🍨 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：
- 根据本文Tensorflow代码，编写对应的Pytorch代码
- 了解ResNetV2与ResNetV的区别
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
python
import torch
import torch.nn as nn
from torchvision import transforms, datasets
import pathlib, warnings
warnings.filterwarnings("ignore") # 忽略警告信息
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")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：
python
data_dir = './data/bird_photos/'
data_dir = pathlib.Path(data_dir)
data_paths = list(data_dir.glob('*'))
classeNames = [str(path).split("\\")[2] for path in data_paths]
print(classeNames)得到如下输出：
python
['Bananaquit', 'Black Skimmer', 'Black Throated Bushtiti', 'Cockatoo']接下来，我们通过transforms.Compose对整个数据集进行预处理：
python
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/bird_photos/", transform=train_transforms)
print(total_data.class_to_idx)得到如下输出：
python
{'Bananaquit': 0, 'Black Skimmer': 1, 'Black Throated Bushtiti': 2, 'Cockatoo': 3}3. 划分数据集
此处数据集需要做按比例划分的操作：
python
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：
python
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：
python
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：
python
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
python
class Block2(nn.Module):
def __init__(self, in_channel, filters, kernel_size=3, stride=1, conv_shortcut=False):
super(Block2, self).__init__()
self.preact = nn.Sequential(
nn.BatchNorm2d(in_channel),
nn.ReLU(True)
)
self.shortcut = conv_shortcut
if self.shortcut:
self.short = nn.Conv2d(in_channel, 4 * filters, 1, stride=stride, padding=0, bias=False)
elif stride > 1:
self.short = nn.MaxPool2d(kernel_size=1, stride=stride, padding=0)
else:
self.short = nn.Identity()
self.conv1 = nn.Sequential(
nn.Conv2d(in_channel, filters, 1, stride=1, bias=False),
nn.BatchNorm2d(filters),
nn.ReLU(True)
)
self.conv2 = nn.Sequential(
nn.Conv2d(filters, filters, kernel_size, stride=stride, padding=1, bias=False),
nn.BatchNorm2d(filters),
nn.ReLU(True)
)
self.conv3 = nn.Conv2d(filters, 4 * filters, 1, stride=1, bias=False)
def forward(self, x):
x1 = self.preact(x)
if self.shortcut:
x2 = self.short(x1)
else:
x2 = self.short(x)
x1 = self.conv1(x1)
x1 = self.conv2(x1)
x1 = self.conv3(x1)
x = x1 + x2
return x
class Stack2(nn.Module):
def __init__(self, in_channel, filters, blocks, stride=2):
super(Stack2, self).__init__()
self.conv = nn.Sequential()
self.conv.add_module(str(0), Block2(in_channel, filters, conv_shortcut=True))
for i in range(1, blocks - 1):
self.conv.add_module(str(i), Block2(4 * filters, filters))
