【PyTorch】Resnet/深度残差网络_loss: 0.5167 -

1 模型描述

深度残差网络（Resnet）是由来自Microsoft Research的4位学者（Kaiming He等人）提出的卷积神经网络，在2015年的ImageNet大规模视觉识别竞赛（ImageNet Large Scale Visual Recognition Challenge, ILSVRC）中获得了图像分类和物体识别的优胜。 残差网络的特点是容易优化，并且能够通过增加相当的深度来提高准确率。其内部的残差块使用了跳跃连接，缓解了在深度神经网络中增加深度带来的梯度消失问题。

本文采用PyTorch来实现Resnet。

2 具体代码

# codes for implementing Resnet
# ---------------------------------------------------------------------------- #
# An implementation of https://arxiv.org/pdf/1512.03385.pdf                    #
# See section 4.2 for the model architecture on CIFAR-10                       #
# Some part of the code was referenced from below                              #
# https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py   #
# ---------------------------------------------------------------------------- #

import torch
import torch.nn as nn
import torchvision 
import torchvision.transforms as transforms

# Device configuration 
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

# Hyper-parameters
num_epochs = 80
batch_size = 100
learning_rate = 0.001

# image preprocessing modules
transform = transforms.Compose([
    transforms.Pad(4), # 填充
    transforms.RandomHorizontalFlip(),# 垂直翻转
    transforms.RandomCrop(32), # 随机裁剪
    transforms.ToTensor()])

# CIFAR-10 dataset
train_dataset = torchvision.datasets.CIFAR10(
    root = 'data/',
    train = True,
    transform=transform,
    download = True
    )
test_dataset = torchvision.datasets.CIFAR10(
    root = 'data/',
    train = False,
    transform = transforms.ToTensor()
)

# Data loader
train_loader = torch.utils.data.DataLoader(
    dataset = train_dataset,
    batch_size = batch_size,
    shuffle = True
)
test_loader = torch.utils.data.DataLoader(
    dataset = test_dataset,
    batch_size = batch_size,
    shuffle = False
)

# 3×3 convolution
def conv3x3(in_channels, out_channels, stride=1):
    return nn.Conv2d(
        in_channels, 
        out_channels, 
        kernel_size = 3, 
        stride = stride, 
        padding = 1,
        bias = False
        )

# Residual block
class ResidualBlock(nn.Module):
    def __init__(self, in_channels, out_channels, stride = 1, downsample = None):
        super(ResidualBlock, self).__init__()
        self.conv1 = conv3x3(in_channels, out_channels, stride)
        self.bn1 = nn.BatchNorm2d(out_channels)
        self.relu = nn.ReLU(inplace = True)
        self.conv2 = conv3x3(out_channels, out_channels)
        self.bn2 = nn.BatchNorm2d(out_channels)
        self.downsample = downsample
    
    def forward(self, x):
        residual = x
        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)
        out = self.conv2(out)
        out = self.bn2(out)
        if self.downsample:
            residual = self.downsample(x)
        out += residual
        out = self.relu(out)
        return out

# ResNet
class ResNet(nn.Module):
    def __init__(self, block, layers, num_classes=10):
        super(ResNet, self).__init__()
        self.in_channels = 16
        self.conv = conv3x3(3, 16)
        self.bn = nn.BatchNorm2d(16)
        self.relu = nn.ReLU(inplace=True)
        self.layer1 = self.make_layer(block, 16, layers[0])
        self.layer2 = self.make_layer(block, 32, layers[1], 2)
        self.layer3 = self.make_layer(block, 64, layers[2], 2)
        self.avg_pool = nn.AvgPool2d(8)
        self.fc = nn.Linear(64, num_classes)

    def make_layer(self, block, out_channels, blocks, stride=1):
        downsample = None
        if (stride != 1) or (self.in_channels != out_channels):
            downsample = nn.Sequential(
                conv3x3(self.in_channels, out_channels, stride=stride),
                nn.BatchNorm2d(out_channels))
        layers = []
        layers.append(block(self.in_channels, out_channels, stride, downsample))
        self.in_channels = out_channels
        for i in range(1, blocks):
            layers.append(block(out_channels, out_channels))
        return nn.Sequential(*layers)

    def forward(self, x):
        out = self.conv(x)
        out = self.bn(out)
        out = self.relu(out)
        out = self.layer1(out)
        out = self.layer2(out)
        out = self.layer3(out)
        out = self.avg_pool(out)
        out = out.view(out.size(0), -1)
        out = self.fc(out)
        return out

model = ResNet(ResidualBlock, [2, 2, 2]).to(device)

