基于rsesnet网络架构的图像分类模型-编程知识

数据预处理部分：

数据增强：torchvision中transforms模块自带功能，比较实用
数据预处理：torchvision中transforms也帮我们实现好了，直接调用即可
DataLoader模块直接读取batch数据

网络模块设置：

加载预训练模型，torchvision中有很多经典网络架构，调用起来十分方便，并且可以用人家训练好的权重参数来继续训练，也就是所谓的迁移学习
需要注意的是别人训练好的任务跟咱们的可不是完全一样，需要把最后的head层改一改，一般也就是最后的全连接层，改成咱们自己的任务
训练时可以全部重头训练，也可以只训练最后咱们任务的层，因为前几层都是做特征提取的，本质任务目标是一致的
resnet只有18、50、101、152层的网络结构

网络模型保存与测试

模型保存的时候可以带有选择性，例如在验证集中如果当前效果好则保存
读取模型进行实际测试

[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-aWFTCjHl-1691645174490)(1.png)]

import os
import matplotlib.pyplot as plt
%matplotlib inline
import numpy as np
import torch
from torch import nn
import torch.optim as optim
import torchvision
#pip install torchvision
from torchvision import transforms, models, datasets
#https://pytorch.org/docs/stable/torchvision/index.html
import imageio
import time
import warnings
warnings.filterwarnings("ignore")
import random
import sys
import copy
import json
from PIL import Image

数据读取与预处理操作

data_dir = './flower_data/'
train_dir = data_dir + '/train'
valid_dir = data_dir + '/valid'

制作好数据源：

data_transforms中指定了所有图像预处理操作
ImageFolder假设所有的文件按文件夹保存好，每个文件夹下面存贮同一类别的图片，文件夹的名字为分类的名字

data_transforms = {'train': transforms.Compose([transforms.Resize([96, 96]),#将每张图片转化为大小相同，但是肯定会丢失一些信息 transforms.RandomRotation(45),#随机旋转，-45到45度之间随机选transforms.CenterCrop(64),#从中心开始裁剪transforms.RandomHorizontalFlip(p=0.5),#随机水平翻转 选择一个概率概率transforms.RandomVerticalFlip(p=0.5),#随机垂直翻转transforms.ColorJitter(brightness=0.2, contrast=0.1, saturation=0.1, hue=0.1),#参数1为亮度，参数2为对比度，参数3为饱和度，参数4为色相transforms.RandomGrayscale(p=0.025),#概率转换成灰度率，3通道就是R=G=Btransforms.ToTensor(),transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])#均值，标准差，这些参数主要是基于大数据算出来的]),'valid': transforms.Compose([transforms.Resize([64, 64]),transforms.ToTensor(),transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])]),
}

batch_size = 128image_datasets = {x: datasets.ImageFolder(os.path.join(data_dir, x), data_transforms[x]) for x in ['train', 'valid']}
dataloaders = {x: torch.utils.data.DataLoader(image_datasets[x], batch_size=batch_size, shuffle=True) for x in ['train', 'valid']}
dataset_sizes = {x: len(image_datasets[x]) for x in ['train', 'valid']}
class_names = image_datasets['train'].classes

读取标签对应的实际名字

#读取标签对应的实际名字
with open('cat_to_name.json','r') as f:cat_to_name = json.load(f)

加载models中提供的模型，并且直接用训练的好权重当做初始化参数

model_name = 'resnet'  #可选的比较多 ['resnet', 'alexnet', 'vgg', 'squeezenet', 'densenet', 'inception']
#是否用人家训练好的特征来做
feature_extract = True #都用人家特征，咱先不更新

# 是否用GPU训练
train_on_gpu = torch.cuda.is_available()if not train_on_gpu:print('CUDA is not available.  Training on CPU ...')
else:print('CUDA is available!  Training on GPU ...')device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

模型参数要不要更新

有时候用人家模型，就一直用了，更不更新咱们可以自己定

model_ft = models.resnet18()#18层的能快点，条件好点的也可以选152
model_ft

def set_parameter_requires_grad(model, feature_extracting):if feature_extracting:for param in model.parameters():param.requires_grad = False #设置成false的话，在反向传播的过程种，参数就不再进行更新了

for name,param in model_ft.named_parameters():if param.requires_grad == True:print("\t",name)