self.conv.add_module(str(blocks - 1), Block2(4 * filters, filters, stride=stride))
def forward(self, x):
x = self.conv(x)
return x
class ResNet50V2(nn.Module):
def __init__(self,
include_top=True, # 是否包含位于网络顶部的全链接层
preact=True, # 是否使用预激活
use_bias=True, # 是否对卷积层使用偏置
input_shape=[224, 224, 3],
classes=1000,
pooling=None): # 用于分类图像的可选类数
super(ResNet50V2, self).__init__()
self.conv1 = nn.Sequential()
self.conv1.add_module('conv', nn.Conv2d(3, 64, 7, stride=2, padding=3, bias=use_bias, padding_mode='zeros'))
if not preact:
self.conv1.add_module('bn', nn.BatchNorm2d(64))
self.conv1.add_module('relu', nn.ReLU())
self.conv1.add_module('max_pool', nn.MaxPool2d(kernel_size=3, stride=2, padding=1))
self.conv2 = Stack2(64, 64, 3)
self.conv3 = Stack2(256, 128, 4)
self.conv4 = Stack2(512, 256, 6)
self.conv5 = Stack2(1024, 512, 3, stride=1)
self.post = nn.Sequential()
if preact:
self.post.add_module('bn', nn.BatchNorm2d(2048))
self.post.add_module('relu', nn.ReLU())
if include_top:
self.post.add_module('avg_pool', nn.AdaptiveAvgPool2d((1, 1)))
self.post.add_module('flatten', nn.Flatten())
self.post.add_module('fc', nn.Linear(2048, classes))
else:
if pooling == 'avg':
self.post.add_module('avg_pool', nn.AdaptiveAvgPool2d((1, 1)))
elif pooling == 'max':
self.post.add_module('max_pool', nn.AdaptiveMaxPool2d((1, 1)))
def forward(self, x):
x = self.conv1(x)
x = self.conv2(x)
x = self.conv3(x)
x = self.conv4(x)
x = self.conv5(x)
x = self.post(x)
return x
model = ResNet50V2().to(device)2.查看模型信息
python
import torchsummary as summary
summary.summary(model, (3, 224, 224))得到如下输出：
python
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [-1, 64, 112, 112] 9,472
MaxPool2d-2 [-1, 64, 56, 56] 0
BatchNorm2d-3 [-1, 64, 56, 56] 128
ReLU-4 [-1, 64, 56, 56] 0
Conv2d-5 [-1, 256, 56, 56] 16,384
Conv2d-6 [-1, 64, 56, 56] 4,096
BatchNorm2d-7 [-1, 64, 56, 56] 128
ReLU-8 [-1, 64, 56, 56] 0
Conv2d-9 [-1, 64, 56, 56] 36,864
BatchNorm2d-10 [-1, 64, 56, 56] 128
ReLU-11 [-1, 64, 56, 56] 0
Conv2d-12 [-1, 256, 56, 56] 16,384
Block2-13 [-1, 256, 56, 56] 0
BatchNorm2d-14 [-1, 256, 56, 56] 512
ReLU-15 [-1, 256, 56, 56] 0
Identity-16 [-1, 256, 56, 56] 0
Conv2d-17 [-1, 64, 56, 56] 16,384
BatchNorm2d-18 [-1, 64, 56, 56] 128
ReLU-19 [-1, 64, 56, 56] 0
Conv2d-20 [-1, 64, 56, 56] 36,864
BatchNorm2d-21 [-1, 64, 56, 56] 128
ReLU-22 [-1, 64, 56, 56] 0
Conv2d-23 [-1, 256, 56, 56] 16,384
Block2-24 [-1, 256, 56, 56] 0
BatchNorm2d-25 [-1, 256, 56, 56] 512
ReLU-26 [-1, 256, 56, 56] 0
MaxPool2d-27 [-1, 256, 28, 28] 0
Conv2d-28 [-1, 64, 56, 56] 16,384
BatchNorm2d-29 [-1, 64, 56, 56] 128
ReLU-30 [-1, 64, 56, 56] 0