# Loss and optimizer
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)

# For updating learning rate
def update_lr(optimizer, lr):    
    for param_group in optimizer.param_groups:
        param_group['lr'] = lr

# Train the model
total_step = len(train_loader)
curr_lr = learning_rate
for epoch in range(num_epochs):
    for i, (images, labels) in enumerate(train_loader):
        images = images.to(device)
        labels = labels.to(device)
        
        # Forward pass
        outputs = model(images)
        loss = criterion(outputs, labels)
        
        # Backward and optimize
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
        
        if (i+1) % 100 == 0:
            print ("Epoch [{}/{}], Step [{}/{}] Loss: {:.4f}"
                   .format(epoch+1, num_epochs, i+1, total_step, loss.item()))
    # Decay learning rate
    if (epoch+1) % 20 == 0:
        curr_lr /= 3
        update_lr(optimizer, curr_lr)

# Test the model
model.eval()
with torch.no_grad():
    correct = 0
    total = 0
    for images, labels in test_loader:
        images = images.to(device)
        labels = labels.to(device)
        outputs = model(images)
        _, predicted = torch.max(outputs.data, 1)
        total += labels.size(0)
        correct += (predicted == labels).sum().item()

    print('Accuracy of the model on the test images: {} %'.format(100 * correct / total))

# Save the model checkpoint
torch.save(model.state_dict(), 'resnet.ckpt')
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