把模型输出层改成自己的

def initialize_model(model_name, num_classes, feature_extract, use_pretrained=True):model_ft = models.resnet18(pretrained=use_pretrained)set_parameter_requires_grad(model_ft, feature_extract)num_ftrs = model_ft.fc.in_featuresmodel_ft.fc = nn.Linear(num_ftrs, num_classes)#类别数自己根据自己任务来input_size = 64#输入大小根据自己配置来return model_ft, input_size

设置哪些层需要训练

model_ft, input_size = initialize_model(model_name, 102, feature_extract, use_pretrained=True)#GPU还是CPU计算
model_ft = model_ft.to(device)# 模型保存，名字自己起
filename='best.pt'# 是否训练所有层
params_to_update = model_ft.parameters()
print("Params to learn:")
if feature_extract:params_to_update = []for name,param in model_ft.named_parameters():if param.requires_grad == True:params_to_update.append(param)print("\t",name)
else:for name,param in model_ft.named_parameters():if param.requires_grad == True:print("\t",name)

优化器设置

# 优化器设置
optimizer_ft = optim.Adam(params_to_update, lr=1e-2)#要训练啥 参数，你来定
scheduler = optim.lr_scheduler.StepLR(optimizer_ft, step_size=10, gamma=0.1)#学习率每7个epoch衰减成原来的1/10
criterion = nn.CrossEntropyLoss()

训练模块

def train_model(model, dataloaders, criterion, optimizer, num_epochs=25,filename='best.pt'):#咱们要算时间的since = time.time()#也要记录最好的那一次best_acc = 0#模型也得放到你的CPU或者GPUmodel.to(device)#训练过程中打印一堆损失和指标val_acc_history = []train_acc_history = []train_losses = []valid_losses = []#学习率LRs = [optimizer.param_groups[0]['lr']]#最好的那次模型，后续会变的，先初始化best_model_wts = copy.deepcopy(model.state_dict())#一个个epoch来遍历for epoch in range(num_epochs):print('Epoch {}/{}'.format(epoch, num_epochs - 1))print('-' * 10)# 训练和验证for phase in ['train', 'valid']:if phase == 'train':model.train()  # 训练else:model.eval()   # 验证running_loss = 0.0running_corrects = 0# 把数据都取个遍for inputs, labels in dataloaders[phase]:inputs = inputs.to(device)#放到你的CPU或GPUlabels = labels.to(device)# 清零optimizer.zero_grad()# 只有训练的时候计算和更新梯度outputs = model(inputs)loss = criterion(outputs, labels)_, preds = torch.max(outputs, 1)# 训练阶段更新权重if phase == 'train':loss.backward()optimizer.step()# 计算损失running_loss += loss.item() * inputs.size(0)#0表示batch那个维度running_corrects += torch.sum(preds == labels.data)#预测结果最大的和真实值是否一致epoch_loss = running_loss / len(dataloaders[phase].dataset)#算平均epoch_acc = running_corrects.double() / len(dataloaders[phase].dataset)time_elapsed = time.time() - since#一个epoch我浪费了多少时间print('Time elapsed {:.0f}m {:.0f}s'.format(time_elapsed // 60, time_elapsed % 60))print('{} Loss: {:.4f} Acc: {:.4f}'.format(phase, epoch_loss, epoch_acc))# 得到最好那次的模型if phase == 'valid' and epoch_acc > best_acc:best_acc = epoch_accbest_model_wts = copy.deepcopy(model.state_dict())state = {'state_dict': model.state_dict(),#字典里key就是各层的名字，值就是训练好的权重'best_acc': best_acc,'optimizer' : optimizer.state_dict(),}torch.save(state, filename)if phase == 'valid':val_acc_history.append(epoch_acc)valid_losses.append(epoch_loss)#scheduler.step(epoch_loss)#学习率衰减if phase == 'train':train_acc_history.append(epoch_acc)train_losses.append(epoch_loss)print('Optimizer learning rate : {:.7f}'.format(optimizer.param_groups[0]['lr']))LRs.append(optimizer.param_groups[0]['lr'])print()scheduler.step()#学习率衰减time_elapsed = time.time() - sinceprint('Training complete in {:.0f}m {:.0f}s'.format(time_elapsed // 60, time_elapsed % 60))print('Best val Acc: {:4f}'.format(best_acc))# 训练完后用最好的一次当做模型最终的结果,等着一会测试model.load_state_dict(best_model_wts)return model, val_acc_history, train_acc_history, valid_losses, train_losses, LRs