Conv2d-31 [-1, 64, 28, 28] 36,864
BatchNorm2d-32 [-1, 64, 28, 28] 128
ReLU-33 [-1, 64, 28, 28] 0
Conv2d-34 [-1, 256, 28, 28] 16,384
Block2-35 [-1, 256, 28, 28] 0
Stack2-36 [-1, 256, 28, 28] 0
BatchNorm2d-37 [-1, 256, 28, 28] 512
ReLU-38 [-1, 256, 28, 28] 0
Conv2d-39 [-1, 512, 28, 28] 131,072
Conv2d-40 [-1, 128, 28, 28] 32,768
BatchNorm2d-41 [-1, 128, 28, 28] 256
ReLU-42 [-1, 128, 28, 28] 0
Conv2d-43 [-1, 128, 28, 28] 147,456
BatchNorm2d-44 [-1, 128, 28, 28] 256
ReLU-45 [-1, 128, 28, 28] 0
Conv2d-46 [-1, 512, 28, 28] 65,536
Block2-47 [-1, 512, 28, 28] 0
BatchNorm2d-48 [-1, 512, 28, 28] 1,024
ReLU-49 [-1, 512, 28, 28] 0
Identity-50 [-1, 512, 28, 28] 0
Conv2d-51 [-1, 128, 28, 28] 65,536
BatchNorm2d-52 [-1, 128, 28, 28] 256
ReLU-53 [-1, 128, 28, 28] 0
Conv2d-54 [-1, 128, 28, 28] 147,456
BatchNorm2d-55 [-1, 128, 28, 28] 256
ReLU-56 [-1, 128, 28, 28] 0
Conv2d-57 [-1, 512, 28, 28] 65,536
Block2-58 [-1, 512, 28, 28] 0
BatchNorm2d-59 [-1, 512, 28, 28] 1,024
ReLU-60 [-1, 512, 28, 28] 0
Identity-61 [-1, 512, 28, 28] 0
Conv2d-62 [-1, 128, 28, 28] 65,536
BatchNorm2d-63 [-1, 128, 28, 28] 256
ReLU-64 [-1, 128, 28, 28] 0
Conv2d-65 [-1, 128, 28, 28] 147,456
BatchNorm2d-66 [-1, 128, 28, 28] 256
ReLU-67 [-1, 128, 28, 28] 0
Conv2d-68 [-1, 512, 28, 28] 65,536
Block2-69 [-1, 512, 28, 28] 0
BatchNorm2d-70 [-1, 512, 28, 28] 1,024
ReLU-71 [-1, 512, 28, 28] 0
MaxPool2d-72 [-1, 512, 14, 14] 0
Conv2d-73 [-1, 128, 28, 28] 65,536
BatchNorm2d-74 [-1, 128, 28, 28] 256
ReLU-75 [-1, 128, 28, 28] 0
Conv2d-76 [-1, 128, 14, 14] 147,456
BatchNorm2d-77 [-1, 128, 14, 14] 256
ReLU-78 [-1, 128, 14, 14] 0
Conv2d-79 [-1, 512, 14, 14] 65,536
Block2-80 [-1, 512, 14, 14] 0
Stack2-81 [-1, 512, 14, 14] 0
BatchNorm2d-82 [-1, 512, 14, 14] 1,024
ReLU-83 [-1, 512, 14, 14] 0
Conv2d-84 [-1, 1024, 14, 14] 524,288
Conv2d-85 [-1, 256, 14, 14] 131,072
BatchNorm2d-86 [-1, 256, 14, 14] 512
ReLU-87 [-1, 256, 14, 14] 0
Conv2d-88 [-1, 256, 14, 14] 589,824
BatchNorm2d-89 [-1, 256, 14, 14] 512
ReLU-90 [-1, 256, 14, 14] 0
Conv2d-91 [-1, 1024, 14, 14] 262,144
Block2-92 [-1, 1024, 14, 14] 0
BatchNorm2d-93 [-1, 1024, 14, 14] 2,048
ReLU-94 [-1, 1024, 14, 14] 0
Identity-95 [-1, 1024, 14, 14] 0
Conv2d-96 [-1, 256, 14, 14] 262,144
BatchNorm2d-97 [-1, 256, 14, 14] 512
ReLU-98 [-1, 256, 14, 14] 0
Conv2d-99 [-1, 256, 14, 14] 589,824
BatchNorm2d-100 [-1, 256, 14, 14] 512
ReLU-101 [-1, 256, 14, 14] 0
Conv2d-102 [-1, 1024, 14, 14] 262,144
Block2-103 [-1, 1024, 14, 14] 0
BatchNorm2d-104 [-1, 1024, 14, 14] 2,048
ReLU-105 [-1, 1024, 14, 14] 0
Identity-106 [-1, 1024, 14, 14] 0
Conv2d-107 [-1, 256, 14, 14] 262,144
BatchNorm2d-108 [-1, 256, 14, 14] 512
