🍨 本文为：[🔗365天深度学习训练营] 中的学习记录博客
🍖 原作者：[K同学啊 | 接辅导、项目定制]
要求：
- 阅读ResNeXt论文，了解作者的构建思路
- 对比我们之前介绍的ResNet50V2、DenseNet算法
- 使用ResNeXt-50算法完成猴痘病识别
一、 基础配置
- 语言环境：Python3.8
- 编译器选择：Pycharm
- 深度学习环境：
-
- torch==1.12.1+cu113
- torchvision==0.13.1+cu113
二、 前期准备
1.设置GPU
import pathlib
import torch
import torch.nn as nn
import torch.nn.functional as F
from torchvision import transforms, datasets
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)2. 导入数据
本项目所采用的数据集未收录于公开数据中，故需要自己在文件目录中导入相应数据集合，并设置对应文件目录，以供后续学习过程中使用。
运行下述代码：
data_dir = './data/'
data_dir = pathlib.Path(data_dir)
data_paths = list(data_dir.glob('*'))
classeNames = [str(path).split("\\")[1] for path in data_paths]
print(classeNames)
image_count = len(list(data_dir.glob('*/*')))
print("图片总数为：", image_count)得到如下输出：
['Monkeypox', 'Others']
图片总数为： 2142接下来，我们通过transforms.Compose对整个数据集进行预处理：
train_transforms = transforms.Compose([
transforms.Resize([224, 224]), # 将输入图片resize成统一尺寸
# transforms.RandomHorizontalFlip(), # 随机水平翻转
transforms.ToTensor(), # 将PIL Image或numpy.ndarray转换为tensor，并归一化到[0,1]之间
transforms.Normalize( # 标准化处理-->转换为标准正太分布（高斯分布），使模型更容易收敛
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]) # 其中 mean=[0.485,0.456,0.406]与std=[0.229,0.224,0.225] 从数据集中随机抽样计算得到的。
])
test_transform = transforms.Compose([
transforms.Resize([224, 224]), # 将输入图片resize成统一尺寸
transforms.ToTensor(), # 将PIL Image或numpy.ndarray转换为tensor，并归一化到[0,1]之间
transforms.Normalize( # 标准化处理-->转换为标准正太分布（高斯分布），使模型更容易收敛
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]) # 其中 mean=[0.485,0.456,0.406]与std=[0.229,0.224,0.225] 从数据集中随机抽样计算得到的。
])
total_data = datasets.ImageFolder("./data/", transform=train_transforms)
print(total_data.class_to_idx)得到如下输出：
{'Monkeypox': 0, 'Others': 1}3. 划分数据集
此处数据集需要做按比例划分的操作：
train_size = int(0.8 * len(total_data))
test_size = len(total_data) - train_size
train_dataset, test_dataset = torch.utils.data.random_split(total_data, [train_size, test_size])
接下来，根据划分得到的训练集和验证集对数据集进行包装：
batch_size = 32
train_dl = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=0)
test_dl = torch.utils.data.DataLoader(test_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=0)并通过：
for X, y in test_dl:
print("Shape of X [N, C, H, W]: ", X.shape)
print("Shape of y: ", y.shape, y.dtype)
break
输出测试数据集的数据分布情况：
Shape of X [N, C, H, W]: torch.Size([32, 3, 224, 224])
Shape of y: torch.Size([32]) torch.int644.搭建模型
1.模型搭建
class BN_Conv2d(nn.Module):
"""
BN_CONV_RELU
"""
def __init__(self, in_channels, out_channels, kernel_size, stride, padding, dilation=1, groups=1, bias=False):
