Today, a picture classification work was carried out using resnet50, and the code is now recorded. In the first half of this article, quickly build a resnet network for training. This part of the code mainly refers to the complete code of the blog using the resnet50 network to train multi-classification models . I have made certain modifications and added comments to the code based on this blog; in addition, in this blog The latter part of the content is mainly a script written by myself to quickly build the previously trained resnet50 network model and perform image detection.
1. The resnet50 model is quickly built and the data set is trained
The rapid deployment here mainly uses the resnet50 pre-trained network that comes with pytorch, which saves the process of adding structures in turn.
Regarding the folder storage structure of the data set before training preparation, as described in the original blog:
--train
|--dog
|--cat ...
--valid
|--dog
|--cat
Create a data folder (the name can be arbitrary) under the project root directory to store the data set.
Create train, valid, and test folders in sequence under the data folder (the test folder can be absent, determined according to your own needs)
Create category folders under the train folder, such as cat, dog, etc.
, and store them in category folders such as cat An image of the cat category.
…
Create category folders under the val folder, such as cat, dog, etc.
Under the category folder such as cat, store pictures of the cat category.
…
——————————————————
Note
: This quote is partly excerpted from the original article of CSDN blogger "Working Quietly"
Original article link: https://blog.csdn.net/weixin_46034990 /article/details/124859877
In addition, it should be noted that a new file named models should be created under the root directory of the code folder to save the model file during the training process. At the same time, according to your own training set, you need to customize and modify the variable num_classes of the number of categories of the target classification in the code.
The following shows the model deployment and training code:
import torch
from torchvision import datasets, models, transforms
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader
import time
import numpy as np
import matplotlib.pyplot as plt
import os
from tqdm import tqdm
# 一、建立数据集
# animals-6
# --train
# |--dog
# |--cat
# ...
# --valid
# |--dog
# |--cat
# ...
# --test
# |--dog
# |--cat
# ...
# 我的数据集中 train 中每个类别60张图片，valid 中每个类别 10 张图片，test 中每个类别几张到几十张不等，一共 6 个类别。
# 二、数据增强
# 建好的数据集在输入网络之前先进行数据增强，包括随机 resize 裁剪到 256 x 256，随机旋转，随机水平翻转，中心裁剪到 224 x 224，转化成 Tensor，正规化等。
image_transforms = {
'train': transforms.Compose([
transforms.RandomResizedCrop(size=256, scale=(0.8, 1.0)),
transforms.RandomRotation(degrees=15),
transforms.RandomHorizontalFlip(),
transforms.CenterCrop(size=224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]),
'valid': transforms.Compose([
transforms.Resize(size=256),
transforms.CenterCrop(size=224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
])
}
# 三、加载数据
# torchvision.transforms包DataLoader是 Pytorch 重要的特性，它们使得数据增加和加载数据变得非常简单。
# 使用 DataLoader 加载数据的时候就会将之前定义的数据 transform 就会应用的数据上了。
dataset = 'data'
train_directory = os.path.join(dataset, 'train')
valid_directory = os.path.join(dataset, 'valid')
batch_size = 32
num_classes = 2 #分类种类数
print(train_directory)
data = {
'train': datasets.ImageFolder(root=train_directory, transform=image_transforms['train']),
'valid': datasets.ImageFolder(root=valid_directory, transform=image_transforms['valid'])
}
print("训练集图片类别及其对应编号(种类名：编号)：",data['train'].class_to_idx)
print("测试集图片类别及其对应编号：",data['valid'].class_to_idx)
train_data_size = len(data['train'])
valid_data_size = len(data['valid'])
train_data = DataLoader(data['train'], batch_size=batch_size, shuffle=True, num_workers=8)
valid_data = DataLoader(data['valid'], batch_size=batch_size, shuffle=True, num_workers=8)
print("训练集图片数量：",train_data_size, "测试集图片数量：",valid_data_size)
# 四、迁移学习
# 这里使用ResNet-50的预训练模型。
resnet50 = models.resnet50(pretrained=True)
# 在PyTorch中加载模型时，所有参数的‘requires_grad’字段默认设置为true。这意味着对参数值的每一次更改都将被存储，以便在用于训练的反向传播图中使用。
# 这增加了内存需求。由于预训练的模型中的大多数参数已经训练好了，因此将requires_grad字段重置为false。
for param in resnet50.parameters():
param.requires_grad = False
# 为了适应自己的数据集，将ResNet-50的最后一层替换为，将原来最后一个全连接层的输入喂给一个有256个输出单元的线性层，接着再连接ReLU层和Dropout层，然后是256 x 6的线性层，输出为6通道的softmax层。
fc_inputs = resnet50.fc.in_features
