Pytorch入门 2.2 deep residual network

最新推荐文章于 2023-09-23 08:28:06 发布
最新推荐文章于 2023-09-23 08:28:06 发布
阅读量415 收藏 1
点赞数
分类专栏： pytorch
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/qq_38682032/article/details/88831894
版权 
# ---------------------------------------------------------------------------- #
# 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  配置设备，GPU加速
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

# Hyper-parameters   超参数配置 ： epochs 和学习率
num_epochs = 80
learning_rate = 0.001

# Image preprocessing modules
transform = transforms.Compose([
    transforms.Pad(4),
    transforms.RandomHorizontalFlip(),
    transforms.RandomCrop(32),
    transforms.ToTensor()])

torchvision.transforms这个包中包含resize、crop等常见的data augmentation操作，基本上PyTorch中的data augmentation操作都可以通过该接口实现。该包主要包含两个脚本：transformas.py和functional.py，前者定义了各种data augmentation的类，在每个类中通过调用functional.py中对应的函数完成data augmentation操作。

有兴趣的可以对pytorch源码进行学习

class Compose(object):
    """Composes several transforms together.
    Args:
        transforms (list of ``Transform`` objects): list of transforms to compose.
    Example:
        >>> transforms.Compose([
        >>>     transforms.CenterCrop(10),
        >>>     transforms.ToTensor(),
        >>> ])
    """

    def __init__(self, transforms):
        self.transforms = transforms

    def __call__(self, img):
        for t in self.transforms:
            img = t(img)
        return img

    def __repr__(self):
        format_string = self.__class__.__name__ + '('
        for t in self.transforms:
            format_string += '\n'
            format_string += '    {0}'.format(t)
        format_string += '\n)'
        return format_string

回到ResNet 的代码


# 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=100, 
                                           shuffle=True)

test_loader = torch.utils.data.DataLoader(dataset=test_dataset,
                                          batch_size=100, 
                                          shuffle=False)