3 程序输出

程序训练时间较长，在笔记本上跑了5~6个小时，但结果还是是挺好的。

Epoch [1/80], Step [100/500] Loss: 1.6903
Epoch [1/80], Step [200/500] Loss: 1.3618
Epoch [1/80], Step [300/500] Loss: 1.1689
Epoch [1/80], Step [400/500] Loss: 1.1824
Epoch [1/80], Step [500/500] Loss: 1.0652
Epoch [2/80], Step [100/500] Loss: 1.1884
Epoch [2/80], Step [200/500] Loss: 0.8875
Epoch [2/80], Step [300/500] Loss: 1.2085
Epoch [2/80], Step [400/500] Loss: 0.9510
Epoch [2/80], Step [500/500] Loss: 0.9154
Epoch [3/80], Step [100/500] Loss: 0.9555
Epoch [3/80], Step [200/500] Loss: 0.9859
Epoch [3/80], Step [300/500] Loss: 0.8526
Epoch [3/80], Step [400/500] Loss: 0.9229
Epoch [3/80], Step [500/500] Loss: 0.7979
Epoch [4/80], Step [100/500] Loss: 0.7952
Epoch [4/80], Step [200/500] Loss: 0.7257
Epoch [4/80], Step [300/500] Loss: 0.7596
Epoch [4/80], Step [400/500] Loss: 0.7599
Epoch [4/80], Step [500/500] Loss: 0.7675
Epoch [5/80], Step [100/500] Loss: 0.6990
Epoch [5/80], Step [200/500] Loss: 0.7193
Epoch [5/80], Step [300/500] Loss: 0.6661
Epoch [5/80], Step [400/500] Loss: 0.4435
Epoch [5/80], Step [500/500] Loss: 0.7103
Epoch [6/80], Step [100/500] Loss: 0.5538
Epoch [6/80], Step [200/500] Loss: 0.6042
Epoch [6/80], Step [300/500] Loss: 0.5002
Epoch [6/80], Step [400/500] Loss: 0.7440
Epoch [6/80], Step [500/500] Loss: 0.5007
Epoch [7/80], Step [100/500] Loss: 0.4111
Epoch [7/80], Step [200/500] Loss: 0.7357
Epoch [7/80], Step [300/500] Loss: 0.5855
Epoch [7/80], Step [400/500] Loss: 0.4054
Epoch [7/80], Step [500/500] Loss: 0.5780
Epoch [8/80], Step [100/500] Loss: 0.4854
Epoch [8/80], Step [200/500] Loss: 0.5895
Epoch [8/80], Step [300/500] Loss: 0.6393
Epoch [8/80], Step [400/500] Loss: 0.5335
Epoch [8/80], Step [500/500] Loss: 0.5157
Epoch [9/80], Step [100/500] Loss: 0.7645
Epoch [9/80], Step [200/500] Loss: 0.4416
Epoch [9/80], Step [300/500] Loss: 0.2846
Epoch [9/80], Step [400/500] Loss: 0.5031
Epoch [9/80], Step [500/500] Loss: 0.4851
Epoch [10/80], Step [100/500] Loss: 0.4381
Epoch [10/80], Step [200/500] Loss: 0.5167
Epoch [10/80], Step [300/500] Loss: 0.4316
Epoch [10/80], Step [400/500] Loss: 0.4434
Epoch [10/80], Step [500/500] Loss: 0.2888
Epoch [11/80], Step [100/500] Loss: 0.4379
Epoch [11/80], Step [200/500] Loss: 0.5113
Epoch [11/80], Step [300/500] Loss: 0.5018
Epoch [11/80], Step [400/500] Loss: 0.5127
Epoch [11/80], Step [500/500] Loss: 0.4458
Epoch [12/80], Step [100/500] Loss: 0.4463
Epoch [12/80], Step [200/500] Loss: 0.5314
Epoch [12/80], Step [300/500] Loss: 0.3969
Epoch [12/80], Step [400/500] Loss: 0.4774
Epoch [12/80], Step [500/500] Loss: 0.7450
Epoch [13/80], Step [100/500] Loss: 0.3637
Epoch [13/80], Step [200/500] Loss: 0.5757
Epoch [13/80], Step [300/500] Loss: 0.3526
Epoch [13/80], Step [400/500] Loss: 0.4996
Epoch [13/80], Step [500/500] Loss: 0.4953
Epoch [14/80], Step [100/500] Loss: 0.4351
Epoch [14/80], Step [200/500] Loss: 0.4416
Epoch [14/80], Step [300/500] Loss: 0.4317