super(BN_Conv2d, self).__init__()
self.seq = nn.Sequential(
nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, stride=stride,
padding=padding, dilation=dilation, groups=groups, bias=bias),
nn.BatchNorm2d(out_channels)
)
def forward(self, x):
return F.relu(self.seq(x))
class ResNeXt_Block(nn.Module):
"""
ResNeXt block with group convolutions
"""
def __init__(self, in_chnls, cardinality, group_depth, stride):
super(ResNeXt_Block, self).__init__()
self.group_chnls = cardinality * group_depth
self.conv1 = BN_Conv2d(in_chnls, self.group_chnls, 1, stride=1, padding=0)
self.conv2 = BN_Conv2d(self.group_chnls, self.group_chnls, 3, stride=stride, padding=1, groups=cardinality)
self.conv3 = nn.Conv2d(self.group_chnls, self.group_chnls * 2, 1, stride=1, padding=0)
self.bn = nn.BatchNorm2d(self.group_chnls * 2)
self.short_cut = nn.Sequential(
nn.Conv2d(in_chnls, self.group_chnls * 2, 1, stride, 0, bias=False),
nn.BatchNorm2d(self.group_chnls * 2)
)
def forward(self, x):
out = self.conv1(x)
out = self.conv2(out)
out = self.bn(self.conv3(out))
out += self.short_cut(x)
return F.relu(out)
class ResNeXt(nn.Module):
"""
ResNeXt builder
"""
def __init__(self, layers: object, cardinality, group_depth, num_classes) -> object:
super(ResNeXt, self).__init__()
self.cardinality = cardinality
self.channels = 64
self.conv1 = BN_Conv2d(3, self.channels, 7, stride=2, padding=3)
d1 = group_depth
self.conv2 = self.___make_layers(d1, layers[0], stride=1)
d2 = d1 * 2
self.conv3 = self.___make_layers(d2, layers[1], stride=2)
d3 = d2 * 2
self.conv4 = self.___make_layers(d3, layers[2], stride=2)
d4 = d3 * 2
self.conv5 = self.___make_layers(d4, layers[3], stride=2)
self.fc = nn.Linear(self.channels, num_classes) # 224x224 input size
def ___make_layers(self, d, blocks, stride):
strides = [stride] + [1] * (blocks - 1)
layers = []
for stride in strides:
layers.append(ResNeXt_Block(self.channels, self.cardinality, d, stride))
self.channels = self.cardinality * d * 2
return nn.Sequential(*layers)
def forward(self, x):
out = self.conv1(x)
out = F.max_pool2d(out, 3, 2, 1)
out = self.conv2(out)
out = self.conv3(out)
out = self.conv4(out)
out = self.conv5(out)
out = F.avg_pool2d(out, 7)
out = out.view(out.size(0), -1)
out = F.softmax(self.fc(out), dim=1)
return out2.查看模型信息
model = ResNeXt([3, 4, 6, 3], 32, 4, 4)
model.to(device)
# 统计模型参数量以及其他指标
import torchsummary as summary
summary.summary(model, (3, 224, 224))得到如下输出：