# 3x3 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')

第一次运行下载太慢了，手动下了cifar10的压缩包 ，然后放到 data文件夹里。

Using downloaded and verified file: …/…/data/cifar-10-python.tar.gz
Epoch [1/80], Step [100/500] Loss: 1.5907
Epoch [1/80], Step [200/500] Loss: 1.5505
Epoch [1/80], Step [300/500] Loss: 1.4105
Epoch [1/80], Step [400/500] Loss: 1.3551
Epoch [1/80], Step [500/500] Loss: 1.2090
Epoch [2/80], Step [100/500] Loss: 1.2112
Epoch [2/80], Step [200/500] Loss: 1.1780
Epoch [2/80], Step [300/500] Loss: 0.8728
Epoch [2/80], Step [400/500] Loss: 1.1701
Epoch [2/80], Step [500/500] Loss: 0.9331
Epoch [3/80], Step [100/500] Loss: 0.9620
Epoch [3/80], Step [200/500] Loss: 0.8221
Epoch [3/80], Step [300/500] Loss: 0.8338
Epoch [3/80], Step [400/500] Loss: 0.9130
Epoch [3/80], Step [500/500] Loss: 0.9109
Epoch [4/80], Step [100/500] Loss: 0.8154
Epoch [4/80], Step [200/500] Loss: 0.8370
Epoch [4/80], Step [300/500] Loss: 0.9625
Epoch [4/80], Step [400/500] Loss: 0.5545
Epoch [4/80], Step [500/500] Loss: 0.7641
Epoch [5/80], Step [100/500] Loss: 0.6688
Epoch [5/80], Step [200/500] Loss: 0.7438
Epoch [5/80], Step [300/500] Loss: 0.7748
Epoch [5/80], Step [400/500] Loss: 0.8090
Epoch [5/80], Step [500/500] Loss: 0.5770
Epoch [6/80], Step [100/500] Loss: 0.7093
Epoch [6/80], Step [200/500] Loss: 0.7038
Epoch [6/80], Step [300/500] Loss: 0.7001
Epoch [6/80], Step [400/500] Loss: 0.5775
Epoch [6/80], Step [500/500] Loss: 0.4946
Epoch [7/80], Step [100/500] Loss: 0.7791
Epoch [7/80], Step [200/500] Loss: 0.7824
Epoch [7/80], Step [300/500] Loss: 0.5866
Epoch [7/80], Step [400/500] Loss: 0.6588
Epoch [7/80], Step [500/500] Loss: 0.6818
Epoch [8/80], Step [100/500] Loss: 0.6786
Epoch [8/80], Step [200/500] Loss: 0.5140
Epoch [8/80], Step [300/500] Loss: 0.5762
Epoch [8/80], Step [400/500] Loss: 0.5273
Epoch [8/80], Step [500/500] Loss: 0.5237
Epoch [9/80], Step [100/500] Loss: 0.6247
Epoch [9/80], Step [200/500] Loss: 0.5715
Epoch [9/80], Step [300/500] Loss: 0.5152
Epoch [9/80], Step [400/500] Loss: 0.5199
Epoch [9/80], Step [500/500] Loss: 0.7008
Epoch [10/80], Step [100/500] Loss: 0.5383
Epoch [10/80], Step [200/500] Loss: 0.4883
Epoch [10/80], Step [300/500] Loss: 0.5236
Epoch [10/80], Step [400/500] Loss: 0.4917
Epoch [10/80], Step [500/500] Loss: 0.6096
Epoch [11/80], Step [100/500] Loss: 0.4984
Epoch [11/80], Step [200/500] Loss: 0.5861
Epoch [11/80], Step [300/500] Loss: 0.4360
Epoch [11/80], Step [400/500] Loss: 0.5420
Epoch [11/80], Step [500/500] Loss: 0.3293
Epoch [12/80], Step [100/500] Loss: 0.4317
Epoch [12/80], Step [200/500] Loss: 0.2819
Epoch [12/80], Step [300/500] Loss: 0.4769
Epoch [12/80], Step [400/500] Loss: 0.4246
Epoch [12/80], Step [500/500] Loss: 0.5061
Epoch [13/80], Step [100/500] Loss: 0.5079
Epoch [13/80], Step [200/500] Loss: 0.3586
Epoch [13/80], Step [300/500] Loss: 0.4981
Epoch [13/80], Step [400/500] Loss: 0.4251
Epoch [13/80], Step [500/500] Loss: 0.4372
Epoch [14/80], Step [100/500] Loss: 0.3423
Epoch [14/80], Step [200/500] Loss: 0.5756
Epoch [14/80], Step [300/500] Loss: 0.3429
Epoch [14/80], Step [400/500] Loss: 0.6109
Epoch [14/80], Step [500/500] Loss: 0.4580
Epoch [15/80], Step [100/500] Loss: 0.4947
Epoch [15/80], Step [200/500] Loss: 0.3373
Epoch [15/80], Step [300/500] Loss: 0.6069
Epoch [15/80], Step [400/500] Loss: 0.5885
Epoch [15/80], Step [500/500] Loss: 0.4452
Epoch [16/80], Step [100/500] Loss: 0.3119
Epoch [16/80], Step [200/500] Loss: 0.4524
Epoch [16/80], Step [300/500] Loss: 0.4564
