5、运动鞋识别
- 🍨 本文为🔗365天深度学习训练营 中的学习记录博客
- 🍦 参考文章：365天深度学习训练营-第P2周：彩色识别
- 🍖 原作者：K同学啊|接辅导、项目定制
In [1]:
import torch
import torch.nn as nn
import torchvision.transforms as transforms
import torchvision
from torchvision import transforms, datasets
from sklearn.model_selection import KFold
from torch.optim.lr_scheduler import StepLR, MultiStepLR, LambdaLR, ExponentialLR, CosineAnnealingLR, ReduceLROnPlateau
import os,PIL,pathlib,random
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
deviceOut[1]:
2、导入数据¶
In [2]:
data_dir = '../data/5-data'
# 通过Path类创建路径对象
data_dir = pathlib.Path(data_dir)
# 获取路径下所有文件路径
paths= list(data_dir.glob('*'))
# 获取所有文件夹的名字，也就是图片类别
classname = [str(path).split("\\")[3] for path in paths]
print(classname)
# 图像transforms
# 关于transforms.Compose的更多介绍可以参考：https://blog.csdn.net/qq_38251616/article/details/124878863
train_transforms = 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和std是从数据中随机抽样计算得到的，其实这个数值是pytorch上给的通用的统计值
])
test_transforms = 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和std是从数据中随机抽样计算得到的，其实这个数值是pytorch上给的通用的统计值
])
train_dataset = datasets.ImageFolder("../data/5-data/train/",transform=train_transforms)
test_dataset = datasets.ImageFolder("../data/5-data/test//",transform=train_transforms)In [3]:
train_dataset.class_to_idxOut[3]:
In [4]:
batch_size = 32
train_dl = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=3)
test_dl = torch.utils.data.DataLoader(test_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=3)In [5]:
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，其中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。
In [6]:
import torch.nn.functional as F
class Network_bn(nn.Module):
def __init__(self):
super(Network_bn, 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(classname))) # K同学这总是写错
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 = Network_bn().to(device)
modelOut[6]:
建立更好的CNN网络模型：加深、dropout与adam、lr_decay¶
第一次调参：¶
首先尝试将kernel_size调整为3，其次由于样本量并不大，考虑将dropout调大成0.5。但是由于一下子调整的好像有点狠，同样40个epoch难以很好的学习模型。
第二次调参：¶
本次针对epoch可能不太够选择了提高学习率，同时删去学习率衰减机制。虽然前期仅20余个epoch就让train_loss低于0.3，但是test_loss却一直卡住降不下去。后面试图通过恢复学习率衰减，但是并不能对test_loss产生很好的效果，反而让train的更快拟合完成。
第三次调参：¶
尝试使用AvgPool2d进行池化操作，但是不仅test_loss不好，连train_loss都进行的并不好。
第四次调参：¶
增加一层的dropout在第一次pooling之后，同时将学习率增大到1e-3，最终成功的在测试集上普遍达到了80%的准确率。
第五次调参：¶
调用Adam优化器进行优化，终于有最佳test_acc提升到86%以上，大量test_acc普遍在84%以上
Epoch:24, Train_acc:98.8%, Train_loss:0.046, Test_acc:85.5%, Test_loss:0.616, Lr:8.01E-04
第六次调参：¶
参照别人的博客文章，运动鞋识别-第五周。
- 将第一个dropout删去。
- 将动态学习率调整为每10个epoch衰减为原来的0.98
- 继续使用Adam优化器
但是对于我的模型效果并不好，因此还是采取原来第五次调参的结果。
In [18]:
import torch.nn.functional as F
class Network_bn(nn.Module):
def __init__(self):
super(Network_bn, self).__init__()
self.conv1=nn.Sequential(
nn.Conv2d(3, 12, kernel_size=3, padding=0), # 12*222*222
nn.BatchNorm2d(12),
nn.ReLU())
self.conv2=nn.Sequential(
nn.Conv2d(12, 12, kernel_size=3, padding=0), # 12*220*220
nn.BatchNorm2d(12),
nn.ReLU())
self.pool3=nn.Sequential(
nn.MaxPool2d(2)) # 12*110*110
self.dropout1 = nn.Sequential(
nn.Dropout(0.5))
self.conv4=nn.Sequential(
nn.Conv2d(12, 24, kernel_size=3, padding=0), # 24*108*108
nn.BatchNorm2d(24),
nn.ReLU())
self.conv5=nn.Sequential(
nn.Conv2d(24, 24, kernel_size=3, padding=0), # 24*106*106
nn.BatchNorm2d(24),
nn.ReLU())
self.pool6=nn.Sequential(
nn.MaxPool2d(2)) # 24*53*53
self.dropout2 = nn.Sequential(
nn.Dropout(0.5))
self.fc=nn.Sequential(
nn.Linear(24*53*53, len(classname))) # K同学这总是写错
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.dropout1(x)
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.dropout2(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 = Network_bn().to(device)
modelOut[18]:
In [22]:
def adjust_learning_rate(optimizer, epoch, start_lr):
# 每 2 个epoch衰减到原来的 0.98
lr = start_lr * (0.98 ** (epoch // 2))
for param_group in optimizer.param_groups:
param_group['lr'] = lr
learn_rate = 1e-3 # 初始学习率
optimizer = torch.optim.Adam(model.parameters(), lr=learn_rate, weight_decay = 0.01)2、编写训练函数¶
In [9]:
# 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，1875（60000/32）
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_loss3、编写测试函数¶
In [10]:
def test (dataloader, model, loss_fn):
size = len(dataloader.dataset) # 测试集的大小，一共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_loss4、正式训练¶
In [23]:
loss_fn = nn.CrossEntropyLoss() # 创建损失函数
epochs = 50
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')In [12]:
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()2、指定图片进行预测¶
In [13]:
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}')In [14]:
# 预测训练集中的某张照片
predict_one_image(image_path='E:/jupyter-notebook/data/5-data/test/adidas/22.jpg',
model=model,
transform=train_transforms,
classes=classes)五、保存并加载模型¶
In [15]:
# 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))Out[15]:
In [16]:
min_loss = 100000#随便设置一个比较大的数
for epoch in range(epochs):
train()
val_loss = val()
if val_loss < min_loss:
min_loss = val_loss
print("save model")
torch.save(net.state_dict(),'model.pth')
【推荐】国内首个AI IDE，深度理解中文开发场景，立即下载体验Trae
【推荐】编程新体验，更懂你的AI，立即体验豆包MarsCode编程助手
【推荐】抖音旗下AI助手豆包，你的智能百科全书，全免费不限次数
【推荐】轻量又高性能的 SSH 工具 IShell：AI 加持，快人一步
· 震惊！C++程序真的从main开始吗？99%的程序员都答错了
· 【硬核科普】Trae如何「偷看」你的代码？零基础破解AI编程运行原理
· 单元测试从入门到精通
· 上周热点回顾（3.3-3.9）
· winform 绘制太阳，地球，月球 运作规律