Epoch [14/80], Step [400/500] Loss: 0.5490
Epoch [14/80], Step [500/500] Loss: 0.3177
Epoch [15/80], Step [100/500] Loss: 0.4292
Epoch [15/80], Step [200/500] Loss: 0.4585
Epoch [15/80], Step [300/500] Loss: 0.4706
Epoch [15/80], Step [400/500] Loss: 0.4133
Epoch [15/80], Step [500/500] Loss: 0.4701
Epoch [16/80], Step [100/500] Loss: 0.3542
Epoch [16/80], Step [200/500] Loss: 0.3555
Epoch [16/80], Step [300/500] Loss: 0.4832
Epoch [16/80], Step [400/500] Loss: 0.4278
Epoch [16/80], Step [500/500] Loss: 0.3719
Epoch [17/80], Step [100/500] Loss: 0.3845
Epoch [17/80], Step [200/500] Loss: 0.3643
Epoch [17/80], Step [300/500] Loss: 0.3886
Epoch [17/80], Step [400/500] Loss: 0.4406
Epoch [17/80], Step [500/500] Loss: 0.3347
Epoch [18/80], Step [100/500] Loss: 0.3234
Epoch [18/80], Step [200/500] Loss: 0.3385
Epoch [18/80], Step [300/500] Loss: 0.2918
Epoch [18/80], Step [400/500] Loss: 0.3457
Epoch [18/80], Step [500/500] Loss: 0.4655
Epoch [19/80], Step [100/500] Loss: 0.2944
Epoch [19/80], Step [200/500] Loss: 0.2373
Epoch [19/80], Step [300/500] Loss: 0.4930
Epoch [19/80], Step [400/500] Loss: 0.5316
Epoch [19/80], Step [500/500] Loss: 0.5754
Epoch [20/80], Step [100/500] Loss: 0.3018
Epoch [20/80], Step [200/500] Loss: 0.2551
Epoch [20/80], Step [300/500] Loss: 0.3542
Epoch [20/80], Step [400/500] Loss: 0.3928
Epoch [20/80], Step [500/500] Loss: 0.4278
Epoch [21/80], Step [100/500] Loss: 0.1989
Epoch [21/80], Step [200/500] Loss: 0.2735
Epoch [21/80], Step [300/500] Loss: 0.3500
Epoch [21/80], Step [400/500] Loss: 0.2338
Epoch [21/80], Step [500/500] Loss: 0.3547
Epoch [22/80], Step [100/500] Loss: 0.4918
Epoch [22/80], Step [200/500] Loss: 0.3486
Epoch [22/80], Step [300/500] Loss: 0.2192
Epoch [22/80], Step [400/500] Loss: 0.2795
Epoch [22/80], Step [500/500] Loss: 0.2835
Epoch [23/80], Step [100/500] Loss: 0.1862
Epoch [23/80], Step [200/500] Loss: 0.2384
Epoch [23/80], Step [300/500] Loss: 0.2469
Epoch [23/80], Step [400/500] Loss: 0.2181
Epoch [23/80], Step [500/500] Loss: 0.1682
Epoch [24/80], Step [100/500] Loss: 0.2193
Epoch [24/80], Step [200/500] Loss: 0.2292
Epoch [24/80], Step [300/500] Loss: 0.3381
Epoch [24/80], Step [400/500] Loss: 0.3113
Epoch [24/80], Step [500/500] Loss: 0.3083
Epoch [25/80], Step [100/500] Loss: 0.3507
Epoch [25/80], Step [200/500] Loss: 0.2731
Epoch [25/80], Step [300/500] Loss: 0.2989
Epoch [25/80], Step [400/500] Loss: 0.2334
Epoch [25/80], Step [500/500] Loss: 0.2419
Epoch [26/80], Step [100/500] Loss: 0.3797
Epoch [26/80], Step [200/500] Loss: 0.1951
Epoch [26/80], Step [300/500] Loss: 0.2815
Epoch [26/80], Step [400/500] Loss: 0.2826
Epoch [26/80], Step [500/500] Loss: 0.2990
Epoch [27/80], Step [100/500] Loss: 0.3639
Epoch [27/80], Step [200/500] Loss: 0.2609
Epoch [27/80], Step [300/500] Loss: 0.2676
Epoch [27/80], Step [400/500] Loss: 0.1734
Epoch [27/80], Step [500/500] Loss: 0.3016