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [-1, 64, 112, 112] 9,408
BatchNorm2d-2 [-1, 64, 112, 112] 128
BN_Conv2d-3 [-1, 64, 112, 112] 0
Conv2d-4 [-1, 128, 56, 56] 8,192
BatchNorm2d-5 [-1, 128, 56, 56] 256
BN_Conv2d-6 [-1, 128, 56, 56] 0
Conv2d-7 [-1, 128, 56, 56] 4,608
BatchNorm2d-8 [-1, 128, 56, 56] 256
BN_Conv2d-9 [-1, 128, 56, 56] 0
Conv2d-10 [-1, 256, 56, 56] 33,024
BatchNorm2d-11 [-1, 256, 56, 56] 512
Conv2d-12 [-1, 256, 56, 56] 16,384
BatchNorm2d-13 [-1, 256, 56, 56] 512
ResNeXt_Block-14 [-1, 256, 56, 56] 0
Conv2d-15 [-1, 128, 56, 56] 32,768
BatchNorm2d-16 [-1, 128, 56, 56] 256
BN_Conv2d-17 [-1, 128, 56, 56] 0
Conv2d-18 [-1, 128, 56, 56] 4,608
BatchNorm2d-19 [-1, 128, 56, 56] 256
BN_Conv2d-20 [-1, 128, 56, 56] 0
Conv2d-21 [-1, 256, 56, 56] 33,024
BatchNorm2d-22 [-1, 256, 56, 56] 512
Conv2d-23 [-1, 256, 56, 56] 65,536
BatchNorm2d-24 [-1, 256, 56, 56] 512
ResNeXt_Block-25 [-1, 256, 56, 56] 0
Conv2d-26 [-1, 128, 56, 56] 32,768
BatchNorm2d-27 [-1, 128, 56, 56] 256
BN_Conv2d-28 [-1, 128, 56, 56] 0
Conv2d-29 [-1, 128, 56, 56] 4,608
BatchNorm2d-30 [-1, 128, 56, 56] 256
BN_Conv2d-31 [-1, 128, 56, 56] 0
Conv2d-32 [-1, 256, 56, 56] 33,024
BatchNorm2d-33 [-1, 256, 56, 56] 512
Conv2d-34 [-1, 256, 56, 56] 65,536
BatchNorm2d-35 [-1, 256, 56, 56] 512
ResNeXt_Block-36 [-1, 256, 56, 56] 0
Conv2d-37 [-1, 256, 56, 56] 65,536
BatchNorm2d-38 [-1, 256, 56, 56] 512
BN_Conv2d-39 [-1, 256, 56, 56] 0
Conv2d-40 [-1, 256, 28, 28] 18,432
BatchNorm2d-41 [-1, 256, 28, 28] 512
BN_Conv2d-42 [-1, 256, 28, 28] 0
Conv2d-43 [-1, 512, 28, 28] 131,584
BatchNorm2d-44 [-1, 512, 28, 28] 1,024
Conv2d-45 [-1, 512, 28, 28] 131,072
BatchNorm2d-46 [-1, 512, 28, 28] 1,024
ResNeXt_Block-47 [-1, 512, 28, 28] 0
Conv2d-48 [-1, 256, 28, 28] 131,072
BatchNorm2d-49 [-1, 256, 28, 28] 512
BN_Conv2d-50 [-1, 256, 28, 28] 0
Conv2d-51 [-1, 256, 28, 28] 18,432
BatchNorm2d-52 [-1, 256, 28, 28] 512
BN_Conv2d-53 [-1, 256, 28, 28] 0
Conv2d-54 [-1, 512, 28, 28] 131,584
BatchNorm2d-55 [-1, 512, 28, 28] 1,024
Conv2d-56 [-1, 512, 28, 28] 262,144
BatchNorm2d-57 [-1, 512, 28, 28] 1,024
ResNeXt_Block-58 [-1, 512, 28, 28] 0
Conv2d-59 [-1, 256, 28, 28] 131,072
BatchNorm2d-60 [-1, 256, 28, 28] 512
BN_Conv2d-61 [-1, 256, 28, 28] 0
Conv2d-62 [-1, 256, 28, 28] 18,432
BatchNorm2d-63 [-1, 256, 28, 28] 512
BN_Conv2d-64 [-1, 256, 28, 28] 0
Conv2d-65 [-1, 512, 28, 28] 131,584
BatchNorm2d-66 [-1, 512, 28, 28] 1,024
Conv2d-67 [-1, 512, 28, 28] 262,144
BatchNorm2d-68 [-1, 512, 28, 28] 1,024
ResNeXt_Block-69 [-1, 512, 28, 28] 0
Conv2d-70 [-1, 256, 28, 28] 131,072
