multiplied (32768x50 and 32768x256)
        # batch_size = 50
        # 要改成 50x32768 and 32768x256
        x = x.view(-1, 32 * 32 * 32)
        x = self.fc1(x)
        x = nn.functional.relu(x)

        x = self.fc2(x)
        x = nn.functional.relu(x)

        x = self.fc3(x)
        x = nn.functional.softmax(x, dim=1)

        return x

注意两个点：

  1. init 写错（int），没有__，没有 super 都会报错 ValueError:optimizer got an empty parameter list；
  2. x = x.view(32 * 32 * 32, -1) 会报错 mat1 and mat2 shapes cannot be multiplied (32768x50 and 32768x256)，要行对列。

训练模型

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
def get_model(train, test, batch_size, gpu, lr, epochs):
    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
    model = Model()
    # SGD
    optimizer = torch.optim.Adam(model.parameters(), lr=lr)
    # RuntimeError: Boolean value of Tensor with more than one value is ambiguous
    # 初始化的时候，后面加个()
    loss_func = nn.CrossEntropyLoss()

    if gpu:
        # Input type (torch.FloatTensor) and weight type (torch.cuda.FloatTensor) should be the same
        # 要model = model.to(device)，不能写model.to(device)
        model = model.to(device)
        loss_func = loss_func.to(device)

    train_dataloader = DataLoader(train,
                                  batch_size=batch_size,
                                  shuffle=True,
                                  num_workers=0,
                                  pin_memory=True)

    test_dataloader = DataLoader(test,
                                 batch_size=batch_size,
                                 shuffle=True,
                                 num_workers=0,
                                 pin_memory=True)

    for epoch in range(epochs):
        tic = time.time()
        print('Training')
        for step, (train_images, train_labels) in enumerate(train_dataloader):

            if gpu:
                train_images = train_images.to(device)
                train_labels = train_labels.to(device)

            prediction = model(train_images)
            loss = loss_func(prediction, train_labels)
            optimizer.zero_grad()
            loss.backward()
            optimizer.step()
            # print('Epoch: ', epoch + 1, '| Step: ', step + 1, '| Loss: ', loss.item())

        toc = time.time()
        print('Train', epoch + 1, 'Over')
        print('Train time:' + str(toc - tic) + 's')
        print('Test Dataset Loading...')

        correct = 0
        for step, (test_images, test_labels) in enumerate(test_dataloader):

            if gpu:
                test_images = test_images.to(device)
                test_labels = test_labels.to(device)

            result = model(test_images)
            prediction = torch.argmax(result, 1)
            correct += (prediction == test_labels).sum().item()
        accuracy = correct / len(test_dataset)
        print('Test Dataset:', len(test_dataset), '| Correct:', correct, '| Accuracy:', accuracy)
        print('')
        time.sleep(3)

    return model

整体代码

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
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
import time
import os
import torch
import torch.nn as nn
from torch.utils.data import Dataset, DataLoader
from torchvision import transforms
from sklearn.model_selection import train_test_split
from PIL import Image


class ImageDataset(Dataset):
    def __init__(self, path, transform):
        self.path = path
        self.transform = transform
        self.txt_path = self.path + '/' + 'Dataset.txt'
        f = open(self.txt_path, 'r')
        data = f.readlines()
        images = []
        labels = []

        for item in data:
            # 去掉空格，并组成数组
            word = item.split()
            images.append(os.path.join(self.path, word[2], word[0]))
            labels.append(int(word[1]))

        # 张量化
        labels = torch.LongTensor(labels)
        self.images = images
        self.labels = labels

    def __getitem__(self, index):
        image = self.images[index]
        label = self.labels[index]
        image = Image.open(image).convert('RGB')
        image = self.transform(image)
        return image, label

    def __len__(self):
        return len(self.labels)


class Model(nn.Module):
    # ValueError:optimizer got an empty parameter list
    # init写错（int），没有__，没有super都会报错
    def __init__(self):
        super(Model, self).__init__()
        # 卷积层
        # original 3 x 256 x 256
        # conv1 -> 8 x 256 x 256
        self.conv1 = nn.Conv2d(in_channels=3, out_channels=8, kernel_size=3, stride=1, padding=1)
        # pool1 -> 8 x 128 x 128
        self.pool1 = nn.MaxPool2d(2, 2)

        # conv2 -> 16 x 128 x 128
        self.conv2 = nn.Conv2d(in_channels=8, out_channels=16, kernel_size=3, stride=1, padding=1)
        # pool2 -> 16 x 64 x 64
        self.pool2 = nn.MaxPool2d(2, 2)

        # conv3 -> 32 x 64 x 64
        self.conv3 = nn.Conv2d(in_channels=16, out_channels=32, kernel_size=3, stride=1, padding=1)
        # pool3 -> 32 x 32 x 32
        self.pool3 = nn.MaxPool2d(2, 2)

        # 全连接层
        self.fc1 = nn.Linear(32 * 32 * 32, 256)
        self.fc2 = nn.Linear(256, 64)
        self.fc3 = nn.Linear(64, 2)

    def forward(self, x):
        x = self.conv1(x)
        x = nn.functional.relu(x)
        x = self.pool1(x)

        x = self.conv2(x)
        x = nn.functional.relu(x)
        x = self.pool2(x)

        x = self.conv3(x)
        x = nn.functional.relu(x)
        x = self.pool3(x)