Epoch [28/80], Step [100/500] Loss: 0.2848
Epoch [28/80], Step [200/500] Loss: 0.2427
Epoch [28/80], Step [300/500] Loss: 0.1968
Epoch [28/80], Step [400/500] Loss: 0.1952
Epoch [28/80], Step [500/500] Loss: 0.1763
Epoch [29/80], Step [100/500] Loss: 0.2007
Epoch [29/80], Step [200/500] Loss: 0.1410
Epoch [29/80], Step [300/500] Loss: 0.4305
Epoch [29/80], Step [400/500] Loss: 0.2089
Epoch [29/80], Step [500/500] Loss: 0.2162
Epoch [30/80], Step [100/500] Loss: 0.1495
Epoch [30/80], Step [200/500] Loss: 0.3787
Epoch [30/80], Step [300/500] Loss: 0.1756
Epoch [30/80], Step [400/500] Loss: 0.3060
Epoch [30/80], Step [500/500] Loss: 0.2241
Epoch [31/80], Step [100/500] Loss: 0.1697
Epoch [31/80], Step [200/500] Loss: 0.1628
Epoch [31/80], Step [300/500] Loss: 0.3507
Epoch [31/80], Step [400/500] Loss: 0.2836
Epoch [31/80], Step [500/500] Loss: 0.2011
Epoch [32/80], Step [100/500] Loss: 0.2271
Epoch [32/80], Step [200/500] Loss: 0.1603
Epoch [32/80], Step [300/500] Loss: 0.2815
Epoch [32/80], Step [400/500] Loss: 0.2119
Epoch [32/80], Step [500/500] Loss: 0.3992
Epoch [33/80], Step [100/500] Loss: 0.2674
Epoch [33/80], Step [200/500] Loss: 0.2192
Epoch [33/80], Step [300/500] Loss: 0.3333
Epoch [33/80], Step [400/500] Loss: 0.3045
Epoch [33/80], Step [500/500] Loss: 0.2048
Epoch [34/80], Step [100/500] Loss: 0.3307
Epoch [34/80], Step [200/500] Loss: 0.2005
Epoch [34/80], Step [300/500] Loss: 0.1421
Epoch [34/80], Step [400/500] Loss: 0.3129
Epoch [34/80], Step [500/500] Loss: 0.2223
Epoch [35/80], Step [100/500] Loss: 0.1297
Epoch [35/80], Step [200/500] Loss: 0.2528
Epoch [35/80], Step [300/500] Loss: 0.1897
Epoch [35/80], Step [400/500] Loss: 0.1646
Epoch [35/80], Step [500/500] Loss: 0.2214
Epoch [36/80], Step [100/500] Loss: 0.2536
Epoch [36/80], Step [200/500] Loss: 0.2118
Epoch [36/80], Step [300/500] Loss: 0.2619
Epoch [36/80], Step [400/500] Loss: 0.2337
Epoch [36/80], Step [500/500] Loss: 0.2445
Epoch [37/80], Step [100/500] Loss: 0.2530
Epoch [37/80], Step [200/500] Loss: 0.3891
Epoch [37/80], Step [300/500] Loss: 0.1512
Epoch [37/80], Step [400/500] Loss: 0.1894
Epoch [37/80], Step [500/500] Loss: 0.1827
Epoch [38/80], Step [100/500] Loss: 0.2923
Epoch [38/80], Step [200/500] Loss: 0.1250
Epoch [38/80], Step [300/500] Loss: 0.1928
Epoch [38/80], Step [400/500] Loss: 0.1417
Epoch [38/80], Step [500/500] Loss: 0.3942
Epoch [39/80], Step [100/500] Loss: 0.2499
Epoch [39/80], Step [200/500] Loss: 0.2135
Epoch [39/80], Step [300/500] Loss: 0.1442
Epoch [39/80], Step [400/500] Loss: 0.1517
Epoch [39/80], Step [500/500] Loss: 0.1288
Epoch [40/80], Step [100/500] Loss: 0.2396
Epoch [40/80], Step [200/500] Loss: 0.1906
Epoch [40/80], Step [300/500] Loss: 0.2148
Epoch [40/80], Step [400/500] Loss: 0.1330
Epoch [40/80], Step [500/500] Loss: 0.2699
Epoch [41/80], Step [100/500] Loss: 0.2349
Epoch [41/80], Step [200/500] Loss: 0.2214