BatchNorm2d-71 [-1, 256, 28, 28] 512
BN_Conv2d-72 [-1, 256, 28, 28] 0
Conv2d-73 [-1, 256, 28, 28] 18,432
BatchNorm2d-74 [-1, 256, 28, 28] 512
BN_Conv2d-75 [-1, 256, 28, 28] 0
Conv2d-76 [-1, 512, 28, 28] 131,584
BatchNorm2d-77 [-1, 512, 28, 28] 1,024
Conv2d-78 [-1, 512, 28, 28] 262,144
BatchNorm2d-79 [-1, 512, 28, 28] 1,024
ResNeXt_Block-80 [-1, 512, 28, 28] 0
Conv2d-81 [-1, 512, 28, 28] 262,144
BatchNorm2d-82 [-1, 512, 28, 28] 1,024
BN_Conv2d-83 [-1, 512, 28, 28] 0
Conv2d-84 [-1, 512, 14, 14] 73,728
BatchNorm2d-85 [-1, 512, 14, 14] 1,024
BN_Conv2d-86 [-1, 512, 14, 14] 0
Conv2d-87 [-1, 1024, 14, 14] 525,312
BatchNorm2d-88 [-1, 1024, 14, 14] 2,048
Conv2d-89 [-1, 1024, 14, 14] 524,288
BatchNorm2d-90 [-1, 1024, 14, 14] 2,048
ResNeXt_Block-91 [-1, 1024, 14, 14] 0
Conv2d-92 [-1, 512, 14, 14] 524,288
BatchNorm2d-93 [-1, 512, 14, 14] 1,024
BN_Conv2d-94 [-1, 512, 14, 14] 0
Conv2d-95 [-1, 512, 14, 14] 73,728
BatchNorm2d-96 [-1, 512, 14, 14] 1,024
BN_Conv2d-97 [-1, 512, 14, 14] 0
Conv2d-98 [-1, 1024, 14, 14] 525,312
BatchNorm2d-99 [-1, 1024, 14, 14] 2,048
Conv2d-100 [-1, 1024, 14, 14] 1,048,576
BatchNorm2d-101 [-1, 1024, 14, 14] 2,048
ResNeXt_Block-102 [-1, 1024, 14, 14] 0
Conv2d-103 [-1, 512, 14, 14] 524,288
BatchNorm2d-104 [-1, 512, 14, 14] 1,024
BN_Conv2d-105 [-1, 512, 14, 14] 0
Conv2d-106 [-1, 512, 14, 14] 73,728
BatchNorm2d-107 [-1, 512, 14, 14] 1,024
BN_Conv2d-108 [-1, 512, 14, 14] 0
Conv2d-109 [-1, 1024, 14, 14] 525,312
BatchNorm2d-110 [-1, 1024, 14, 14] 2,048
Conv2d-111 [-1, 1024, 14, 14] 1,048,576
BatchNorm2d-112 [-1, 1024, 14, 14] 2,048
ResNeXt_Block-113 [-1, 1024, 14, 14] 0
Conv2d-114 [-1, 512, 14, 14] 524,288
BatchNorm2d-115 [-1, 512, 14, 14] 1,024
BN_Conv2d-116 [-1, 512, 14, 14] 0
Conv2d-117 [-1, 512, 14, 14] 73,728
BatchNorm2d-118 [-1, 512, 14, 14] 1,024
BN_Conv2d-119 [-1, 512, 14, 14] 0
Conv2d-120 [-1, 1024, 14, 14] 525,312
BatchNorm2d-121 [-1, 1024, 14, 14] 2,048
Conv2d-122 [-1, 1024, 14, 14] 1,048,576
BatchNorm2d-123 [-1, 1024, 14, 14] 2,048
ResNeXt_Block-124 [-1, 1024, 14, 14] 0
Conv2d-125 [-1, 512, 14, 14] 524,288
BatchNorm2d-126 [-1, 512, 14, 14] 1,024
BN_Conv2d-127 [-1, 512, 14, 14] 0
Conv2d-128 [-1, 512, 14, 14] 73,728
BatchNorm2d-129 [-1, 512, 14, 14] 1,024
BN_Conv2d-130 [-1, 512, 14, 14] 0
Conv2d-131 [-1, 1024, 14, 14] 525,312
BatchNorm2d-132 [-1, 1024, 14, 14] 2,048
Conv2d-133 [-1, 1024, 14, 14] 1,048,576
BatchNorm2d-134 [-1, 1024, 14, 14] 2,048
ResNeXt_Block-135 [-1, 1024, 14, 14] 0
Conv2d-136 [-1, 512, 14, 14] 524,288
