- 🍨 本## 标题文为🔗365天深度学习训练营 中的学习记录博客
- 🍦 参考文章:Pytorch实战 | 第P5周:运动鞋识别
- 🍖 原作者:K同学啊 | 接辅导、项目定制
- 🚀 文章来源:K同学的学习圈子
注:本文加粗加斜体部分字体都是笔者基于训练营教案实践学习中的个人思考和一些资料的查找。
一、前期准备
1.设置GPU
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’)
1.os就是“operating system”的缩写,顾名思义,os模块提供的就是各种 Python 程序与操作系统进行交互的接口。通过使用os模块,一方面可以方便地与操作系统进行交互,另一方面页可以极大增强代码的可移植性。如果该模块中相关功能出错,会抛出OSError异常或其子类异常。转载自轩辕御龙的“Python os 模块详解”
2.PIL,全称 Python Imaging Library,是 Python 平台一个功能非常强大而且简单易用的图像处理库。来源于seniusen的“Python 学习笔记之—— PIL 库”
3.pathlib是跨平台的、面向对象的路径操作模块,可适用于不同的操作系统,其操作对象是各种操作系统中使用的路径(包括绝对路径和相对路径),pathlib有两个主要的类,分别为PurePath和Path。转载自Okmfj的“Python路径操作模块pathlib,看这篇就够了!”
2.导入数据
import os,PIL,random,pathlib
data_dir = './shoe brand/'
data_dir = pathlib.Path(data_dir)
data_paths = list(data_dir.glob('*'))
classeNames = [str(path).split("\\")[1] for path in data_paths]
classeNames
[‘test’, ‘train’]
● 第一步:使用pathlib.Path()函数将字符串类型的文件夹路径转换为pathlib.Path对象。
● 第二步:使用glob()方法获取data_dir路径下的所有文件路径,并以列表形式存储在data_paths中。
● 第三步:通过split()函数对data_paths中的每个文件路径执行分割操作,获得各个文件所属的类别名称,并存储在classeNames中
● 第四步:打印classeNames列表,显示每个文件所属的类别名称。
个人理解:在获取训练测试集的时候将类型统一,这是导入数据最基本的要求。最后的打印辅助检查的了数据上传是否成功和符合。
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。
])
train_dataset = datasets.ImageFolder("./5-data/train/",transform=train_transforms)
test_dataset = datasets.ImageFolder("./5-data/test/",transform=train_transforms)
torchvision是pytorch的一个图形库,它服务于PyTorch深度学习框架的,主要用来构建计算机视觉模型。torchvision.transforms主要是用于常见的一些图形变换。以下是torchvision的构成:
1.torchvision.datasets: 一些加载数据的函数及常用的数据集接口;
2.torchvision.models: 包含常用的模型结构(含预训练模型),例如AlexNet、VGG、ResNet等;
3.torchvision.transforms: 常用的图片变换,例如裁剪、旋转等;
4.image_mean=[0.485,0.456,0.406]、image_std=[0.229,0.224,0.225]是Imagenet数据集的均值和标准差
K同学啊torchvision.transforms.Compose()详解【Pytorch入门手册】
————————————————
版权声明:本文为CSDN博主「K同学啊」的原创文章,遵循CC 4.0 BY-SA版权协议,转载请附上原文出处链接及本声明。
原文链接:https://blog.csdn.net/qq_38251616/article/details/124878863
train_dataset.class_to_idx
{‘adidas’: 0, ‘nike’: 1}
train_dataset.class_to_idx 返回一个字典,即{root下包含的第一个文件夹:0,root下包含的第二个文件夹名:1,root下包含的第三个文件夹名:2}。
来自m0_59559920的深度学习2:加载数据集
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
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self):
super(Model, self).__init__()
self.conv1=nn.Sequential(
nn.Conv2d(3, 12, kernel_size=5, padding=0), # 12*220*220
nn.BatchNorm2d(12),
nn.ReLU())
self.conv2=nn.Sequential(
nn.Conv2d(12, 12, kernel_size=5, padding=0), # 12*216*216
nn.BatchNorm2d(12),
nn.ReLU())
self.pool3=nn.Sequential(