Epoch [16/80], Step [400/500] Loss: 0.3127
Epoch [16/80], Step [500/500] Loss: 0.3666
Epoch [17/80], Step [100/500] Loss: 0.4241
Epoch [17/80], Step [200/500] Loss: 0.3478
Epoch [17/80], Step [300/500] Loss: 0.4995
Epoch [17/80], Step [400/500] Loss: 0.4871
Epoch [17/80], Step [500/500] Loss: 0.4783
Epoch [18/80], Step [100/500] Loss: 0.5313
Epoch [18/80], Step [200/500] Loss: 0.4338
Epoch [18/80], Step [300/500] Loss: 0.3254
Epoch [18/80], Step [400/500] Loss: 0.5164
Epoch [18/80], Step [500/500] Loss: 0.3072
Epoch [19/80], Step [100/500] Loss: 0.4340
Epoch [19/80], Step [200/500] Loss: 0.4137
Epoch [19/80], Step [300/500] Loss: 0.5225
Epoch [19/80], Step [400/500] Loss: 0.4492
Epoch [19/80], Step [500/500] Loss: 0.4146
Epoch [20/80], Step [100/500] Loss: 0.3948
Epoch [20/80], Step [200/500] Loss: 0.4375
Epoch [20/80], Step [300/500] Loss: 0.4595
Epoch [20/80], Step [400/500] Loss: 0.4922
Epoch [20/80], Step [500/500] Loss: 0.2847
Epoch [21/80], Step [100/500] Loss: 0.2246
Epoch [21/80], Step [200/500] Loss: 0.3152
Epoch [21/80], Step [300/500] Loss: 0.2093
Epoch [21/80], Step [400/500] Loss: 0.3289
Epoch [21/80], Step [500/500] Loss: 0.3332
Epoch [22/80], Step [100/500] Loss: 0.3782
Epoch [22/80], Step [200/500] Loss: 0.2125
Epoch [22/80], Step [300/500] Loss: 0.3163
Epoch [22/80], Step [400/500] Loss: 0.2698
Epoch [22/80], Step [500/500] Loss: 0.2988
Epoch [23/80], Step [100/500] Loss: 0.2566
Epoch [23/80], Step [200/500] Loss: 0.3730
Epoch [23/80], Step [300/500] Loss: 0.1468
Epoch [23/80], Step [400/500] Loss: 0.3783
Epoch [23/80], Step [500/500] Loss: 0.3516
Epoch [24/80], Step [100/500] Loss: 0.2937
Epoch [24/80], Step [200/500] Loss: 0.2234
Epoch [24/80], Step [300/500] Loss: 0.2574
Epoch [24/80], Step [400/500] Loss: 0.3449
Epoch [24/80], Step [500/500] Loss: 0.2348
Epoch [25/80], Step [100/500] Loss: 0.2861
Epoch [25/80], Step [200/500] Loss: 0.2639
Epoch [25/80], Step [300/500] Loss: 0.2189
Epoch [25/80], Step [400/500] Loss: 0.1530
Epoch [25/80], Step [500/500] Loss: 0.2656
Epoch [26/80], Step [100/500] Loss: 0.2503
Epoch [26/80], Step [200/500] Loss: 0.3219
Epoch [26/80], Step [300/500] Loss: 0.2858
Epoch [26/80], Step [400/500] Loss: 0.2349
Epoch [26/80], Step [500/500] Loss: 0.2122
Epoch [27/80], Step [100/500] Loss: 0.2785
Epoch [27/80], Step [200/500] Loss: 0.3244
Epoch [27/80], Step [300/500] Loss: 0.2613
Epoch [27/80], Step [400/500] Loss: 0.4803
Epoch [27/80], Step [500/500] Loss: 0.1870
Epoch [28/80], Step [100/500] Loss: 0.4138
Epoch [28/80], Step [200/500] Loss: 0.2766
Epoch [28/80], Step [300/500] Loss: 0.2770
Epoch [28/80], Step [400/500] Loss: 0.2199
Epoch [28/80], Step [500/500] Loss: 0.1671
Epoch [29/80], Step [100/500] Loss: 0.2494
Epoch [29/80], Step [200/500] Loss: 0.3742
Epoch [29/80], Step [300/500] Loss: 0.3162
Epoch [29/80], Step [400/500] Loss: 0.1809
Epoch [29/80], Step [500/500] Loss: 0.2658
Epoch [30/80], Step [100/500] Loss: 0.1920
Epoch [30/80], Step [200/500] Loss: 0.2074
Epoch [30/80], Step [300/500] Loss: 0.4392
Epoch [30/80], Step [400/500] Loss: 0.2384
Epoch [30/80], Step [500/500] Loss: 0.2517
Epoch [31/80], Step [100/500] Loss: 0.2161
Epoch [31/80], Step [200/500] Loss: 0.1608
Epoch [31/80], Step [300/500] Loss: 0.3108
Epoch [31/80], Step [400/500] Loss: 0.3299
Epoch [31/80], Step [500/500] Loss: 0.1993
Epoch [32/80], Step [100/500] Loss: 0.2255
Epoch [32/80], Step [200/500] Loss: 0.2504
Epoch [32/80], Step [300/500] Loss: 0.2334
Epoch [32/80], Step [400/500] Loss: 0.2160