BatchNorm2d-137 [-1, 512, 14, 14] 1,024
BN_Conv2d-138 [-1, 512, 14, 14] 0
Conv2d-139 [-1, 512, 14, 14] 73,728
BatchNorm2d-140 [-1, 512, 14, 14] 1,024
BN_Conv2d-141 [-1, 512, 14, 14] 0
Conv2d-142 [-1, 1024, 14, 14] 525,312
BatchNorm2d-143 [-1, 1024, 14, 14] 2,048
Conv2d-144 [-1, 1024, 14, 14] 1,048,576
BatchNorm2d-145 [-1, 1024, 14, 14] 2,048
ResNeXt_Block-146 [-1, 1024, 14, 14] 0
Conv2d-147 [-1, 1024, 14, 14] 1,048,576
BatchNorm2d-148 [-1, 1024, 14, 14] 2,048
BN_Conv2d-149 [-1, 1024, 14, 14] 0
Conv2d-150 [-1, 1024, 7, 7] 294,912
BatchNorm2d-151 [-1, 1024, 7, 7] 2,048
BN_Conv2d-152 [-1, 1024, 7, 7] 0
Conv2d-153 [-1, 2048, 7, 7] 2,099,200
BatchNorm2d-154 [-1, 2048, 7, 7] 4,096
Conv2d-155 [-1, 2048, 7, 7] 2,097,152
BatchNorm2d-156 [-1, 2048, 7, 7] 4,096
ResNeXt_Block-157 [-1, 2048, 7, 7] 0
Conv2d-158 [-1, 1024, 7, 7] 2,097,152
BatchNorm2d-159 [-1, 1024, 7, 7] 2,048
BN_Conv2d-160 [-1, 1024, 7, 7] 0
Conv2d-161 [-1, 1024, 7, 7] 294,912
BatchNorm2d-162 [-1, 1024, 7, 7] 2,048
BN_Conv2d-163 [-1, 1024, 7, 7] 0
Conv2d-164 [-1, 2048, 7, 7] 2,099,200
BatchNorm2d-165 [-1, 2048, 7, 7] 4,096
Conv2d-166 [-1, 2048, 7, 7] 4,194,304
BatchNorm2d-167 [-1, 2048, 7, 7] 4,096
ResNeXt_Block-168 [-1, 2048, 7, 7] 0
Conv2d-169 [-1, 1024, 7, 7] 2,097,152
BatchNorm2d-170 [-1, 1024, 7, 7] 2,048
BN_Conv2d-171 [-1, 1024, 7, 7] 0
Conv2d-172 [-1, 1024, 7, 7] 294,912
BatchNorm2d-173 [-1, 1024, 7, 7] 2,048
BN_Conv2d-174 [-1, 1024, 7, 7] 0
Conv2d-175 [-1, 2048, 7, 7] 2,099,200
BatchNorm2d-176 [-1, 2048, 7, 7] 4,096
Conv2d-177 [-1, 2048, 7, 7] 4,194,304
BatchNorm2d-178 [-1, 2048, 7, 7] 4,096
ResNeXt_Block-179 [-1, 2048, 7, 7] 0
Linear-180 [-1, 4] 8,196
================================================================
Total params: 37,574,724
Trainable params: 37,574,724
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.57
Forward/backward pass size (MB): 379.37
Params size (MB): 143.34
Estimated Total Size (MB): 523.28
----------------------------------------------------------------三、 训练模型
1. 编写训练函数
# 训练循环
def train(dataloader, model, loss_fn, optimizer):
size = len(dataloader.dataset) # 训练集的大小
num_batches = len(dataloader) # 批次数目, (size/batch_size，向上取整)
train_loss, train_acc = 0, 0 # 初始化训练损失和正确率
for X, y in dataloader: # 获取图片及其标签
X, y = X.to(device), y.to(device)
# 计算预测误差
pred = model(X) # 网络输出
loss = loss_fn(pred, y) # 计算网络输出和真实值之间的差距，targets为真实值，计算二者差值即为损失
# 反向传播
optimizer.zero_grad() # grad属性归零
loss.backward() # 反向传播
optimizer.step() # 每一步自动更新
# 记录acc与loss
train_acc += (pred.argmax(1) == y).type(torch.float).sum().item()
