- 🍨 本文为🔗365天深度学习练习营 中的学习记录博客
- 🍦 参阅文章:Pytorch实战 | 第P3周:彩色图片辨认:气候辨认
- 🍖 原作者:K同学啊|接辅导、项目定制
⏲往期文章:
- 深度学习实战练习 | 第8周:猫狗辨认
- 深度学习实战练习 | 第7周:咖啡豆辨认
- 深度学习实战练习 | 第6周:好莱坞明星辨认
- 深度学习实战练习 | 第5周:运动鞋品牌辨认
☕ Pytorch实战
- Pytorch实战 | 第P1周:完成mnist手写数字辨认
- Pytorch实战 | 第P2周:彩色图片辨认
难度:新手入门⭐
🍺要求:
- 本地读取并加载数据。
- 测验集accuracy抵达93%
🍻拔高:
- 测验集accuracy抵达95%
- 调用模型辨认一张本地图片
🏡 我的环境:
- 语言环境:Python3.8
- 编译器:jupyter notebook
- 深度学习环境:Pytorch
- 数据:大众号:K同学啊
一、 前期准备
1. 设置GPU
假如设备上支撑GPU就运用GPU,否则运用CPU
import torch
import torch.nn as nn
import torchvision.transforms as transforms
import torchvision
from torchvision import transforms, datasets
import os,PIL,pathlib
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
device
device(type='cuda')
2. 导入数据
import os,PIL,random,pathlib
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]
classeNames
['cloudy', 'rain', 'shine', 'sunrise']
total_datadir = './data/'
# 关于transforms.Compose的更多介绍可以参阅:https://blog.csdn.net/qq_38251616/article/details/124878863
train_transforms = transforms.Compose([
transforms.Resize([224, 224]), # 将输入图片resize成一致尺度
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-->转换为标准正太散布(高斯散布),使模型更简单收敛
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]) # 其间 mean=[0.485,0.456,0.406]与std=[0.229,0.224,0.225] 从数据集中随机抽样核算得到的。
])
total_data = datasets.ImageFolder(total_datadir,transform=train_transforms)
total_data
Dataset ImageFolder
Number of datapoints: 1125
Root location: ./data/
StandardTransform
Transform: Compose(
Resize(size=[224, 224], interpolation=bilinear, max_size=None, antialias=None)
ToTensor()
Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
)
3. 划分数据集
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])
train_dataset, test_dataset
(<torch.utils.data.dataset.Subset at 0x1cd91e01ee0>,
<torch.utils.data.dataset.Subset at 0x1cd91e01f70>)
train_size,test_size
(900, 225)
batch_size = 32
train_dl = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=1)
test_dl = torch.utils.data.DataLoader(test_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=1)
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
Shape of X [N, C, H, W]: torch.Size([32, 3, 224, 224])
Shape of y: torch.Size([32]) torch.int64
二、构建简单的CNN网络
对于一般的CNN网络来说,都是由特征提取网络和分类网络构成,其间特征提取网络用于提取图片的特征,分类网络用于将图片进行分类。
⭐1. torch.nn.Conv2d()详解
函数原型:
torch.nn.Conv2d(in_channels, out_channels, kernel_size, stride=1, padding=0, dilation=1, groups=1, bias=True, padding_mode=’zeros’, device=None, dtype=None)
要害参数阐明:
- in_channels ( int ) – 输入图像中的通道数
- out_channels ( int ) – 卷积发生的通道数
- kernel_size ( int or tuple ) – 卷积核的巨细
- stride ( int or tuple , optional ) — 卷积的步幅。默许值:1
- padding ( int , tuple或str , optional ) – 添加到输入的一切四个边的填充。默许值:0
- padding_mode (字符串,可选) – ‘zeros’, ‘reflect’, ‘replicate’或’circular’. 默许:’zeros’
⭐2. torch.nn.Linear()详解
函数原型:
torch.nn.Linear(in_features, out_features, bias=True, device=None, dtype=None)
要害参数阐明:
- in_features:每个输入样本的巨细
- out_features:每个输出样本的巨细
⭐3. torch.nn.MaxPool2d()详解
函数原型:
torch.nn.MaxPool2d(kernel_size, stride=None, padding=0, dilation=1, return_indices=False, ceil_mode=False)
要害参数阐明:
- kernel_size:最大的窗口巨细
- stride:窗口的步幅,默许值为
kernel_size
- padding:填充值,默许为0
- dilation:操控窗口中元素步幅的参数
import torch.nn.functional as F
class Network_bn(nn.Module):
def __init__(self):
super(Network_bn, self).__init__()