Epoch [32/80], Step [500/500] Loss: 0.1846
Epoch [33/80], Step [100/500] Loss: 0.1758
Epoch [33/80], Step [200/500] Loss: 0.2508
Epoch [33/80], Step [300/500] Loss: 0.1809
Epoch [33/80], Step [400/500] Loss: 0.2126
Epoch [33/80], Step [500/500] Loss: 0.1541
Epoch [34/80], Step [100/500] Loss: 0.2740
Epoch [34/80], Step [200/500] Loss: 0.2125
Epoch [34/80], Step [300/500] Loss: 0.2800
Epoch [34/80], Step [400/500] Loss: 0.3252
Epoch [34/80], Step [500/500] Loss: 0.1676
Epoch [35/80], Step [100/500] Loss: 0.2941
Epoch [35/80], Step [200/500] Loss: 0.3275
Epoch [35/80], Step [300/500] Loss: 0.1462
Epoch [35/80], Step [400/500] Loss: 0.1550
Epoch [35/80], Step [500/500] Loss: 0.2294
Epoch [36/80], Step [100/500] Loss: 0.1557
Epoch [36/80], Step [200/500] Loss: 0.1995
Epoch [36/80], Step [300/500] Loss: 0.2436
Epoch [36/80], Step [400/500] Loss: 0.1950
Epoch [36/80], Step [500/500] Loss: 0.2539
Epoch [37/80], Step [100/500] Loss: 0.2753
Epoch [37/80], Step [200/500] Loss: 0.2348
Epoch [37/80], Step [300/500] Loss: 0.1384
Epoch [37/80], Step [400/500] Loss: 0.3876
Epoch [37/80], Step [500/500] Loss: 0.2565
Epoch [38/80], Step [100/500] Loss: 0.2558
Epoch [38/80], Step [200/500] Loss: 0.2371
Epoch [38/80], Step [300/500] Loss: 0.1932
Epoch [38/80], Step [400/500] Loss: 0.2206
Epoch [38/80], Step [500/500] Loss: 0.3186
Epoch [39/80], Step [100/500] Loss: 0.2072
Epoch [39/80], Step [200/500] Loss: 0.3059
Epoch [39/80], Step [300/500] Loss: 0.2730
Epoch [39/80], Step [400/500] Loss: 0.2736
Epoch [39/80], Step [500/500] Loss: 0.1822
Epoch [40/80], Step [100/500] Loss: 0.1110
Epoch [40/80], Step [200/500] Loss: 0.2265
Epoch [40/80], Step [300/500] Loss: 0.1517
Epoch [40/80], Step [400/500] Loss: 0.2840
Epoch [40/80], Step [500/500] Loss: 0.2290
Epoch [41/80], Step [100/500] Loss: 0.2211
Epoch [41/80], Step [200/500] Loss: 0.1697
Epoch [41/80], Step [300/500] Loss: 0.2321
Epoch [41/80], Step [400/500] Loss: 0.3043
Epoch [41/80], Step [500/500] Loss: 0.3035
Epoch [42/80], Step [100/500] Loss: 0.2421
Epoch [42/80], Step [200/500] Loss: 0.1752
Epoch [42/80], Step [300/500] Loss: 0.2698
Epoch [42/80], Step [400/500] Loss: 0.1251
Epoch [42/80], Step [500/500] Loss: 0.1092
Epoch [43/80], Step [100/500] Loss: 0.1583
Epoch [43/80], Step [200/500] Loss: 0.2241
Epoch [43/80], Step [300/500] Loss: 0.1903
Epoch [43/80], Step [400/500] Loss: 0.2138
Epoch [43/80], Step [500/500] Loss: 0.1410
Epoch [44/80], Step [100/500] Loss: 0.2107
Epoch [44/80], Step [200/500] Loss: 0.1608
Epoch [44/80], Step [300/500] Loss: 0.1878
Epoch [44/80], Step [400/500] Loss: 0.1976
Epoch [44/80], Step [500/500] Loss: 0.1361
Epoch [45/80], Step [100/500] Loss: 0.1442
Epoch [45/80], Step [200/500] Loss: 0.2911
Epoch [45/80], Step [300/500] Loss: 0.1455
Epoch [45/80], Step [400/500] Loss: 0.2712
Epoch [45/80], Step [500/500] Loss: 0.1288
Epoch [46/80], Step [100/500] Loss: 0.2417
Epoch [46/80], Step [200/500] Loss: 0.1889
Epoch [46/80], Step [300/500] Loss: 0.1500
Epoch [46/80], Step [400/500] Loss: 0.1471
Epoch [46/80], Step [500/500] Loss: 0.1577
Epoch [47/80], Step [100/500] Loss: 0.1232
Epoch [47/80], Step [200/500] Loss: 0.1485
Epoch [47/80], Step [300/500] Loss: 0.2873
Epoch [47/80], Step [400/500] Loss: 0.3067
Epoch [47/80], Step [500/500] Loss: 0.1435
Epoch [48/80], Step [100/500] Loss: 0.2490
Epoch [48/80], Step [200/500] Loss: 0.2533
Epoch [48/80], Step [300/500] Loss: 0.1119
Epoch [48/80], Step [400/500] Loss: 0.1546
Epoch [48/80], Step [500/500] Loss: 0.2637