ReLU-109 [-1, 256, 14, 14] 0
Conv2d-110 [-1, 256, 14, 14] 589,824
BatchNorm2d-111 [-1, 256, 14, 14] 512
ReLU-112 [-1, 256, 14, 14] 0
Conv2d-113 [-1, 1024, 14, 14] 262,144
Block2-114 [-1, 1024, 14, 14] 0
BatchNorm2d-115 [-1, 1024, 14, 14] 2,048
ReLU-116 [-1, 1024, 14, 14] 0
Identity-117 [-1, 1024, 14, 14] 0
Conv2d-118 [-1, 256, 14, 14] 262,144
BatchNorm2d-119 [-1, 256, 14, 14] 512
ReLU-120 [-1, 256, 14, 14] 0
Conv2d-121 [-1, 256, 14, 14] 589,824
BatchNorm2d-122 [-1, 256, 14, 14] 512
ReLU-123 [-1, 256, 14, 14] 0
Conv2d-124 [-1, 1024, 14, 14] 262,144
Block2-125 [-1, 1024, 14, 14] 0
BatchNorm2d-126 [-1, 1024, 14, 14] 2,048
ReLU-127 [-1, 1024, 14, 14] 0
Identity-128 [-1, 1024, 14, 14] 0
Conv2d-129 [-1, 256, 14, 14] 262,144
BatchNorm2d-130 [-1, 256, 14, 14] 512
ReLU-131 [-1, 256, 14, 14] 0
Conv2d-132 [-1, 256, 14, 14] 589,824
BatchNorm2d-133 [-1, 256, 14, 14] 512
ReLU-134 [-1, 256, 14, 14] 0
Conv2d-135 [-1, 1024, 14, 14] 262,144
Block2-136 [-1, 1024, 14, 14] 0
BatchNorm2d-137 [-1, 1024, 14, 14] 2,048
ReLU-138 [-1, 1024, 14, 14] 0
MaxPool2d-139 [-1, 1024, 7, 7] 0
Conv2d-140 [-1, 256, 14, 14] 262,144
BatchNorm2d-141 [-1, 256, 14, 14] 512
ReLU-142 [-1, 256, 14, 14] 0
Conv2d-143 [-1, 256, 7, 7] 589,824
BatchNorm2d-144 [-1, 256, 7, 7] 512
ReLU-145 [-1, 256, 7, 7] 0
Conv2d-146 [-1, 1024, 7, 7] 262,144
Block2-147 [-1, 1024, 7, 7] 0
Stack2-148 [-1, 1024, 7, 7] 0
BatchNorm2d-149 [-1, 1024, 7, 7] 2,048
ReLU-150 [-1, 1024, 7, 7] 0
Conv2d-151 [-1, 2048, 7, 7] 2,097,152
Conv2d-152 [-1, 512, 7, 7] 524,288
BatchNorm2d-153 [-1, 512, 7, 7] 1,024
ReLU-154 [-1, 512, 7, 7] 0
Conv2d-155 [-1, 512, 7, 7] 2,359,296
BatchNorm2d-156 [-1, 512, 7, 7] 1,024
ReLU-157 [-1, 512, 7, 7] 0
Conv2d-158 [-1, 2048, 7, 7] 1,048,576
Block2-159 [-1, 2048, 7, 7] 0
BatchNorm2d-160 [-1, 2048, 7, 7] 4,096
ReLU-161 [-1, 2048, 7, 7] 0
Identity-162 [-1, 2048, 7, 7] 0
Conv2d-163 [-1, 512, 7, 7] 1,048,576
BatchNorm2d-164 [-1, 512, 7, 7] 1,024
ReLU-165 [-1, 512, 7, 7] 0
Conv2d-166 [-1, 512, 7, 7] 2,359,296
BatchNorm2d-167 [-1, 512, 7, 7] 1,024
ReLU-168 [-1, 512, 7, 7] 0
Conv2d-169 [-1, 2048, 7, 7] 1,048,576
Block2-170 [-1, 2048, 7, 7] 0
BatchNorm2d-171 [-1, 2048, 7, 7] 4,096
ReLU-172 [-1, 2048, 7, 7] 0
Identity-173 [-1, 2048, 7, 7] 0
Conv2d-174 [-1, 512, 7, 7] 1,048,576
BatchNorm2d-175 [-1, 512, 7, 7] 1,024
ReLU-176 [-1, 512, 7, 7] 0
Conv2d-177 [-1, 512, 7, 7] 2,359,296
BatchNorm2d-178 [-1, 512, 7, 7] 1,024
ReLU-179 [-1, 512, 7, 7] 0
Conv2d-180 [-1, 2048, 7, 7] 1,048,576
Block2-181 [-1, 2048, 7, 7] 0
Stack2-182 [-1, 2048, 7, 7] 0
BatchNorm2d-183 [-1, 2048, 7, 7] 4,096
ReLU-184 [-1, 2048, 7, 7] 0