开始训练！

我们现在只训练了输出层

model_ft, val_acc_history, train_acc_history, valid_losses, train_losses, LRs  = train_model(model_ft, dataloaders, criterion, optimizer_ft, num_epochs=2)

再继续训练所有层

for param in model_ft.parameters():param.requires_grad = True# 再继续训练所有的参数，学习率调小一点
optimizer = optim.Adam(model_ft.parameters(), lr=1e-3)
scheduler = optim.lr_scheduler.StepLR(optimizer_ft, step_size=7, gamma=0.1)# 损失函数
criterion = nn.CrossEntropyLoss()

# 加载之前训练好的权重参数
checkpoint = torch.load(filename)
best_acc = checkpoint['best_acc']
model_ft.load_state_dict(checkpoint['state_dict'])

model_ft, val_acc_history, train_acc_history, valid_losses, train_losses, LRs  = train_model(model_ft, dataloaders, criterion, optimizer, num_epochs=10,)

加载训练好的模型

model_ft, input_size = initialize_model(model_name, 102, feature_extract, use_pretrained=True)# GPU模式
model_ft = model_ft.to(device)# 保存文件的名字
filename='best.pt'# 加载模型
checkpoint = torch.load(filename)
best_acc = checkpoint['best_acc']
model_ft.load_state_dict(checkpoint['state_dict'])

测试数据预处理

测试数据处理方法需要跟训练时一直才可以
crop操作的目的是保证输入的大小是一致的
标准化操作也是必须的，用跟训练数据相同的mean和std,但是需要注意一点训练数据是在0-1上进行标准化，所以测试数据也需要先归一化
最后一点，PyTorch中颜色通道是第一个维度，跟很多工具包都不一样，需要转换

# 得到一个batch的测试数据
dataiter = iter(dataloaders['valid'])
images, labels = dataiter.next()model_ft.eval()if train_on_gpu:output = model_ft(images.cuda())
else:output = model_ft(images)

output表示对一个batch中每一个数据得到其属于各个类别的可能性

output.shape

得到概率最大的那个

_, preds_tensor = torch.max(output, 1)preds = np.squeeze(preds_tensor.numpy()) if not train_on_gpu else np.squeeze(preds_tensor.cpu().numpy())
preds

展示预测结果

def im_convert(tensor):""" 展示数据"""image = tensor.to("cpu").clone().detach()image = image.numpy().squeeze()image = image.transpose(1,2,0)image = image * np.array((0.229, 0.224, 0.225)) + np.array((0.485, 0.456, 0.406))image = image.clip(0, 1)return image

fig=plt.figure(figsize=(20, 20))
columns =4
rows = 2for idx in range (columns*rows):ax = fig.add_subplot(rows, columns, idx+1, xticks=[], yticks=[])plt.imshow(im_convert(images[idx]))ax.set_title("{} ({})".format(cat_to_name[str(preds[idx])], cat_to_name[str(labels[idx].item())]),color=("green" if cat_to_name[str(preds[idx])]==cat_to_name[str(labels[idx].item())] else "red"))
plt.show()