Epoch [41/80], Step [300/500] Loss: 0.2116
Epoch [41/80], Step [400/500] Loss: 0.1685
Epoch [41/80], Step [500/500] Loss: 0.1872
Epoch [42/80], Step [100/500] Loss: 0.2569
Epoch [42/80], Step [200/500] Loss: 0.2714
Epoch [42/80], Step [300/500] Loss: 0.2066
Epoch [42/80], Step [400/500] Loss: 0.1444
Epoch [42/80], Step [500/500] Loss: 0.0552
Epoch [43/80], Step [100/500] Loss: 0.1116
Epoch [43/80], Step [200/500] Loss: 0.1501
Epoch [43/80], Step [300/500] Loss: 0.1773
Epoch [43/80], Step [400/500] Loss: 0.2420
Epoch [43/80], Step [500/500] Loss: 0.1612
Epoch [44/80], Step [100/500] Loss: 0.1579
Epoch [44/80], Step [200/500] Loss: 0.1696
Epoch [44/80], Step [300/500] Loss: 0.1036
Epoch [44/80], Step [400/500] Loss: 0.1548
Epoch [44/80], Step [500/500] Loss: 0.1066
Epoch [45/80], Step [100/500] Loss: 0.1976
Epoch [45/80], Step [200/500] Loss: 0.2174
Epoch [45/80], Step [300/500] Loss: 0.1403
Epoch [45/80], Step [400/500] Loss: 0.1959
Epoch [45/80], Step [500/500] Loss: 0.1865
Epoch [46/80], Step [100/500] Loss: 0.2320
Epoch [46/80], Step [200/500] Loss: 0.1705
Epoch [46/80], Step [300/500] Loss: 0.2263
Epoch [46/80], Step [400/500] Loss: 0.2542
Epoch [46/80], Step [500/500] Loss: 0.1033
Epoch [47/80], Step [100/500] Loss: 0.2302
Epoch [47/80], Step [200/500] Loss: 0.1596
Epoch [47/80], Step [300/500] Loss: 0.1005
Epoch [47/80], Step [400/500] Loss: 0.1006
Epoch [47/80], Step [500/500] Loss: 0.1116
Epoch [48/80], Step [100/500] Loss: 0.1319
Epoch [48/80], Step [200/500] Loss: 0.1520
Epoch [48/80], Step [300/500] Loss: 0.2459
Epoch [48/80], Step [400/500] Loss: 0.1722
Epoch [48/80], Step [500/500] Loss: 0.0748
Epoch [49/80], Step [100/500] Loss: 0.1780
Epoch [49/80], Step [200/500] Loss: 0.2386
Epoch [49/80], Step [300/500] Loss: 0.1755
Epoch [49/80], Step [400/500] Loss: 0.1421
Epoch [49/80], Step [500/500] Loss: 0.1742
Epoch [50/80], Step [100/500] Loss: 0.1197
Epoch [50/80], Step [200/500] Loss: 0.1366
Epoch [50/80], Step [300/500] Loss: 0.1197
Epoch [50/80], Step [400/500] Loss: 0.0538
Epoch [50/80], Step [500/500] Loss: 0.1342
Epoch [51/80], Step [100/500] Loss: 0.2153
Epoch [51/80], Step [200/500] Loss: 0.2324
Epoch [51/80], Step [300/500] Loss: 0.2810
Epoch [51/80], Step [400/500] Loss: 0.1406
Epoch [51/80], Step [500/500] Loss: 0.1157
Epoch [52/80], Step [100/500] Loss: 0.2096
Epoch [52/80], Step [200/500] Loss: 0.2195
Epoch [52/80], Step [300/500] Loss: 0.1312
Epoch [52/80], Step [400/500] Loss: 0.0810
Epoch [52/80], Step [500/500] Loss: 0.1606
Epoch [53/80], Step [100/500] Loss: 0.1212
Epoch [53/80], Step [200/500] Loss: 0.1205
Epoch [53/80], Step [300/500] Loss: 0.1437
Epoch [53/80], Step [400/500] Loss: 0.3666
Epoch [53/80], Step [500/500] Loss: 0.2045
Epoch [54/80], Step [100/500] Loss: 0.1223
Epoch [54/80], Step [200/500] Loss: 0.1315
Epoch [54/80], Step [300/500] Loss: 0.1991
Epoch [54/80], Step [400/500] Loss: 0.2255
Epoch [54/80], Step [500/500] Loss: 0.1454