        # x = x.view(32 * 32 * 32, -1)
        # mat1 and mat2 shapes cannot be multiplied (32768x50 and 32768x256)
        # batch_size = 50
        # 要改成 50x32768 and 32768x256
        x = x.view(-1, 32 * 32 * 32)
        x = self.fc1(x)
        x = nn.functional.relu(x)

        x = self.fc2(x)
        x = nn.functional.relu(x)

        x = self.fc3(x)
        x = nn.functional.softmax(x, dim=1)

        return x


def get_txt(path, file_name, label):
    # path：dataset的根目录
    # file_path：dataset文件夹
    file_path = os.path.join(path, file_name)
    data = os.listdir(file_path)
    # 'w' is mean write
    f = open(path + '/' + file_name + 'Data.txt', 'w')
    for item in data:
        f.write(item + ' ' + str(label) + ' ' + file_name + '\n')
    f.close()


def mix_txt(path, file1_name, file2_name):
    f = open(path + '/' + 'DataSet.txt', 'w')
    # 'r' is mean read
    f1 = open(path + '/' + file1_name + 'Data.txt', 'r')
    # readline/readlines
    data1 = f1.readlines()
    # PC 上生成的 txt 有 DStore 信息。合并 txt 时需要只读取第一行之后的内容
    # data1 = data1[1:]
    for item in data1:
        f.write(item)

    f2 = open(path + '/' + file2_name + 'Data.txt', 'r')
    data2 = f2.readlines()
    # data2 = data2[1:]
    for item in data2:
        f.write(item)

    f.close()
    f1.close()
    f2.close()


def get_dataset(path, transform):
    tic = time.time()
    print('Loading Dataset...')
    dataset = ImageDataset(path, transform)
    train, test = train_test_split(dataset, test_size=0.2)
    toc = time.time()
    print('Loading Over, Cost Time: ' + str(toc - tic) + 's')
    return train, test


def get_model(train, test, batch_size, gpu, lr, epochs):
    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
    model = Model()
    # SGD
    optimizer = torch.optim.Adam(model.parameters(), lr=lr)
    # RuntimeError: Boolean value of Tensor with more than one value is ambiguous
    # 初始化的时候，后面加个()
    loss_func = nn.CrossEntropyLoss()

    if gpu:
        # Input type (torch.FloatTensor) and weight type (torch.cuda.FloatTensor) should be the same
        # 要model = model.to(device)，不能写model.to(device)
        model = model.to(device)
        loss_func = loss_func.to(device)

    train_dataloader = DataLoader(train,
                                  batch_size=batch_size,
                                  shuffle=True,
                                  num_workers=0,
                                  pin_memory=True)

    test_dataloader = DataLoader(test,
                                 batch_size=batch_size,
                                 shuffle=True,
                                 num_workers=0,
                                 pin_memory=True)

    for epoch in range(epochs):
        tic = time.time()
        print('Training')
        for step, (train_images, train_labels) in enumerate(train_dataloader):

            if gpu:
                train_images = train_images.to(device)
                train_labels = train_labels.to(device)

            prediction = model(train_images)
            loss = loss_func(prediction, train_labels)
            optimizer.zero_grad()
            loss.backward()
            optimizer.step()
            # print('Epoch: ', epoch + 1, '| Step: ', step + 1, '| Loss: ', loss.item())

        toc = time.time()
        print('Train', epoch + 1, 'Over')
        print('Train time:' + str(toc - tic) + 's')
        print('Test Dataset Loading...')

        correct = 0
        for step, (test_images, test_labels) in enumerate(test_dataloader):

            if gpu:
                test_images = test_images.to(device)
                test_labels = test_labels.to(device)

            result = model(test_images)
            prediction = torch.argmax(result, 1)
            correct += (prediction == test_labels).sum().item()
        accuracy = correct / len(test_dataset)
        print('Test Dataset:', len(test_dataset), '| Correct:', correct, '| Accuracy:', accuracy)
        print('')
        time.sleep(3)

    return model


if __name__ == '__main__':
    # root_path = 'E:/Mac/PetImages'
    root_path = '/Users/bakako/Downloads/PetImages'
    get_txt(root_path, file_name='Cat', label=0)
    get_txt(root_path, file_name='Dog', label=1)
    mix_txt(root_path, file1_name='Cat', file2_name='Dog')
    custom_transform = transforms.Compose([
        transforms.Resize(288),
        transforms.CenterCrop(256),
        transforms.ToTensor()
    ])

    # batch_size
    # https://blog.51cto.com/u_15127688/4678771
    # 这相比于全数据集训练，相当于是在寻找最优时人为增加了一些随机噪声，来修正由局部数据得到的梯度，尽量避免因 batch size 过大陷入局部最优
    # 吃内存

    train_dataset, test_dataset = get_dataset(root_path, custom_transform)
    cat_vs_dog_model = get_model(train=train_dataset, test=test_dataset,
                                 batch_size=25,
                                 gpu=False,
                                 lr=0.0001,
                                 epochs=30)


随笔
#PyTorch
PyTorch 猫狗识别深度学习 I 低准确度版本
https://wonderhoi.com/2023/08/07/PyTorch-猫狗识别深度学习-I-低准确度版本/
作者
wonderhoi
发布于
2023年8月7日
许可协议