在这里插入图片描述 ### 完整代码

import os
import matplotlib.pyplot as plt
import numpy as np
import torch
from torch import nn
import torch.optim as optim
import torchvision
#pip install torchvision
from torchvision import transforms, models, datasets
#https://pytorch.org/docs/stable/torchvision/index.html
import imageio
import time
import warnings
warnings.filterwarnings("ignore")
import random
import sys
import copy
import json
from PIL import Image#检验torch（GPU）是否可以用
print(torch.cuda.is_available())#读取数据
data_dir = "./flower_data/"
train_dir = data_dir + '/train'
valid_dir = data_dir + '/valid'#制作数据源
data_transforms = {'train':transforms.Compose([transforms.Resize([96, 96]),#将每张图片转化为大小相同，但是肯定会丢失一些信息transforms.RandomRotation(45),#随机旋转，-45到45度之间随机选transforms.CenterCrop(64),#从中心开始裁剪transforms.RandomHorizontalFlip(p=0.5),#随机水平翻转 选择一个概率概率transforms.RandomVerticalFlip(p=0.5),#随机垂直翻转transforms.ColorJitter(brightness=0.2, contrast=0.1, saturation=0.1, hue=0.1),#参数1为亮度，参数2为对比度，参数3为饱和度，参数4为色相transforms.RandomGrayscale(p=0.025),#概率转换成灰度率，3通道就是R=G=Btransforms.ToTensor(),transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])#均值，标准差，这些参数主要是基于大数据算出来的]),'valid':transforms.Compose([transforms.Resize([64, 64]),transforms.ToTensor(),transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])]),
}
batch_size = 128
image_datasets = {x: datasets.ImageFolder(os.path.join(data_dir, x), data_transforms[x]) for x in ['train', 'valid']}
dataloaders = {x: torch.utils.data.DataLoader(image_datasets[x], batch_size=batch_size, shuffle=True) for x in ['train', 'valid']}
dataset_sizes = {x: len(image_datasets[x]) for x in ['train', 'valid']}
class_names = image_datasets['train'].classes#读取标签对应的实际名字
with open('cat_to_name.json','r') as f:cat_to_name = json.load(f)model_name = 'resnet'  #可选的比较多 ['resnet', 'alexnet', 'vgg', 'squeezenet', 'densenet', 'inception']
#是否用人家训练好的特征来做
feature_extract = True #都用人家特征，咱先不更新# 是否用GPU训练
train_on_gpu = torch.cuda.is_available()
if not train_on_gpu:print('CUDA is not available.  Training on CPU ...')
else:print('CUDA is available!  Training on GPU ...')
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")#模型参数要不要更新
#有时候用人家模型，就一直用了，更不更新咱们可以自己定
model_ft = models.resnet18()#18层的能快点，条件好点的也可以选152def set_parameter_requires_grad(model,feature_extracting):if feature_extracting:for param in model.parameters():param.requires_grad = False#设置成false的话，在反向传播的过程种，参数就不再进行更新了# 把模型输出层改成自己的
def initialize_model(model_name,num_classes,feature_extract, use_pretrained=True):model_ft = models.resnet18(pretrained=use_pretrained)set_parameter_requires_grad(model_name, feature_extract)num_ftrs = model_name.fc.in_featuresmodel_name.fc = nn.Linear(num_ftrs, num_classes)  # 类别数自己根据自己任务来input_size = 64  # 输入大小根据自己配置来return model_name,input_sizemodel_ft, input_size = initialize_model(model_ft, 102, feature_extract, use_pretrained=True)
#GPU还是CPU计算
model_ft = model_ft.to(device)
# 模型保存，名字自己起
filename='best.pt'#设置哪些层需要训练
if feature_extract:params_to_update = []for name,param in model_ft.named_parameters():if param.requires_grad == True:params_to_update.append(param)print("\t",name)
else:for name,param in model_ft.named_parameters():if param.requires_grad == True:print("\t",name)optimizer_ft = optim.Adam(params=params_to_update,lr =1e-2)#要训练啥 参数，你来定
scheduler = optim.lr_scheduler.StepLR(optimizer_ft, step_size=10, gamma=0.1)#学习率每7个epoch衰减成原来的1/10
criterion = nn.CrossEntropyLoss()#训练模块