Epoch [55/80], Step [100/500] Loss: 0.1639
Epoch [55/80], Step [200/500] Loss: 0.1156
Epoch [55/80], Step [300/500] Loss: 0.1618
Epoch [55/80], Step [400/500] Loss: 0.1641
Epoch [55/80], Step [500/500] Loss: 0.2260
Epoch [56/80], Step [100/500] Loss: 0.2281
Epoch [56/80], Step [200/500] Loss: 0.1662
Epoch [56/80], Step [300/500] Loss: 0.2069
Epoch [56/80], Step [400/500] Loss: 0.0737
Epoch [56/80], Step [500/500] Loss: 0.1835
Epoch [57/80], Step [100/500] Loss: 0.1179
Epoch [57/80], Step [200/500] Loss: 0.1045
Epoch [57/80], Step [300/500] Loss: 0.1651
Epoch [57/80], Step [400/500] Loss: 0.0943
Epoch [57/80], Step [500/500] Loss: 0.1596
Epoch [58/80], Step [100/500] Loss: 0.0833
Epoch [58/80], Step [200/500] Loss: 0.2477
Epoch [58/80], Step [300/500] Loss: 0.2278
Epoch [58/80], Step [400/500] Loss: 0.1701
Epoch [58/80], Step [500/500] Loss: 0.1955
Epoch [59/80], Step [100/500] Loss: 0.1032
Epoch [59/80], Step [200/500] Loss: 0.1117
Epoch [59/80], Step [300/500] Loss: 0.1026
Epoch [59/80], Step [400/500] Loss: 0.1879
Epoch [59/80], Step [500/500] Loss: 0.1909
Epoch [60/80], Step [100/500] Loss: 0.1473
Epoch [60/80], Step [200/500] Loss: 0.1257
Epoch [60/80], Step [300/500] Loss: 0.1188
Epoch [60/80], Step [400/500] Loss: 0.1970
Epoch [60/80], Step [500/500] Loss: 0.1073
Epoch [61/80], Step [100/500] Loss: 0.0380
Epoch [61/80], Step [200/500] Loss: 0.1633
Epoch [61/80], Step [300/500] Loss: 0.1887
Epoch [61/80], Step [400/500] Loss: 0.1534
Epoch [61/80], Step [500/500] Loss: 0.1583
Epoch [62/80], Step [100/500] Loss: 0.1815
Epoch [62/80], Step [200/500] Loss: 0.2359
Epoch [62/80], Step [300/500] Loss: 0.0880
Epoch [62/80], Step [400/500] Loss: 0.1913
Epoch [62/80], Step [500/500] Loss: 0.1737
Epoch [63/80], Step [100/500] Loss: 0.2746
Epoch [63/80], Step [200/500] Loss: 0.2172
Epoch [63/80], Step [300/500] Loss: 0.1420
Epoch [63/80], Step [400/500] Loss: 0.0832
Epoch [63/80], Step [500/500] Loss: 0.1406
Epoch [64/80], Step [100/500] Loss: 0.1082
Epoch [64/80], Step [200/500] Loss: 0.2194
Epoch [64/80], Step [300/500] Loss: 0.1298
Epoch [64/80], Step [400/500] Loss: 0.1292
Epoch [64/80], Step [500/500] Loss: 0.2240
Epoch [65/80], Step [100/500] Loss: 0.0852
Epoch [65/80], Step [200/500] Loss: 0.1746
Epoch [65/80], Step [300/500] Loss: 0.1457
Epoch [65/80], Step [400/500] Loss: 0.1916
Epoch [65/80], Step [500/500] Loss: 0.1710
Epoch [66/80], Step [100/500] Loss: 0.1483
Epoch [66/80], Step [200/500] Loss: 0.1403
Epoch [66/80], Step [300/500] Loss: 0.1248
Epoch [66/80], Step [400/500] Loss: 0.2174
Epoch [66/80], Step [500/500] Loss: 0.1166
Epoch [67/80], Step [100/500] Loss: 0.1912
Epoch [67/80], Step [200/500] Loss: 0.1391
Epoch [67/80], Step [300/500] Loss: 0.1057
Epoch [67/80], Step [400/500] Loss: 0.2443
Epoch [67/80], Step [500/500] Loss: 0.0956
Epoch [68/80], Step [100/500] Loss: 0.1503
Epoch [68/80], Step [200/500] Loss: 0.1808
