深度学习系列3——Pytorch 图像分类(VGG)

最新推荐文章于 2024-01-28 19:38:16 发布

霁风AI

最新推荐文章于 2024-01-28 19:38:16 发布

阅读量3.3k

点赞数 3

分类专栏： DL and ML 我用Python Machine Learning 文章标签：深度学习 pytorch 分类

本文链接：https://blog.csdn.net/wwt18811707971/article/details/127990561

版权

DL and ML 同时被 3 个专栏收录

11 篇文章 4 订阅

订阅专栏

我用Python

11 篇文章 0 订阅

订阅专栏

Machine Learning

7 篇文章 0 订阅

订阅专栏

1. 概述

本文主要是参照 B 站 UP 主霹雳吧啦Wz 的视频学习笔记，参考的相关资料在文末参照栏给出，包括实现代码和文中用的一些图片。

整个工程已经上传个人的 github https://github.com/lovewinds13/QYQXDeepLearning ，下载即可直接测试，数据集文件因为比较大，已经删除了，按照下文教程下载即可。

论文下载：Very Deep Convolutional Networks For Large-Scale Image Recognition

2. VGG

VGG 网络是 2014 年由牛津大学著名研究组 VGG (Visual Geometry Group) 提出，斩获该年 ImageNet 竞赛中 Localization Task (定位任务) 第一名和 Classification Task (分类任务) 第二名。

VGG 主要是卷积神经网络的深度大规模图像识别精度的影响，使用 3 x 3 的卷积核构建各种深度的卷积神经网络结构，减少了所需的参数。常用的深度是 VGG16 和 VGG 19。

2.1 网络框架

结构框图

VGG 不同深度的卷积层数配置如下：

VGG不同深度

VGG 的特点：

(1) 输入为 224 x 224 x 3 的图像；
(2) 卷积层步长 (stride = 1) ，补位 (padding = 1) ；
(3) 最大池化 (maxpool) 尺寸 (size = 2)，步长 (stride = 2)，5 个最大池化层；
(4) 连续的卷积层后接池化层，降低图像的分辨率；
(5) 相比于ALexNet，卷积深度更深；
(6) VGG 的结构一致，卷积核尺寸 3 x 3，池化核尺寸 2 x 2，不断叠加；
(7) VGG 参数更多，计算耗费计算资源；

2.2 感受野

定义： 卷积网络中，决定某一层输出结果中一个元素所对应输入层区域的大小，感受野用来表示网络内部的不同神经元对原图像感受范围的大小，即每一层输出的特征图 (feature map) 上的像素点在原始图像上映射的区域大小。

感受野计算
可以看待，通过两次 3x3 的卷积与一次 5x5 的卷积，都最终输出一个特征图。

针对两次 3x3 的卷积，经过第一次卷积之后，输出 3x3 的特征图，每个像素点对应的上一层 3*3 的区域，即感受野大小为 3，再经过第二次 3x3 的卷积，输出为 1x1 的特征图，这个像素点对应原始图像 5x5 的区域，即感受野为 5。

对 5x5 的图像，经过一次 5x5 的卷积核，同样输出 1x1 特征图，它的感受野也是 5。

感受野计算：

感受野计算
堆叠两个3x3的卷积核替代5x5的卷积核，假设输入、输出通道为 C。

卷积核尺寸	参数
5 * 5	5 * 5 * C * C = 25C^2
3 * 3	3 * 3 * C * C + 3 * 3 * C * C = 18C^2

在输出相同的特征图下，小卷积核的参数明显小于大卷积核。

3. demo 实现

3.1 数据集

本文使用花分类数据集，下载链接: 花分类数据集——http://download.tensorflow.org/example_images/flower_photos.tgz

数据集划分参考这个pytorch图像分类篇：3.搭建AlexNet并训练花分类数据集

3.1 demo 结构：

代码结构
在 CPU 训练的基础上，为了修改为 GPU 训练，因此单独修改了一个文件 train_gpu.py。

3.2 model.py


"""
VGG模型
"""


import torch.nn as nn
import torch


# official pretrain weights
model_urls = {
    'vgg11': 'https://download.pytorch.org/models/vgg11-bbd30ac9.pth',
    'vgg13': 'https://download.pytorch.org/models/vgg13-c768596a.pth',
    'vgg16': 'https://download.pytorch.org/models/vgg16-397923af.pth',
    'vgg19': 'https://download.pytorch.org/models/vgg19-dcbb9e9d.pth'
}

cfgs = {
    'vgg11': [64, 'M', 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'],
    'vgg13': [64, 64, 'M', 128, 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'],
    'vgg16': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 'M', 512, 512, 512, 'M', 512, 512, 512, 'M'],
    'vgg19': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 256, 'M', 512, 512, 512, 512, 'M', 512, 512, 512, 512, 'M'],
}