train_loss += loss.item()
train_acc /= size
train_loss /= num_batches
return train_acc, train_loss2. 编写测试函数
测试函数和训练函数大致相同，但是由于不进行梯度下降对网络权重进行更新，所以不需要传入优化器
def test(dataloader, model, loss_fn):
size = len(dataloader.dataset) # 测试集的大小
num_batches = len(dataloader) # 批次数目
test_loss, test_acc = 0, 0
# 当不进行训练时，停止梯度更新，节省计算内存消耗
with torch.no_grad():
for imgs, target in dataloader:
imgs, target = imgs.to(device), target.to(device)
# 计算loss
target_pred = model(imgs)
loss = loss_fn(target_pred, target)
test_loss += loss.item()
test_acc += (target_pred.argmax(1) == target).type(torch.float).sum().item()
test_acc /= size
test_loss /= num_batches
return test_acc, test_loss
3.正式训练
import copy
optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
loss_fn = nn.CrossEntropyLoss() # 创建损失函数
epochs = 10
train_loss = []
train_acc = []
test_loss = []
test_acc = []
best_acc = 0 # 设置一个最佳准确率，作为最佳模型的判别指标
for epoch in range(epochs):
# 更新学习率（使用自定义学习率时使用）
# adjust_learning_rate(optimizer, epoch, learn_rate)
model.train()
epoch_train_acc, epoch_train_loss = train(train_dl, model, loss_fn, optimizer)
# scheduler.step() # 更新学习率（调用官方动态学习率接口时使用）
model.eval()
epoch_test_acc, epoch_test_loss = test(test_dl, model, loss_fn)
# 保存最佳模型到 best_model
if epoch_test_acc > best_acc:
best_acc = epoch_test_acc
best_model = copy.deepcopy(model)
train_acc.append(epoch_train_acc)
train_loss.append(epoch_train_loss)
test_acc.append(epoch_test_acc)
test_loss.append(epoch_test_loss)
# 获取当前的学习率
lr = optimizer.state_dict()['param_groups'][0]['lr']
template = ('Epoch:{:2d}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%, Test_loss:{:.3f}, Lr:{:.2E}')
print(template.format(epoch + 1, epoch_train_acc * 100, epoch_train_loss,
epoch_test_acc * 100, epoch_test_loss, lr))
# 保存最佳模型到文件中
PATH = './best_model.pth' # 保存的参数文件名
torch.save(model.state_dict(), PATH)
print('Done')得到如下输出：
Epoch: 1, Train_acc:54.9%, Train_loss:1.188, Test_acc:55.2%, Test_loss:1.196, Lr:1.00E-04
Epoch: 2, Train_acc:62.9%, Train_loss:1.104, Test_acc:67.6%, Test_loss:1.064, Lr:1.00E-04
Epoch: 3, Train_acc:66.1%, Train_loss:1.075, Test_acc:68.5%, Test_loss:1.058, Lr:1.00E-04
Epoch: 4, Train_acc:66.9%, Train_loss:1.071, Test_acc:69.9%, Test_loss:1.045, Lr:1.00E-04
Epoch: 5, Train_acc:68.1%, Train_loss:1.060, Test_acc:66.0%, Test_loss:1.081, Lr:1.00E-04
Epoch: 6, Train_acc:68.9%, Train_loss:1.049, Test_acc:67.6%, Test_loss:1.060, Lr:1.00E-04
Epoch: 7, Train_acc:70.3%, Train_loss:1.041, Test_acc:71.8%, Test_loss:1.029, Lr:1.00E-04