nn.MaxPool2d(2)) # 12*108*108
self.conv4=nn.Sequential(
nn.Conv2d(12, 24, kernel_size=5, padding=0), # 24*104*104
nn.BatchNorm2d(24),
nn.ReLU())
self.conv5=nn.Sequential(
nn.Conv2d(24, 24, kernel_size=5, padding=0), # 24*100*100
nn.BatchNorm2d(24),
nn.ReLU())
self.pool6=nn.Sequential(
nn.MaxPool2d(2)) # 24*50*50
self.dropout = nn.Sequential(
nn.Dropout(0.2))
self.fc=nn.Sequential(
nn.Linear(24*50*50, len(classeNames)))
def forward(self, x):
batch_size = x.size(0)
x = self.conv1(x) # 卷积-BN-激活
x = self.conv2(x) # 卷积-BN-激活
x = self.pool3(x) # 池化
x = self.conv4(x) # 卷积-BN-激活
x = self.conv5(x) # 卷积-BN-激活
x = self.pool6(x) # 池化
x = self.dropout(x)
x = x.view(batch_size, -1) # flatten 变成全连接网络需要的输入 (batch, 24*50*50) ==> (batch, -1), -1 此处自动算出的是24*50*50
x = self.fc(x)
return x
device = "cuda" if torch.cuda.is_available() else "cpu"
print("Using {} device".format(device))
model = Model().to(device)
model
构建的网络使用了四层卷积和两层池化,最后用x.view对tensor进行reshape,使输出结果更规范
Using cuda device Model( (conv1): Sequential(
(0): Conv2d(3, 12, kernel_size=(5, 5), stride=(1, 1))
(1): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU() ) (conv2): Sequential(
(0): Conv2d(12, 12, kernel_size=(5, 5), stride=(1, 1))
(1): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU() ) (pool3): Sequential(
(0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False) ) (conv4): Sequential(
(0): Conv2d(12, 24, kernel_size=(5, 5), stride=(1, 1))
(1): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU() ) (conv5): Sequential(
(0): Conv2d(24, 24, kernel_size=(5, 5), stride=(1, 1))
(1): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU() ) (pool6): Sequential(
(0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False) ) (dropout): Sequential(
(0): Dropout(p=0.2, inplace=False) ) (fc): Sequential(
(0): Linear(in_features=60000, out_features=2, bias=True) ) )
使用的cuda运行,将运行过程结果数据打印出来
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_loss
2.编写测试函数
def test (dataloader, model, loss_fn):
size = len(dataloader.dataset) # 测试集的大小
num_batches = len(dataloader) # 批次数目, (size/batch_size,向上取整)
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
测试集跟训练集的处理方式有很多类似,但它只是测试集,不用权重更新反向传播等,只需要一次测试就行,这些工作在训练集已经做过,测试集是检验训练成果的。
3.设置动态学习率
def adjust_learning_rate(optimizer, epoch, start_lr):
# 每 2 个epoch衰减到原来的 0.98
lr = start_lr * (0.92 ** (epoch // 2))
for param_group in optimizer.param_groups:
param_group['lr'] = lr
learn_rate = 1e-4 # 初始学习率
optimizer = torch.optim.SGD(model.parameters(), lr=learn_rate)
为什么要设置动态学习率?