class VGG(nn.Module):
    def __init__(self, features, num_classes=1000, init_weights=False):
        super(VGG, self).__init__()
        self.features = features
        self.classifier = nn.Sequential(
            nn.Linear(512*7*7, 2048),   # 第1线性层, 2048 减少参数
            nn.ReLU(True),
            nn.Dropout(p=0.5),
            nn.Linear(2048, 2048),  # 第2线性层
            nn.ReLU(True),
            nn.Dropout(p=0.5),
            nn.Linear(2048, num_classes),   # 第3线性层
        )
        if init_weights:
            self._initialize_weights()

    def forward(self, x):
        x = self.features(x)    # N x 3 x 224 x 224
        x = torch.flatten(x, start_dim=1)   # N x 512 x 7 x 7
        x = self.classifier(x)   # N x 512*7*7

        return x

    def _initialize_weights(self):
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.xavier_uniform_(m.weight)
                if m.bias is not None:
                    nn.init.constant_(m.bias, 0)
            elif isinstance(m, nn.Linear):
                nn.init.xavier_uniform_(m.weight)
                nn.init.constant_(m.bias, 0)

def make_features(cfg: list):
    layers = []
    in_channels = 3
    for v in cfg:
        if v == "M":
            layers += [nn.MaxPool2d(kernel_size=2, stride=2)]
        else:
            conv2d = nn.Conv2d(in_channels, v, kernel_size=3, padding=1)
            layers += [conv2d, nn.ReLU(True)]
            in_channels = v

    return nn.Sequential(*layers)

def vgg(model_name="vgg16", **kwargs):
    assert model_name in cfgs, "Warning: model number {} not in cfgs dist!".format(model_name)
    cfg = cfgs[model_name]
    model = VGG(make_features(cfg), **kwargs)

    return model

"""
测试模型
"""
# if __name__ == '__main__':
#     input1 = torch.rand([224, 3, 224, 224])
#     model_name = "vgg16"
#     model_x = vgg(model_name=model_name, num_classes=5, init_weights=True)
#     print(model_x)
    # output = AlexNet(input1)

此处的网络模型仅使用了一半的参数，即线性层的 2048，减少参数。

3.2.1 Tensor 展平

直接使用 torch.flatten 方法展平张量


 x = torch.flatten(x, start_dim=1)	# 二维平坦

3.3 train.py

3.3.1 导入包


"""
训练(CPU)
"""
import os
import sys
import json
import time
import torch
import torch.nn as nn
from torchvision import transforms, datasets, utils
import matplotlib.pyplot as plt
import numpy as np
import torch.optim as optim
from tqdm import tqdm   # 显示进度条模块

from model import vgg

3.3.2 数据集预处理


    data_transform = {
        "train": transforms.Compose([
                                    transforms.RandomResizedCrop(224),  # 随机裁剪, 再缩放为 224*224
                                    transforms.RandomHorizontalFlip(),  # 水平随机翻转
                                    transforms.ToTensor(),
                                    transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
        ]),
        "val": transforms.Compose([
                                    transforms.Resize((224, 224)),  # 元组(224, 224)
                                    transforms.ToTensor(),
                                    transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
        ])
    }

3.3.3 加载数据集

3.3.3.1 读取数据路径


# data_root = os.path.abspath(os.path.join(os.getcwd(), "../..")) # 读取数据路径
data_root = os.path.abspath(os.path.join(os.getcwd(), "./"))
image_path = os.path.join(data_root, "data_set", "flower_data")
# image_path = data_root + "/data_set/flower_data/"
assert os.path.exists(image_path), "{} path does not exist.".format(image_path)

此处相比于 UP 主教程，修改了读取路径。

3.3.3.2 加载训练集


 train_dataset = datasets.ImageFolder(root=os.path.join(image_path, "train"),
                                         transform=data_transform["train"]
                                         )
 train_loader = torch.utils.data.DataLoader(train_dataset,
                                               batch_size=batch_size,
                                               shuffle=True,
                                               num_workers=nw
                                               )

3.3.3.3 加载验证集


val_dataset = datasets.ImageFolder(root=os.path.join(image_path, "val"),
                                       transform=data_transform["val"]
                                       )
val_num = len(val_dataset)
val_loader = torch.utils.data.DataLoader(val_dataset,
                                             batch_size=4,
                                             shuffle=False,
                                             num_workers=nw
                                             )

3.3.3.4 保存数据索引


flower_list = train_dataset.class_to_idx
    cla_dict = dict((val, key) for key, val in flower_list.items())
    json_str = json.dumps(cla_dict, indent=4)
    with open("calss_indices.json", 'w') as json_file:
        json_file.write(json_str)