"""
nn.Conv2d()函数:
第一个参数(in_channels)是输入的channel数量
第二个参数(out_channels)是输出的channel数量
第三个参数(kernel_size)是卷积核巨细
第四个参数(stride)是步长,默许为1
第五个参数(padding)是填充巨细,默许为0
"""
self.conv1 = nn.Conv2d(in_channels=3, out_channels=12, kernel_size=5, stride=1, padding=0)
self.bn1 = nn.BatchNorm2d(12)
self.conv2 = nn.Conv2d(in_channels=12, out_channels=12, kernel_size=5, stride=1, padding=0)
self.bn2 = nn.BatchNorm2d(12)
self.pool = nn.MaxPool2d(2,2)
self.conv4 = nn.Conv2d(in_channels=12, out_channels=24, kernel_size=5, stride=1, padding=0)
self.bn4 = nn.BatchNorm2d(24)
self.conv5 = nn.Conv2d(in_channels=24, out_channels=24, kernel_size=5, stride=1, padding=0)
self.bn5 = nn.BatchNorm2d(24)
self.fc1 = nn.Linear(24*50*50, len(classeNames))
def forward(self, x):
x = F.relu(self.bn1(self.conv1(x)))
x = F.relu(self.bn2(self.conv2(x)))
x = self.pool(x)
x = F.relu(self.bn4(self.conv4(x)))
x = F.relu(self.bn5(self.conv5(x)))
x = self.pool(x)
x = x.view(-1, 24*50*50)
x = self.fc1(x)
return x
device = "cuda" if torch.cuda.is_available() else "cpu"
print("Using {} device".format(device))
model = Network_bn().to(device)
model
Using cuda device
Network_bn(
(conv1): Conv2d(3, 12, kernel_size=(5, 5), stride=(1, 1))
(bn1): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(12, 12, kernel_size=(5, 5), stride=(1, 1))
(bn2): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(pool): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(conv4): Conv2d(12, 24, kernel_size=(5, 5), stride=(1, 1))
(bn4): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv5): Conv2d(24, 24, kernel_size=(5, 5), stride=(1, 1))
(bn5): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(fc1): Linear(in_features=60000, out_features=4, bias=True)
)
三、 练习模型
1. 设置超参数
loss_fn = nn.CrossEntropyLoss() # 创立丢失函数
learn_rate = 1e-4 # 学习率
opt = torch.optim.SGD(model.parameters(),lr=learn_rate)
2. 编写练习函数
1. optimizer.zero_grad()
函数会遍历模型的一切参数,经过内置办法截断反向传播的梯度流,再将每个参数的梯度值设为0,即上一次的梯度记录被清空。
2. loss.backward()
PyTorch的反向传播(即tensor.backward()
)是经过autograd包来完成的,autograd包会根据tensor进行过的数学运算来自动核算其对应的梯度。
具体来说,torch.tensor是autograd包的根底类,假如你设置tensor的requires_grads为True,就会开端盯梢这个tensor上面的一切运算,假如你做完运算后运用tensor.backward()
,一切的梯度就会自动运算,tensor的梯度将会累加到它的.grad特点里边去。
更具体地说,丢失函数loss是由模型的一切权重w经过一系列运算得到的,若某个w的requires_grads为True,则w的一切上层参数(后面层的权重w)的.grad_fn特点中就保存了对应的运算,然后在运用loss.backward()
后,会一层层的反向传播核算每个w的梯度值,并保存到该w的.grad特点中。
假如没有进行tensor.backward()
的话,梯度值将会是None,因而loss.backward()
要写在optimizer.step()
之前。
3. optimizer.step()
step()函数的作用是履行一次优化过程,经过梯度下降法来更新参数的值。因为梯度下降是基于梯度的,所以在履行optimizer.step()
函数前应先履行loss.backward()
函数来核算梯度。
留意:optimizer只担任经过梯度下降进行优化,而不担任发生梯度,梯度是tensor.backward()
办法发生的。
# 练习循环
def train(dataloader, model, loss_fn, optimizer):
size = len(dataloader.dataset) # 练习集的巨细,总共60000张图片
num_batches = len(dataloader) # 批次数目,1875(60000/32)
train_loss, train_acc = 0, 0 # 初始化练习丢失和正确率
for X, y in dataloader: # 获取图片及其标签
X, y = X.to(device), y.to(device)
# 核算猜测误差
pred = model(X) # 网络输出
loss = loss_fn(pred, y) # 核算网络输出和实在值之间的差距,targets为实在值,核算二者差值即为丢失
# 反向传播
optimizer.zero_grad() # grad特点归零
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_loss
3. 编写测验函数
测验函数和练习函数大致相同,可是因为不进行梯度下降对网络权重进行更新,所以不需求传入优化器
def test (dataloader, model, loss_fn):
size = len(dataloader.dataset) # 测验集的巨细,总共10000张图片