AdaptiveAvgPool2d-185 [-1, 2048, 1, 1] 0
Flatten-186 [-1, 2048] 0
Linear-187 [-1, 1000] 2,049,000
================================================================
Total params: 25,549,416
Trainable params: 25,549,416
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.57
Forward/backward pass size (MB): 241.69
Params size (MB): 97.46
Estimated Total Size (MB): 339.73
----------------------------------------------------------------三、 训练模型
1. 编写训练函数
python
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
python
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_loss3.正式训练
python
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：
python
Epoch: 1, Train_acc:38.3%, Train_loss:5.263, Test_acc:31.9%, Test_loss:5.151, Lr:1.00E-04
Epoch: 2, Train_acc:71.9%, Train_loss:1.759, Test_acc:31.0%, Test_loss:3.492, Lr:1.00E-04
Epoch: 3, Train_acc:82.7%, Train_loss:0.822, Test_acc:85.8%, Test_loss:0.620, Lr:1.00E-04
Epoch: 4, Train_acc:89.2%, Train_loss:0.478, Test_acc:83.2%, Test_loss:0.762, Lr:1.00E-04
Epoch: 5, Train_acc:89.2%, Train_loss:0.444, Test_acc:86.7%, Test_loss:0.629, Lr:1.00E-04
Epoch: 6, Train_acc:91.2%, Train_loss:0.359, Test_acc:73.5%, Test_loss:0.802, Lr:1.00E-04
Epoch: 7, Train_acc:95.1%, Train_loss:0.173, Test_acc:79.6%, Test_loss:0.689, Lr:1.00E-04
Epoch: 8, Train_acc:96.5%, Train_loss:0.141, Test_acc:80.5%, Test_loss:0.704, Lr:1.00E-04
Epoch: 9, Train_acc:98.5%, Train_loss:0.089, Test_acc:78.8%, Test_loss:0.879, Lr:1.00E-04
Epoch:10, Train_acc:95.8%, Train_loss:0.196, Test_acc:81.4%, Test_loss:0.718, Lr:1.00E-04
Done
预测结果是：Bananaquit
0.8672566371681416 0.5955437496304512
0.8672566371681416
Process finished with exit code 0四、 结果可视化
1. Loss&Accuracy
python
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：
python
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：
python
predict_one_image(image_path='./data/bird_photos/Bananaquit/007.jpg',
model=model,
transform=train_transforms,
classes=classes)得到如下结果：
python
预测结果是：Bananaquit3.模型评估
将模型调至评估模式：
python
best_model.eval()
epoch_test_acc, epoch_test_loss = test(test_dl, best_model, loss_fn)
print(epoch_test_acc, epoch_test_loss)得到如下输出：
python
0.8672566371681416 0.5955437496304512观察得到和前文中一致。
五、个人理解
除完成Tensorflow与pytorch之间的代码转换，还需了解ResNetV2及ResNetV之间的关系及区别：
首先，对比两个残差结构：
可以看出(a)结构先卷积后进行 BN 和激活函数计算，最后执行 addition 后再进行ReLU 计算； (b)结构先进行 BN 和激活函数计算后卷积，把 addition 后的 ReLU 计算放到了残差结构内部。
ResNetV2的最终确定经过了两轮尝试：
5.1关于残差结构的尝试
作者用不同 shortcut 结构的 ResNet-110 在 CIFAR-10 数据集上做测试，发现最原始的(a)original 结构是最好的，也就是 identity mapping 恒等映射是最好的
5.2关于激活的尝试
经实验发现，最好的结果是(e)full pre-activation，其次到(a)original。