Epoch [49/80], Step [100/500] Loss: 0.1515
Epoch [49/80], Step [200/500] Loss: 0.1934
Epoch [49/80], Step [300/500] Loss: 0.0766
Epoch [49/80], Step [400/500] Loss: 0.1556
Epoch [49/80], Step [500/500] Loss: 0.2014
Epoch [50/80], Step [100/500] Loss: 0.2393
Epoch [50/80], Step [200/500] Loss: 0.3324
Epoch [50/80], Step [300/500] Loss: 0.1303
Epoch [50/80], Step [400/500] Loss: 0.2781
Epoch [50/80], Step [500/500] Loss: 0.1666
Epoch [51/80], Step [100/500] Loss: 0.1734
Epoch [51/80], Step [200/500] Loss: 0.2402
Epoch [51/80], Step [300/500] Loss: 0.2121
Epoch [51/80], Step [400/500] Loss: 0.1578
Epoch [51/80], Step [500/500] Loss: 0.2454
Epoch [52/80], Step [100/500] Loss: 0.1594
Epoch [52/80], Step [200/500] Loss: 0.2134
Epoch [52/80], Step [300/500] Loss: 0.0919
Epoch [52/80], Step [400/500] Loss: 0.1769
Epoch [52/80], Step [500/500] Loss: 0.2247
Epoch [53/80], Step [100/500] Loss: 0.2414
Epoch [53/80], Step [200/500] Loss: 0.1700
Epoch [53/80], Step [300/500] Loss: 0.1553
Epoch [53/80], Step [400/500] Loss: 0.2141
Epoch [53/80], Step [500/500] Loss: 0.1380
Epoch [54/80], Step [100/500] Loss: 0.1040
Epoch [54/80], Step [200/500] Loss: 0.1396
Epoch [54/80], Step [300/500] Loss: 0.1406
Epoch [54/80], Step [400/500] Loss: 0.1281
Epoch [54/80], Step [500/500] Loss: 0.1767
Epoch [55/80], Step [100/500] Loss: 0.1912
Epoch [55/80], Step [200/500] Loss: 0.1481
Epoch [55/80], Step [300/500] Loss: 0.1444
Epoch [55/80], Step [400/500] Loss: 0.1703
Epoch [55/80], Step [500/500] Loss: 0.1799
Epoch [56/80], Step [100/500] Loss: 0.1166
Epoch [56/80], Step [200/500] Loss: 0.1637
Epoch [56/80], Step [300/500] Loss: 0.1259
Epoch [56/80], Step [400/500] Loss: 0.1635
Epoch [56/80], Step [500/500] Loss: 0.2844
Epoch [57/80], Step [100/500] Loss: 0.1429
Epoch [57/80], Step [200/500] Loss: 0.1619
Epoch [57/80], Step [300/500] Loss: 0.2326
Epoch [57/80], Step [400/500] Loss: 0.1635
Epoch [57/80], Step [500/500] Loss: 0.2051
Epoch [58/80], Step [100/500] Loss: 0.1408
Epoch [58/80], Step [200/500] Loss: 0.1411
Epoch [58/80], Step [300/500] Loss: 0.2641
Epoch [58/80], Step [400/500] Loss: 0.2165
Epoch [58/80], Step [500/500] Loss: 0.1642
Epoch [59/80], Step [100/500] Loss: 0.2740
Epoch [59/80], Step [200/500] Loss: 0.1128
Epoch [59/80], Step [300/500] Loss: 0.2121
Epoch [59/80], Step [400/500] Loss: 0.2078
Epoch [59/80], Step [500/500] Loss: 0.1159
Epoch [60/80], Step [100/500] Loss: 0.1053
Epoch [60/80], Step [200/500] Loss: 0.2107
Epoch [60/80], Step [300/500] Loss: 0.1609
Epoch [60/80], Step [400/500] Loss: 0.1636
Epoch [60/80], Step [500/500] Loss: 0.1829
Epoch [61/80], Step [100/500] Loss: 0.1145
Epoch [61/80], Step [200/500] Loss: 0.1602
Epoch [61/80], Step [300/500] Loss: 0.0962
Epoch [61/80], Step [400/500] Loss: 0.1498
Epoch [61/80], Step [500/500] Loss: 0.1702
Epoch [62/80], Step [100/500] Loss: 0.0975
Epoch [62/80], Step [200/500] Loss: 0.1107
Epoch [62/80], Step [300/500] Loss: 0.1446
Epoch [62/80], Step [400/500] Loss: 0.2560
Epoch [62/80], Step [500/500] Loss: 0.1480
Epoch [63/80], Step [100/500] Loss: 0.2328
Epoch [63/80], Step [200/500] Loss: 0.1314
Epoch [63/80], Step [300/500] Loss: 0.2381
Epoch [63/80], Step [400/500] Loss: 0.1002
Epoch [63/80], Step [500/500] Loss: 0.1364
Epoch [64/80], Step [100/500] Loss: 0.2337
Epoch [64/80], Step [200/500] Loss: 0.1603
Epoch [64/80], Step [300/500] Loss: 0.1622
Epoch [64/80], Step [400/500] Loss: 0.2281
Epoch [64/80], Step [500/500] Loss: 0.2363