拿打高尔夫球举例子,因为把球从起点打进洞和函数优化到最低点很像。把学习率比喻成击球的力气,整个高尔夫球场比喻成待优化的函数,球洞比作函数最低点。在起点的时候,球洞离我们很远,这时候为了减少击球的杆数,需要用大力把球朝着洞口的方向击打;在球离洞口很近的时候,则不需要使用那么大力气,而是需要精准地使用小力气把球朝着洞口的方向击打,不然使大力气很容易让球在洞口的周围来回变换位置但就是不进洞。通过上边的例子就说明了我们为什么要动态调整学习率,根本上还是要让网络朝着正确的方向加速收敛。不至于在最低点附近震荡,也不至于一直缓慢地朝着最低点的方向前进。
————————————
摘自CSDN博主「一颗磐石」的原创文章,遵循CC 4.0 BY-SA版权协议,转载请附上原文出处链接及本声明。
4.正式训练
loss_fn = nn.CrossEntropyLoss() # 创建损失函数
epochs = 40
train_loss = []
train_acc = []
test_loss = []
test_acc = []
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)
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))
print('Done')
这是我自己跑出来的数据,目前还不会调参数来增加学习率,下周争取学会
Epoch: 1, Train_acc:47.0%, Train_loss:0.798, Test_acc:51.3%, Test_loss:0.703, Lr:1.00E-04
Epoch: 2, Train_acc:61.6%, Train_loss:0.682, Test_acc:63.2%, Test_loss:0.622, Lr:1.00E-04
Epoch: 3, Train_acc:68.7%, Train_loss:0.601, Test_acc:67.1%, Test_loss:0.633, Lr:9.20E-05
Epoch: 4, Train_acc:69.7%, Train_loss:0.583, Test_acc:65.8%, Test_loss:0.577, Lr:9.20E-05
Epoch: 5, Train_acc:74.9%, Train_loss:0.540, Test_acc:67.1%, Test_loss:0.608, Lr:8.46E-05
Epoch: 6, Train_acc:75.5%, Train_loss:0.506, Test_acc:73.7%, Test_loss:0.537, Lr:8.46E-05
Epoch: 7, Train_acc:78.5%, Train_loss:0.494, Test_acc:72.4%, Test_loss:0.551, Lr:7.79E-05
Epoch: 8, Train_acc:82.5%, Train_loss:0.456, Test_acc:73.7%, Test_loss:0.521, Lr:7.79E-05
Epoch: 9, Train_acc:81.3%, Train_loss:0.445, Test_acc:73.7%, Test_loss:0.578, Lr:7.16E-05
Epoch:10, Train_acc:85.1%, Train_loss:0.417, Test_acc:76.3%, Test_loss:0.556, Lr:7.16E-05
Epoch:11, Train_acc:83.7%, Train_loss:0.412, Test_acc:76.3%, Test_loss:0.526, Lr:6.59E-05
Epoch:12, Train_acc:85.3%, Train_loss:0.397, Test_acc:75.0%, Test_loss:0.566, Lr:6.59E-05
Epoch:13, Train_acc:87.1%, Train_loss:0.384, Test_acc:76.3%, Test_loss:0.494, Lr:6.06E-05
Epoch:14, Train_acc:85.5%, Train_loss:0.389, Test_acc:77.6%, Test_loss:0.503, Lr:6.06E-05
Epoch:15, Train_acc:86.5%, Train_loss:0.377, Test_acc:76.3%, Test_loss:0.494, Lr:5.58E-05
Epoch:16, Train_acc:89.8%, Train_loss:0.363, Test_acc:76.3%, Test_loss:0.510, Lr:5.58E-05
Epoch:17, Train_acc:87.3%, Train_loss:0.365, Test_acc:78.9%, Test_loss:0.498, Lr:5.13E-05
Epoch:18, Train_acc:91.0%, Train_loss:0.342, Test_acc:78.9%, Test_loss:0.502, Lr:5.13E-05
Epoch:19, Train_acc:91.2%, Train_loss:0.326, Test_acc:77.6%, Test_loss:0.479, Lr:4.72E-05
Epoch:20, Train_acc:91.2%, Train_loss:0.326, Test_acc:77.6%, Test_loss:0.490, Lr:4.72E-05
Epoch:21, Train_acc:91.4%, Train_loss:0.329, Test_acc:80.3%, Test_loss:0.491, Lr:4.34E-05
Epoch:22, Train_acc:93.0%, Train_loss:0.320, Test_acc:78.9%, Test_loss:0.478, Lr:4.34E-05
Epoch:23, Train_acc:92.8%, Train_loss:0.316, Test_acc:76.3%, Test_loss:0.485, Lr:4.00E-05