Epoch: 8, Train_acc:71.1%, Train_loss:1.030, Test_acc:71.8%, Test_loss:1.016, Lr:1.00E-04
Epoch: 9, Train_acc:71.3%, Train_loss:1.027, Test_acc:71.3%, Test_loss:1.016, Lr:1.00E-04
Epoch:10, Train_acc:69.2%, Train_loss:1.052, Test_acc:68.5%, Test_loss:1.057, Lr:1.00E-04
Done四、 结果可视化
1. Loss&Accuracy
import matplotlib.pyplot as plt
# 隐藏警告
import warnings
warnings.filterwarnings("ignore") # 忽略警告信息
plt.rcParams['font.sans-serif'] = ['SimHei'] # 用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False # 用来正常显示负号
plt.rcParams['figure.dpi'] = 100 # 分辨率
epochs_range = range(epochs)
plt.figure(figsize=(12, 3))
plt.subplot(1, 2, 1)
plt.plot(epochs_range, train_acc, label='Training Accuracy')
plt.plot(epochs_range, test_acc, label='Test Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')
plt.subplot(1, 2, 2)
plt.plot(epochs_range, train_loss, label='Training Loss')
plt.plot(epochs_range, test_loss, label='Test Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
plt.show()
得到的可视化结果：
2. 指定图片进行预测
首先，先定义出一个用于预测的函数：
from PIL import Image
classes = list(total_data.class_to_idx)
def predict_one_image(image_path, model, transform, classes):
test_img = Image.open(image_path).convert('RGB')
plt.imshow(test_img) # 展示预测的图片
test_img = transform(test_img)
img = test_img.to(device).unsqueeze(0)
model.eval()
output = model(img)
_, pred = torch.max(output, 1)
pred_class = classes[pred]
print(f'预测结果是：{pred_class}')接着调用函数对指定图片进行预测：
# 预测训练集中的某张照片
predict_one_image(image_path='./data/Monkeypox/M01_01_00.jpg',
model=model,
transform=train_transforms,
classes=classes)得到如下结果：
预测结果是：Others五、网络介绍
ResNeXt是ResNet的升级版，在ResNet的基础上，引入了cardinality的概念，类似于ResNet，ResNeXt也有ResNeXt-50，ResNeXt-101的版本。
这篇文章介绍了一种用于图像分类的简单而有效的网络架构，称为Aggregated Residual Transformations for Deep Neural Networks。该网络采用了VGG/ResNets的策略，通过重复层来增加深度和宽度，并利用分裂-变换-合并策略以易于扩展的方式进行转换。文章还提出了一个新的维度——“基数”，它是指转换集合的大小，可以在保持复杂性不变的情况下提高分类准确性。作者在ImageNet-1K数据集上进行了实证研究，证明了这种方法的有效性。
下图是ResNet（左）与ResNeXt（右）block的差异。在ResNet中，输入的具有256个通道的特征经过1×1卷积压缩4倍到64个通道，之后3×3的卷积核用于处理特征，经1×1卷积扩大通道数与原特征残差连接后输出。
ResNeXt也是相同的处理策略，但在ResNeXt中，输入的具有256个通道的特征被分为32个组，每组被压缩64倍到4个通道后进行处理。32个组相加后与原特征残差连接后输出。这里cardinatity指的是一个block中所具有的相同分支的数目。
分组卷积：
ResNeXt中采用的分组卷机简单来说就是将特征图分为不同的组，再对每组特征图分别进行卷积，这个操作可以有效的降低计算量。
在分组卷积中，每个卷积核只处理部分通道，比如下图中，红色卷积核只处理红色的通道，绿色卷积核只处理绿色通道，黄色卷积核只处理黄色通道。此时每个卷积核有2个通道，每个卷积核生成一张特征图。
总结来说就是：ResNeXt-50网络简单讲就是在ResNet结构的基础上采用了聚合残差结构和局部连接结构，同时引入了Random Erasing和Mixup等数据增强和正则化方法。
扫一扫
1293
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