Epoch [68/80], Step [300/500] Loss: 0.1577
Epoch [68/80], Step [400/500] Loss: 0.1655
Epoch [68/80], Step [500/500] Loss: 0.1433
Epoch [69/80], Step [100/500] Loss: 0.0869
Epoch [69/80], Step [200/500] Loss: 0.0797
Epoch [69/80], Step [300/500] Loss: 0.0691
Epoch [69/80], Step [400/500] Loss: 0.2408
Epoch [69/80], Step [500/500] Loss: 0.2299
Epoch [70/80], Step [100/500] Loss: 0.2348
Epoch [70/80], Step [200/500] Loss: 0.0870
Epoch [70/80], Step [300/500] Loss: 0.2738
Epoch [70/80], Step [400/500] Loss: 0.1694
Epoch [70/80], Step [500/500] Loss: 0.0701
Epoch [71/80], Step [100/500] Loss: 0.0986
Epoch [71/80], Step [200/500] Loss: 0.1484
Epoch [71/80], Step [300/500] Loss: 0.1326
Epoch [71/80], Step [400/500] Loss: 0.1700
Epoch [71/80], Step [500/500] Loss: 0.1414
Epoch [72/80], Step [100/500] Loss: 0.1369
Epoch [72/80], Step [200/500] Loss: 0.2127
Epoch [72/80], Step [300/500] Loss: 0.1270
Epoch [72/80], Step [400/500] Loss: 0.1060
Epoch [72/80], Step [500/500] Loss: 0.0910
Epoch [73/80], Step [100/500] Loss: 0.0716
Epoch [73/80], Step [200/500] Loss: 0.1834
Epoch [73/80], Step [300/500] Loss: 0.1633
Epoch [73/80], Step [400/500] Loss: 0.0962
Epoch [73/80], Step [500/500] Loss: 0.2256
Epoch [74/80], Step [100/500] Loss: 0.1159
Epoch [74/80], Step [200/500] Loss: 0.1332
Epoch [74/80], Step [300/500] Loss: 0.0759
Epoch [74/80], Step [400/500] Loss: 0.1058
Epoch [74/80], Step [500/500] Loss: 0.0650
Epoch [75/80], Step [100/500] Loss: 0.1064
Epoch [75/80], Step [200/500] Loss: 0.2050
Epoch [75/80], Step [300/500] Loss: 0.1925
Epoch [75/80], Step [400/500] Loss: 0.1693
Epoch [75/80], Step [500/500] Loss: 0.1059
Epoch [76/80], Step [100/500] Loss: 0.1471
Epoch [76/80], Step [200/500] Loss: 0.2312
Epoch [76/80], Step [300/500] Loss: 0.1288
Epoch [76/80], Step [400/500] Loss: 0.1261
Epoch [76/80], Step [500/500] Loss: 0.0943
Epoch [77/80], Step [100/500] Loss: 0.1281
Epoch [77/80], Step [200/500] Loss: 0.1067
Epoch [77/80], Step [300/500] Loss: 0.1063
Epoch [77/80], Step [400/500] Loss: 0.0741
Epoch [77/80], Step [500/500] Loss: 0.0722
Epoch [78/80], Step [100/500] Loss: 0.1122
Epoch [78/80], Step [200/500] Loss: 0.1469
Epoch [78/80], Step [300/500] Loss: 0.1075
Epoch [78/80], Step [400/500] Loss: 0.1998
Epoch [78/80], Step [500/500] Loss: 0.1542
Epoch [79/80], Step [100/500] Loss: 0.1831
Epoch [79/80], Step [200/500] Loss: 0.0942
Epoch [79/80], Step [300/500] Loss: 0.0833
Epoch [79/80], Step [400/500] Loss: 0.1147
Epoch [79/80], Step [500/500] Loss: 0.2378
Epoch [80/80], Step [100/500] Loss: 0.1526
Epoch [80/80], Step [200/500] Loss: 0.1237
Epoch [80/80], Step [300/500] Loss: 0.0743
Epoch [80/80], Step [400/500] Loss: 0.1042
Epoch [80/80], Step [500/500] Loss: 0.1178
Accuracy of the model on the test images: 88.14 %
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
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401

Loss随着训练次数增加的变换情况，可以看出loss下降还是很明显的，但到后面阶段，也趋于收敛。