Epoch:24, Train_acc:92.6%, Train_loss:0.307, Test_acc:78.9%, Test_loss:0.459, Lr:4.00E-05
Epoch:25, Train_acc:92.6%, Train_loss:0.295, Test_acc:78.9%, Test_loss:0.490, Lr:3.68E-05
Epoch:26, Train_acc:91.6%, Train_loss:0.312, Test_acc:80.3%, Test_loss:0.508, Lr:3.68E-05
Epoch:27, Train_acc:93.4%, Train_loss:0.297, Test_acc:81.6%, Test_loss:0.481, Lr:3.38E-05
Epoch:28, Train_acc:93.0%, Train_loss:0.295, Test_acc:80.3%, Test_loss:0.464, Lr:3.38E-05
Epoch:29, Train_acc:94.4%, Train_loss:0.286, Test_acc:80.3%, Test_loss:0.458, Lr:3.11E-05
Epoch:30, Train_acc:93.2%, Train_loss:0.288, Test_acc:80.3%, Test_loss:0.457, Lr:3.11E-05
Epoch:31, Train_acc:93.4%, Train_loss:0.279, Test_acc:81.6%, Test_loss:0.454, Lr:2.86E-05
Epoch:32, Train_acc:94.2%, Train_loss:0.286, Test_acc:81.6%, Test_loss:0.499, Lr:2.86E-05
Epoch:33, Train_acc:94.4%, Train_loss:0.272, Test_acc:81.6%, Test_loss:0.472, Lr:2.63E-05
Epoch:34, Train_acc:94.2%, Train_loss:0.275, Test_acc:80.3%, Test_loss:0.456, Lr:2.63E-05
Epoch:35, Train_acc:93.8%, Train_loss:0.272, Test_acc:80.3%, Test_loss:0.464, Lr:2.42E-05
Epoch:36, Train_acc:93.8%, Train_loss:0.265, Test_acc:81.6%, Test_loss:0.467, Lr:2.42E-05
Epoch:37, Train_acc:93.6%, Train_loss:0.276, Test_acc:81.6%, Test_loss:0.470, Lr:2.23E-05
Epoch:38, Train_acc:94.4%, Train_loss:0.267, Test_acc:81.6%, Test_loss:0.465, Lr:2.23E-05
Epoch:39, Train_acc:94.8%, Train_loss:0.266, Test_acc:80.3%, Test_loss:0.447, Lr:2.05E-05
Epoch:40, Train_acc:95.0%, Train_loss:0.257, Test_acc:80.3%, Test_loss:0.466, Lr:2.05E-05
Done
四、结果可视化
1. Loss与Accuracy图
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()
warnings.filterwarnings(“ignore”) 是 运用警告过滤器来避免代码正常运行被警告的情况
plt.rcParams可以将matplotlib中的默认的库的配置进行修改
plt.subplot()函数用于直接指定划分方式和位置进行绘图。
plt.plot是matplotlib.pyplot模块下的一个函数, 用于画图,如上面代码的plt.subplot(1, 2, 1),表示将整个图像窗口分为1行2列, 当前位置为1.
plt.legend用于创建图例
2.指定图片进行预测
>>> x = torch.zeros(2, 1, 2, 1, 2)
>>> x.size()
torch.Size([2, 1, 2, 1, 2])
>>> y = torch.squeeze(x)
>>> y.size()
torch.Size([2, 2, 2])
>>> y = torch.squeeze(x, 0)
>>> y.size()
torch.Size([2, 1, 2, 1, 2])
>>> y = torch.squeeze(x, 1)
>>> y.size()
torch.Size([2, 2, 1, 2])
from PIL import Image
classes = list(train_dataset.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}')
# 预测训练集中的某张照片
predict_one_image(image_path='./5-data/test/adidas/1.jpg',
model=model,
transform=train_transforms,
classes=classes)
预测结果是:adidas
torch.zeros()函数返回一个形状为为size,类型为torch.dtype,里面的每一个值都是0的tensor
squeeze()函数的功能是维度压缩。返回一个tensor(张量),其中 input 中大小为1的所有维都已删除。
五、保存并加载模型
# 模型保存
PATH = './model.pth' # 保存的参数文件名
torch.save(model.state_dict(), PATH)
# 将参数加载到model当中
model.load_state_dict(torch.load(PATH, map_location=device))
< All keys matched successfully>
近期学业上很忙,没有时间学习,之前没有掌握到学习神经网络的方法,现在初有眉目,之后找个时间系统地学习一下,好好调参数,增加准确率。