num_batches = len(dataloader) # 批次数目,313(10000/32=312.5,向上取整)
test_loss, test_acc = 0, 0
# 当不进行练习时,中止梯度更新,节省核算内存消耗
with torch.no_grad():
for imgs, target in dataloader:
imgs, target = imgs.to(device), target.to(device)
# 核算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
4. 正式练习
1. model.train()
model.train()
的作用是启用 Batch Normalization 和 Dropout。
假如模型中有BN
层(Batch Normalization)和Dropout
,需求在练习时添加model.train()
。model.train()
是确保BN层可以用到每一批数据的均值和方差。对于Dropout
,model.train()
是随机取一部分网络连接来练习更新参数。
2. model.eval()
model.eval()
的作用是不启用 Batch Normalization 和 Dropout。
假如模型中有BN层(Batch Normalization)和Dropout,在测验时添加model.eval()
。model.eval()
是确保BN层可以用全部练习数据的均值和方差,即测验过程中要确保BN层的均值和方差不变。对于Dropout
,model.eval()
是利用到了一切网络连接,即不进行随机舍弃神经元。
练习完train样本后,生成的模型model要用来测验样本。在model(test)
之前,需求加上model.eval()
,否则的话,有输入数据,即便不练习,它也会改变权值。这是model中含有BN层和Dropout所带来的的性质。
epochs = 20
train_loss = []
train_acc = []
test_loss = []
test_acc = []
for epoch in range(epochs):
model.train()
epoch_train_acc, epoch_train_loss = train(train_dl, model, loss_fn, opt)
model.eval()
epoch_test_acc, epoch_test_loss = test(test_dl, model, loss_fn)
train_acc.append(epoch_train_acc)
train_loss.append(epoch_train_loss)
test_acc.append(epoch_test_acc)
test_loss.append(epoch_test_loss)
template = ('Epoch:{:2d}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%,Test_loss:{:.3f}')
print(template.format(epoch+1, epoch_train_acc*100, epoch_train_loss, epoch_test_acc*100, epoch_test_loss))
print('Done')
Epoch: 1, Train_acc:61.4%, Train_loss:0.986, Test_acc:72.0%,Test_loss:0.865
Epoch: 2, Train_acc:76.7%, Train_loss:0.674, Test_acc:83.6%,Test_loss:0.558
Epoch: 3, Train_acc:80.8%, Train_loss:0.561, Test_acc:88.4%,Test_loss:0.447
Epoch: 4, Train_acc:83.6%, Train_loss:0.485, Test_acc:90.2%,Test_loss:0.431
Epoch: 5, Train_acc:86.3%, Train_loss:0.423, Test_acc:89.8%,Test_loss:0.354
Epoch: 6, Train_acc:86.3%, Train_loss:0.418, Test_acc:88.4%,Test_loss:0.306
Epoch: 7, Train_acc:87.6%, Train_loss:0.389, Test_acc:88.4%,Test_loss:0.401
Epoch: 8, Train_acc:90.0%, Train_loss:0.340, Test_acc:92.9%,Test_loss:0.488
Epoch: 9, Train_acc:90.7%, Train_loss:0.321, Test_acc:92.4%,Test_loss:0.260
Epoch:10, Train_acc:91.0%, Train_loss:0.316, Test_acc:92.9%,Test_loss:0.240
Epoch:11, Train_acc:92.6%, Train_loss:0.288, Test_acc:93.3%,Test_loss:0.254
Epoch:12, Train_acc:91.3%, Train_loss:0.291, Test_acc:92.4%,Test_loss:0.231
Epoch:13, Train_acc:93.9%, Train_loss:0.238, Test_acc:92.4%,Test_loss:0.226
Epoch:14, Train_acc:93.9%, Train_loss:0.255, Test_acc:93.3%,Test_loss:0.200
Epoch:15, Train_acc:93.7%, Train_loss:0.239, Test_acc:94.7%,Test_loss:0.236
Epoch:16, Train_acc:93.4%, Train_loss:0.224, Test_acc:93.3%,Test_loss:0.201
Epoch:17, Train_acc:94.1%, Train_loss:0.265, Test_acc:94.7%,Test_loss:0.187
Epoch:18, Train_acc:93.7%, Train_loss:0.222, Test_acc:94.2%,Test_loss:0.193
Epoch:19, Train_acc:95.4%, Train_loss:0.224, Test_acc:93.8%,Test_loss:0.199
Epoch:20, Train_acc:95.1%, Train_loss:0.201, Test_acc:93.3%,Test_loss:0.175
Done
四、 成果可视化
import matplotlib.pyplot as plt
#隐藏正告
import warnings
warnings.filterwarnings("ignore") #忽略正告信息
plt.rcParams['font.sans-serif'] = ['SimHei'] # 用来正常显现中文标签
plt.rcParams['axes.unicode_minus'] = False # 用来正常显现负号
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()