resnet50.fc = nn.Sequential(
nn.Linear(fc_inputs, 256),
nn.ReLU(),
nn.Dropout(0.4),
nn.Linear(256, num_classes),
nn.LogSoftmax(dim=1)
)
# 用GPU进行训练。
resnet50 = resnet50.to('cuda:0')
# 定义损失函数和优化器。
loss_func = nn.NLLLoss()
optimizer = optim.Adam(resnet50.parameters())
# 五、训练
def train_and_valid(model, loss_function, optimizer, epochs=25):
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
history = []
best_acc = 0.0
best_epoch = 0
for epoch in range(epochs):
epoch_start = time.time()
print("Epoch: {}/{}".format(epoch+1, epochs))
model.train()
train_loss = 0.0
train_acc = 0.0
valid_loss = 0.0
valid_acc = 0.0
for i, (inputs, labels) in enumerate(tqdm(train_data)):
inputs = inputs.to(device)
labels = labels.to(device)
#因为这里梯度是累加的，所以每次记得清零
optimizer.zero_grad()
outputs = model(inputs)
loss = loss_function(outputs, labels)
print("标签值：",labels)
print("输出值：",outputs)
loss.backward()
optimizer.step()
train_loss += loss.item() * inputs.size(0)
ret, predictions = torch.max(outputs.data, 1)
correct_counts = predictions.eq(labels.data.view_as(predictions))
acc = torch.mean(correct_counts.type(torch.FloatTensor))
train_acc += acc.item() * inputs.size(0)
with torch.no_grad():
model.eval()
for j, (inputs, labels) in enumerate(tqdm(valid_data)):
inputs = inputs.to(device)
labels = labels.to(device)
outputs = model(inputs)
loss = loss_function(outputs, labels)
valid_loss += loss.item() * inputs.size(0)
ret, predictions = torch.max(outputs.data, 1)
correct_counts = predictions.eq(labels.data.view_as(predictions))
acc = torch.mean(correct_counts.type(torch.FloatTensor))
valid_acc += acc.item() * inputs.size(0)
avg_train_loss = train_loss/train_data_size
avg_train_acc = train_acc/train_data_size
avg_valid_loss = valid_loss/valid_data_size
avg_valid_acc = valid_acc/valid_data_size
history.append([avg_train_loss, avg_valid_loss, avg_train_acc, avg_valid_acc])
if best_acc < avg_valid_acc:
best_acc = avg_valid_acc
best_epoch = epoch + 1
epoch_end = time.time()
print("Epoch: {:03d}, Training: Loss: {:.4f}, Accuracy: {:.4f}%, \n\t\tValidation: Loss: {:.4f}, Accuracy: {:.4f}%, Time: {:.4f}s".format(
epoch+1, avg_valid_loss, avg_train_acc*100, avg_valid_loss, avg_valid_acc*100, epoch_end-epoch_start
))
print("Best Accuracy for validation : {:.4f} at epoch {:03d}".format(best_acc, best_epoch))
torch.save(model, 'models/'+dataset+'_model_'+str(epoch+1)+'.pt')
return model, history
num_epochs = 3 #训练周期数
trained_model, history = train_and_valid(resnet50, loss_func, optimizer, num_epochs)
torch.save(history, 'models/'+dataset+'_history.pt')
history = np.array(history)
plt.plot(history[:, 0:2])
plt.legend(['Tr Loss', 'Val Loss'])
plt.xlabel('Epoch Number')
plt.ylabel('Loss')
plt.ylim(0, 1)
plt.savefig(dataset+'_loss_curve.png')
plt.show()
plt.plot(history[:, 2:4])
plt.legend(['Tr Accuracy', 'Val Accuracy'])
plt.xlabel('Epoch Number')
plt.ylabel('Accuracy')
plt.ylim(0, 1)
plt.savefig(dataset+'_accuracy_curve.png')
plt.show()
2. Re-extract the saved file of the trained model when using it, and transfer the picture to be detected to the trained model and read the result.
This part is mainly to read the saved file of the model trained in the previous part (this step needs to build a model with the same structure as the model we used for training, and import the trained model weights into this new model ), and then input the picture to be detected to the model for detection, and interpret and read the output result of the model (the output result is a tensor vector).
The specific process code is as follows:
import torch
from torchvision import models, transforms
import torch.nn as nn
import cv2
classes = ["1","2"] #识别种类名称（顺序要与训练时的数据导入编号顺序对应，可以使用datasets.ImageFolder().class_to_idx来查看）
transf = transforms.ToTensor()
device = torch.device('cuda:0')
num_classes = 2
model_path = "models/data_model_3.pt"
image_input = cv2.imread("test_image1.jpg")
image_input = transf(image_input)
image_input = torch.unsqueeze(image_input,dim=0).cuda()
#搭建模型
resnet50 = models.resnet50(pretrained=True)
for param in resnet50.parameters():
param.requires_grad = False
fc_inputs = resnet50.fc.in_features
resnet50.fc = nn.Sequential(
nn.Linear(fc_inputs, 256),
nn.ReLU(),
nn.Dropout(0.4),
nn.Linear(256, num_classes),
nn.LogSoftmax(dim=1)
)
resnet50 = torch.load(model_path)
outputs = resnet50(image_input)
value,id =torch.max(outputs,1)
print(outputs,"\n","结果是：",classes[id])