Epoch [65/80], Step [100/500] Loss: 0.1970
Epoch [65/80], Step [200/500] Loss: 0.1200
Epoch [65/80], Step [300/500] Loss: 0.0948
Epoch [65/80], Step [400/500] Loss: 0.1586
Epoch [65/80], Step [500/500] Loss: 0.1901
Epoch [66/80], Step [100/500] Loss: 0.1846
Epoch [66/80], Step [200/500] Loss: 0.2095
Epoch [66/80], Step [300/500] Loss: 0.1828
Epoch [66/80], Step [400/500] Loss: 0.0947
Epoch [66/80], Step [500/500] Loss: 0.2168
Epoch [67/80], Step [100/500] Loss: 0.1777
Epoch [67/80], Step [200/500] Loss: 0.2234
Epoch [67/80], Step [300/500] Loss: 0.1363
Epoch [67/80], Step [400/500] Loss: 0.1051
Epoch [67/80], Step [500/500] Loss: 0.1794
Epoch [68/80], Step [100/500] Loss: 0.1194
Epoch [68/80], Step [200/500] Loss: 0.1159
Epoch [68/80], Step [300/500] Loss: 0.1734
Epoch [68/80], Step [400/500] Loss: 0.1428
Epoch [68/80], Step [500/500] Loss: 0.1188
Epoch [69/80], Step [100/500] Loss: 0.1475
Epoch [69/80], Step [200/500] Loss: 0.1046
Epoch [69/80], Step [300/500] Loss: 0.2766
Epoch [69/80], Step [400/500] Loss: 0.2422
Epoch [69/80], Step [500/500] Loss: 0.1225
Epoch [70/80], Step [100/500] Loss: 0.2478
Epoch [70/80], Step [200/500] Loss: 0.2029
Epoch [70/80], Step [300/500] Loss: 0.1470
Epoch [70/80], Step [400/500] Loss: 0.2930
Epoch [70/80], Step [500/500] Loss: 0.1295
Epoch [71/80], Step [100/500] Loss: 0.2208
Epoch [71/80], Step [200/500] Loss: 0.1250
Epoch [71/80], Step [300/500] Loss: 0.1159
Epoch [71/80], Step [400/500] Loss: 0.1022
Epoch [71/80], Step [500/500] Loss: 0.0812
Epoch [72/80], Step [100/500] Loss: 0.0846
Epoch [72/80], Step [200/500] Loss: 0.0844
Epoch [72/80], Step [300/500] Loss: 0.1620
Epoch [72/80], Step [400/500] Loss: 0.2203
Epoch [72/80], Step [500/500] Loss: 0.0930
Epoch [73/80], Step [100/500] Loss: 0.0905
Epoch [73/80], Step [200/500] Loss: 0.1385
Epoch [73/80], Step [300/500] Loss: 0.1368
Epoch [73/80], Step [400/500] Loss: 0.1394
Epoch [73/80], Step [500/500] Loss: 0.1557
Epoch [74/80], Step [100/500] Loss: 0.1351
Epoch [74/80], Step [200/500] Loss: 0.1484
Epoch [74/80], Step [300/500] Loss: 0.1033
Epoch [74/80], Step [400/500] Loss: 0.1216
Epoch [74/80], Step [500/500] Loss: 0.1550
Epoch [75/80], Step [100/500] Loss: 0.0889
Epoch [75/80], Step [200/500] Loss: 0.0894
Epoch [75/80], Step [300/500] Loss: 0.3254
Epoch [75/80], Step [400/500] Loss: 0.2428
Epoch [75/80], Step [500/500] Loss: 0.1525
Epoch [76/80], Step [100/500] Loss: 0.2483
Epoch [76/80], Step [200/500] Loss: 0.1099
Epoch [76/80], Step [300/500] Loss: 0.1486
Epoch [76/80], Step [400/500] Loss: 0.2034
Epoch [76/80], Step [500/500] Loss: 0.0944
Epoch [77/80], Step [100/500] Loss: 0.1246
Epoch [77/80], Step [200/500] Loss: 0.0717
Epoch [77/80], Step [300/500] Loss: 0.1999
Epoch [77/80], Step [400/500] Loss: 0.1432
Epoch [77/80], Step [500/500] Loss: 0.2132
Epoch [78/80], Step [100/500] Loss: 0.0835
Epoch [78/80], Step [200/500] Loss: 0.1805
Epoch [78/80], Step [300/500] Loss: 0.2742
Epoch [78/80], Step [400/500] Loss: 0.1205
Epoch [78/80], Step [500/500] Loss: 0.0664
Epoch [79/80], Step [100/500] Loss: 0.0953
Epoch [79/80], Step [200/500] Loss: 0.1801
Epoch [79/80], Step [300/500] Loss: 0.2401
Epoch [79/80], Step [400/500] Loss: 0.0929
Epoch [79/80], Step [500/500] Loss: 0.1187
Epoch [80/80], Step [100/500] Loss: 0.1222
Epoch [80/80], Step [200/500] Loss: 0.1596
Epoch [80/80], Step [300/500] Loss: 0.1393
Epoch [80/80], Step [400/500] Loss: 0.0888
Epoch [80/80], Step [500/500] Loss: 0.1377
Accuracy of the model on the test images: 87.97 %

