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 = downsampledef forward(self, x):residual = xout = 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 += residualout = 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 = 16self.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 = Noneif (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_channelsfor 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 outmodel = 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 passoutputs = model(images)loss = criterion(outputs, labels)# Backward and optimizeoptimizer.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 rateif (epoch+1) % 20 == 0:curr_lr /= 3update_lr(optimizer, curr_lr)# Test the model
model.eval()
with torch.no_grad():correct = 0total = 0for 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')

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 %

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。

【PyTorch】Resnet/深度残差网络相关推荐

  1. 详解ResNet(深度残差网络)

    引言 深度残差网络(Deep residual network, ResNet)的提出是CNN图像史上的一件里程碑事件,让我们先看一下ResNet在ILSVRC和COCO 2015上的战绩: ResN ...

  2. ResNet深度残差网络

    ResNet 论文:"Deep Residual Learning for Image Recognition" 面临的问题 ResNet解决的是深度神经网络的"退化&q ...

  3. 7 Resnet深度残差网络实现102种花卉分类

    Resnet(Deep residual network, ResNet),深度残差神经网络,卷积神经网络历史在具有划时代意义的神经网络.与Alexnet和VGG不同的是,网络结构上就有很大的改变,在 ...

  4. ResNet(深度残差网络）

    参考:同济子豪兄

  5. 42_ResNet (深度残差网络)---学习笔记

    1.39.ResNet (深度残差网络) Why call Residual # -*- coding: UTF-8 -*-import torch.nn as nn import torch.nn. ...

  6. 通过深度残差网络ResNet进行图像分类（pytorch网络多网络集成配置）

    通过深度残差网络进行图像分类(pytorch网络多网络集成配置) 简介 本项目通过配置文件修改,实现pytorch的ResNet18, ResNet34, ResNet50, ResNet101, R ...

  7. 何恺明编年史之深度残差网络ResNet

    文章目录 前言 一.提出ResNet原因 二.深度残差模块 1.数学理论基础 2.深度网络结构 三.Pytorch代码实现 四.总结 前言 图像分类是计算机视觉任务的基石,在目标监测.图像分割等任务中 ...

  8. 【深度学习之ResNet】——深度残差网络—ResNet总结

    目录 论文名称:Deep Residual Learning for Image Recognition 摘要: 1.引言 2.为什么会提出ResNet残差网络呢? 3.深度残差网络结构学习(Deep ...

  9. 深度残差网络ResNet解析

    ResNet在2015年被提出,在ImageNet比赛classification任务上获得第一名,因为它"简单与实用"并存,之后很多方法都建立在ResNet50或者ResNet1 ...

最新文章

  1. sql注入，预编译和事务
  2. Linux_RHEL7_YUM
  3. python基础包括什么-Python基础题
  4. 信息学奥赛一本通 1982：【19CSPJ普及组】数字游戏
  5. The Network Adapter could not establish the connection解决
  6. java存放的位置_java数据类型的种类以及存放的位置
  7. 一些看起来有用但没用过的函数
  8. Explorer.exe报错故障解决一例
  9. wms仓库管理系统选型要注意哪几点?
  10. 【腾讯云技术沙龙预告】云端数据库的设计之美
  11. tcl android tv系统,TCL to launch motion-controlled Android TV first in China
  12. excel白屏未响应_Excel中出现无响应的解决方法
  13. 美通社企业新闻汇总 | 2019.1.16 | 微软与药店巨头沃博联达成战略合作，科勒中国第13家工厂落成...
  14. 全球37%手机或因芯片漏洞遭攻击、黑客利用微软漏洞窃取登录凭据｜11月25日全球网络安全热点
  15. C语言 求100--200之间的素数
  16. Fibonacci Sum（二项式求和）
  17. win10停止更新_烦人的Windows10更新，教你如何有效的关闭。
  18. linux 浏览器崩溃,Firefox DoS漏洞导致浏览器崩溃 影响到Windows操作系统
  19. python生成小学数学练习题
  20. Windows运行单个.bat文件或运行多个.bat文件

热门文章

  1. xunsearch mysql 索引_xunsearch增量索引改进版
  2. MAC M1 安装 Vue CLI
  3. Java工程师成神之路~
  4. 视觉笔记:产品设计速写_图标设计：速写Vs展示。 准备
  5. Android10定位服务的开启状态判断和申请
  6. 【安装Linux（巨详细步骤从VMware到安装Linux）】
  7. 正视自己，金无足赤——BCH不断发现与修正缺陷
  8. A、数据分析18种最常用的可视化分析模式之——对标分析
  9. Aseprite终极白嫖教程
  10. 你什么都没错，只是太老了。