def train_model(model, dataloaders, criterion, optimizer, num_epochs=25,filename='best.pt'):#咱们要算时间的since = time.time()#也要记录最好的那一次best_acc = 0#模型也得放到你的CPU或者GPUmodel.to(device)#训练过程中打印一堆损失和指标val_acc_history = []train_acc_history = []train_losses = []valid_losses = []#学习率LRs = [optimizer.param_groups[0]['lr']]#最好的那次模型，后续会变的，先初始化best_model_wts = copy.deepcopy(model.state_dict())#一个个epoch来遍历for epoch in range(num_epochs):print(f"Epoch {epoch}/{num_epochs - 1}")print('-'*10)#训练和验证# 训练和验证for phase in ['train', 'valid']:if phase == 'train':model.train()  # 训练else:model.eval()  # 验证running_loss = 0.0running_corrects = 0# 把数据都取个遍for inputs, labels in dataloaders[phase]:inputs = inputs.to(device)  # 放到你的CPU或GPUlabels = labels.to(device)# 清零optimizer.zero_grad()# 只有训练的时候计算和更新梯度outputs = model(inputs)loss = criterion(outputs, labels)_, preds = torch.max(outputs, 1)# 训练阶段更新权重if phase == 'train':loss.backward()optimizer.step()# 计算损失running_loss += loss.item() * inputs.size(0)  # 0表示batch那个维度running_corrects += torch.sum(preds == labels.data)  # 预测结果最大的和真实值是否一致epoch_loss = running_loss / len(dataloaders[phase].dataset)  # 算平均epoch_acc = running_corrects.double() / len(dataloaders[phase].dataset)time_elapsed = time.time() - since  # 一个epoch我浪费了多少时间print('Time elapsed {:.0f}m {:.0f}s'.format(time_elapsed // 60, time_elapsed % 60))print('{} Loss: {:.4f} Acc: {:.4f}'.format(phase, epoch_loss, epoch_acc))# 得到最好那次的模型if phase == 'valid' and epoch_acc > best_acc:best_acc = epoch_accbest_model_wts = copy.deepcopy(model.state_dict())state = {'state_dict': model.state_dict(),  # 字典里key就是各层的名字，值就是训练好的权重'best_acc': best_acc,'optimizer': optimizer.state_dict(),}torch.save(state, filename)if phase == 'valid':val_acc_history.append(epoch_acc)valid_losses.append(epoch_loss)# scheduler.step(epoch_loss)#学习率衰减if phase == 'train':train_acc_history.append(epoch_acc)train_losses.append(epoch_loss)print('Optimizer learning rate : {:.7f}'.format(optimizer.param_groups[0]['lr']))LRs.append(optimizer.param_groups[0]['lr'])print()scheduler.step()#学习率衰减time_elapsed = time.time() - sinceprint('Training complete in {:.0f}m {:.0f}s'.format(time_elapsed // 60, time_elapsed % 60))print('Best val Acc: {:4f}'.format(best_acc))# 训练完后用最好的一次当做模型最终的结果,等着一会测试model.load_state_dict(best_model_wts)return model, val_acc_history, train_acc_history, valid_losses, train_losses, LRs#开始训练模型
model_ft, val_acc_history, train_acc_history, valid_losses, train_losses, LRs  = train_model(model_ft, dataloaders, criterion, optimizer_ft, num_epochs=2)#再继续训练所有层
for param in model_ft.parameters():param.requires_grad = True
# 再继续训练所有的参数，学习率调小一点
optimizer = optim.Adam(model_ft.parameters(), lr=1e-3)
scheduler = optim.lr_scheduler.StepLR(optimizer_ft, step_size=7, gamma=0.1)
# 损失函数
criterion = nn.CrossEntropyLoss()# 加载之前训练好的权重参数
checkpoint = torch.load(filename)
best_acc = checkpoint['best_acc']
model_ft.load_state_dict(checkpoint['state_dict'])model_ft, val_acc_history, train_acc_history, valid_losses, train_losses, LRs  = train_model(model_ft, dataloaders, criterion, optimizer, num_epochs=10,)#加载训练好的模型
model_ft, input_size = initialize_model(model_ft, 102, feature_extract, use_pretrained=True)
# GPU模式
model_ft = model_ft.to(device)
# 保存文件的名字
filename='best.pt'
# 加载模型
checkpoint = torch.load(filename)
best_acc = checkpoint['best_acc']
model_ft.load_state_dict(checkpoint['state_dict'])# 测试数据预处理
# 得到一个batch的测试数据
dataiter = iter(dataloaders['valid'])
images, labels = dataiter.next()model_ft.eval()if train_on_gpu:output = model_ft(images.cuda())
else:output = model_ft(images)_, preds_tensor = torch.max(output, 1)# 得到概率最大的那个
preds = np.squeeze(preds_tensor.numpy()) if not train_on_gpu else np.squeeze(preds_tensor.cpu().numpy())#展示预测的结果
def im_convert(tensor):""" 展示数据"""image = tensor.to("cpu").clone().detach()image = image.numpy().squeeze()image = image.transpose(1, 2, 0)image = image * np.array((0.229, 0.224, 0.225)) + np.array((0.485, 0.456, 0.406))image = image.clip(0, 1)return imagefig=plt.figure(figsize=(20, 20))
columns =4
rows = 2for idx in range (columns*rows):ax = fig.add_subplot(rows, columns, idx+1, xticks=[], yticks=[])plt.imshow(im_convert(images[idx]))ax.set_title("{} ({})".format(cat_to_name[str(preds[idx])], cat_to_name[str(labels[idx].item())]),color=("green" if cat_to_name[str(preds[idx])]==cat_to_name[str(labels[idx].item())] else "red"))
plt.show()