DZzz丶
关注
专栏目录
深度残差收缩网络（Deep Residual Shrinkage Networks for Fault Diagnosis ）
qq_40107571的博客
10-30 1509
深度残差收缩网络（Deep Residual Shrinkage Networks for Fault Diagnosis ）
Reverse Attention-Based Residual Network for Salient Object Detection论文解读以及代码复现时所遇问题
weixin_45908408的博客
09-30 1340
论文原文下载：原文 代码下载：源代码 该博客只是在学习本篇论文时自己所遇问题以及一些心得，如有侵权，请联系本人删除即可。 该论文主要针对的是SOD遇到的一些问题，如：模型架构复杂、模型参数较多的问题，作者提出了一种框架即：Reverse Attention-Based Residual Network。 前述部分作者讲的很仔细，我就不多赘述，只进行模型的讲解。 Initial Saliency Prediction 首先经过VGG-16进行处理,因为VGG16独特的网络框架，具有五个最大池化层，因此在每个池
ResidualDenseNetwork-Pytorch:火炬工具
05-16
残差密集网络-Pytorch Pytorch工具： 本文的两个优点： 将denese连接层加入ResNet 层次特征的串联 与本文不同的是，我仅使用RDB（剩余密集块），每个RDB都有三个密集层。 因此，这是作者提出的RDN（剩余密集网络）的示例实现。 要求 python3.5 / 3.6 火炬> = 0.2 OpenCV 用法 您需要准备DIV2K数据集（./data/） 火车模型： python3 main.py --model_name 'RDN' --load demo_x3_RDN --dataDir ./DIV2K/ --need_patch True --patchSize 144 --nDenselayer 3 --nFeat 64 --growthRate 32 --scale 3 --epoch 10000 --lrDecay 2000 --lr 1
PyTorch学习笔记（六）Residual Network
午夜的行人
07-23 167
# 导入相关库函数 import torch import torch.nn as nn import torchvision import torchvision.transforms as transforms # 配置设备 device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') # 设置超参数 num_epochs = 80 batch_size = 100 learning_rate = 0.001 # 图像预
【吴恩达深度学习】Residual Networks（PyTorch）
无
03-15 1627
keras版本链接 导包 import torch from torch import nn from torch import optim import torch.nn.functional as F from torch.utils.data import Dataset from torch.utils.data import DataLoader from resnets_utils import * Dataset类 class MyDataset(Dataset): def __i
《动手学深度学习 Pytorch版》 7.6 残差网络(ResNet)
最新发布
AncilunKiang的博客
09-23 470
ResNet 在后面使用了 4 个由残差块组成的模块，每个模块使用若干个同样输出通道数的残差块。（4）在 ResNet 的后续版本中，作者将“卷积层、批量规范化层和激活层”架构更改为“批量规范化层、激活层和卷积层”架构。如果把模型看作一个函数，我们设计的更强大的模型则可以看作范围更大的函数。，这里的加法会更有益于靠近数据端的层的训练，因为乘法中的梯度波动会极大的影响链式法则的结果，而在残差块中输入可以通过加法通路更快的前向传播。只有当较复杂的函数类包含较小的函数类时，我们才能确保提高它们的性能。
深度学习——Residual Attention Network
weixin_51352656的博客
02-28 528
CVPR2017原论文：Residual Attention Network for Image Classification 开源代码（caffe框架）：https://github.com/fwang91/residual-attention-network 1.主要内容： 提出了残差注意力深度模型网络。 该网络具有以下特点：1）其由多个注意力模块堆叠而成，每个注意力模块能够捕获不同的注意类型。2）采用残差连接，避免梯度消失，能够扩展很大的深度，且能完成端到端的训练。 2.主体结构 如上图，有三个At
Deep Neural Networks with Pytorch（1）
qq_37840762的博客
07-31 380
目录 1.Overview of Tensors 1维和2维张量 2.导数和梯度 3.Datasets类 1.1Tensor 1D 1.数据类型 2.索引和切片 3.基本操作 4.通用函数 1.2 Tensor 2D 12D Tensor例子 2.创建2D Tensor 3. 2D Tensor索引和切片 4.常用操作 1.3 pytorch中导数和梯度 1.4 简单的Dataset 1.如何创建一个Dataset类 2.如何创建Transform类 3.如何Comp.
Deep Neural Networks with Pytorch（3）
qq_37840762的博客
07-31 154
目录 3.1 Stochastic Gradient Descent 1.随机梯度下降 2.Data Loader+SGD 3.2 Mini-batch Gradient Descent 1.Mini-batch梯度下降 2.pytorch中的mini-batch梯度下降 3.3 pytorch中的Optimization 3.4 Training,Validation and Test Split 1 训练集和验证集的作用 2 如何训练，验证模型并保存模型 3.1 ...
Deep Residual Network工具下载
05-31
如果你想使用Deep Residual Network作为你的神经网络模型，你可以使用一些深度学习框架来实现这个模型，比如TensorFlow、PyTorch等。 这些深度学习框架都提供了相应的工具和库，可以方便地实现Deep Residual ...
Resnet论文解读 （Deep Residual Learning for Image Recognition）
weixin_43179892的博客
09-14 1717
论文基本信息 标题：Deep Residual Learning for Image Recognition 作者：Kaiming He，Xiangyu Zhang， Shaoqing Ren ，and Jian Sun 机构：Microsoft Research 来源：CVPR 时间：2016 链接：[https://openaccess.thecvf.com/content_cvpr_2016/html/He_Deep_Residual_Learning_CVPR_2016_paper.html] 成
d2lzh_pytorch.zip
03-03
在强化学习部分，教程会讲解Q-learning、Deep Q-Network (DQN)等经典算法，以及如何在PyTorch中实现它们。这部分对于理解智能体与环境的交互至关重要。 自然语言处理（NLP）章节则涵盖了词嵌入、Transformer模型等...
论文《Dilated Residual Networks》的pytorch源码
08-29
论文《Dilated Residual Networks》的pytorch源码，python3环境。
[2021-10-15] Residual Attention Network in Image Classification PyTorch 小结
bossWww的博客
10-15 407
Residual Attention Network in Image Classification PyTorch 小结Residual Attention Network整体的网络结构Attention Module Residual Attention Network 论文地址: http://openaccess.thecvf.com/content_cvpr_2017/html/Wang_Residual_Attention_Network_CVPR_2017_paper.html 整体的网络结构
pytorch实现Deep_Residual_network,resnet源码解读
To be a better man
12-11 1293
今天带大家学习resnet网络系列，resnet的发展与思考，任何一篇网络的理解不应该单单局限在一篇或几篇博客上，复制别人的代码，不追根溯源是很难有深度的理解。 所以今天，我整理这篇博客带大家从头到位看resnet为什么有用（理论层面），怎么实现，如何在版本上的迭代（技术层面），pytorch里面关于torchvison.models里面的resnet.py源码已经迭代过几次version了，所以就在这个层面大家也应该知道，一个网络的实现和优化绝不是博...
deep_residual_network
咕噜咕噜
04-07 221
# ---------------------------------------------------------------------------- # # An implementation of https://arxiv.org/pdf/1512.03385.pdf # # See section 4.2 for the model archi...
DeepAR代码详析（pytorch版）实现用电量预测
weixin_52185313的博客
05-19 6259
DeepAR代码详析（pytorch版）实现用电量预测 – 潘登同学的RNN学习笔记 文章目录DeepAR代码详析（pytorch版）实现用电量预测 -- 潘登同学的RNN学习笔记数据集说明数据预处理代码构造模型Loss函数评估指标相关utils工具类训练模型 前言： 上次用Amazon中的glount-ts框架做了一个deepar的股价预测，但是我感觉用的是人家的API，不太好，所以今天来搂一把pytorch的deepar，看看效果如何 数据集说明 数据集说明 2011 ~2014期间； 370 个 家
十二、Pytorch复现Residual Block
beyond谚语的博客
04-06 801
《PyTorch深度学习实践》课后作业
Pseudo-3D Residual Networks算法的pytorch代码
guifeng93的博客
11-12 716
作者：AI之路 原文：https://blog.csdn.net/u014380165/article/details/78986430 本篇博客是对第三方实现的Pseudo-3D Residual Networks算法的pytorch代码进行介绍，介绍顺序为代码调试顺序，建议先阅读论文或相关博客。 论文：Learning Spatio-Temporal Representation with ...
评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值