3.3.4 训练过程


	model_name = "vgg16"
    net = vgg(model_name=model_name, num_classes=5, init_weights=True)     # 实例化网络(5分类)
    # net.to(device)
    net.to("cpu")   # 直接指定 cpu
    loss_function = nn.CrossEntropyLoss()   # 交叉熵损失
    optimizer = optim.Adam(net.parameters(), lr=0.0001)     # 优化器(训练参数, 学习率)

    epochs = 10     # 训练轮数
    save_path = "./VGGNet.pth"
    best_accuracy = 0.0
    train_steps = len(train_loader)
    for epoch in range(epochs):
        net.train()     # 开启Dropout
        running_loss = 0.0
        train_bar = tqdm(train_loader, file=sys.stdout)     # 设置进度条图标
        for step, data in enumerate(train_bar):     # 遍历训练集,
            images, labels = data   # 获取训练集图像和标签
            optimizer.zero_grad()   # 清除历史梯度
            outputs = net(images)   # 正向传播
            loss = loss_function(outputs, labels)   # 计算损失值
            loss.backward()     # 方向传播
            optimizer.step()    # 更新优化器参数
            running_loss += loss.item()
            train_bar.desc = "train epoch [{}/{}] loss:{:.3f}".format(epoch + 1,
                                                                      epochs,
                                                                      loss
                                                                      )
        # 验证
        net.eval()      # 关闭Dropout
        acc = 0.0
        with torch.no_grad():
            val_bar = tqdm(val_loader, file=sys.stdout)
            for val_data in val_bar:
                val_images, val_labels = val_data
                outputs = net(val_images)
                predict_y = torch.max(outputs, dim=1)[1]
                acc += torch.eq(predict_y, val_labels).sum().item()
        val_accuracy = acc / val_num
        print("[epoch %d ] train_loss: %3f    val_accurancy: %3f" %
              (epoch + 1, running_loss / train_steps, val_accuracy))
        if val_accuracy > best_accuracy:    # 保存准确率最高的
            best_accuracy = val_accuracy
            torch.save(net.state_dict(), save_path)
    print("Finshed Training.")

训练过程可视化信息输出：

训练过程

GPU 训练代码： 仅在 CPU 训练的基础上做了数据转换处理。


"""
训练(GPU)
"""

import os
import sys
import json
import time
import torch
import torch.nn as nn
from torchvision import transforms, datasets, utils
import matplotlib.pyplot as plt
import numpy as np
import torch.optim as optim
from tqdm import tqdm

from model import vgg


def main():
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    print(f"use device is {device}")

    data_transform = {
        "train": transforms.Compose([
                                    transforms.RandomResizedCrop(224),
                                    transforms.RandomHorizontalFlip(),
                                    transforms.ToTensor(),
                                    transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
        ]),
        "val": transforms.Compose([
                                    transforms.Resize((224, 224)),
                                    transforms.ToTensor(),
                                    transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
        ])
    }
    # data_root = os.path.abspath(os.path.join(os.getcwd(), "../..")) # 读取数据路径
    data_root = os.path.abspath(os.path.join(os.getcwd(), "./"))
    image_path = os.path.join(data_root, "data_set", "flower_data")
    # image_path = data_root + "/data_set/flower_data/"
    assert os.path.exists(image_path), "{} path does not exist.".format(image_path)
    train_dataset = datasets.ImageFolder(root=os.path.join(image_path, "train"),
                                         transform=data_transform["train"]
                                         )
    train_num = len(train_dataset)
    flower_list = train_dataset.class_to_idx
    cla_dict = dict((val, key) for key, val in flower_list.items())
    json_str = json.dumps(cla_dict, indent=4)
    with open("calss_indices.json", 'w') as json_file:
        json_file.write(json_str)

    batch_size = 32
    nw = min([os.cpu_count(), batch_size if batch_size > 1 else 0, 8])  # 线程数计算
    nw = 0
    print(f"Using {nw} dataloader workers every process.")

    train_loader = torch.utils.data.DataLoader(train_dataset,
                                               batch_size=batch_size,
                                               shuffle=True,
                                               num_workers=nw
                                               )
    val_dataset = datasets.ImageFolder(root=os.path.join(image_path, "val"),
                                       transform=data_transform["val"]
                                       )
    val_num = len(val_dataset)
    val_loader = torch.utils.data.DataLoader(val_dataset,
                                             batch_size=4,
                                             shuffle=False,
                                             num_workers=nw
                                             )
    print(f"Using {train_num} images for training, {val_num} images for validation.")

    # test_data_iter = iter(val_loader)
    # test_image, test_label = next(test_data_iter)

    """ 测试数据集图片"""
    # def imshow(img):
    #     img = img / 2 + 0.5
    #     np_img = img.numpy()
    #     plt.imshow(np.transpose(np_img, (1, 2, 0)))
    #     plt.show()
    # print(' '.join('%5s' % cla_dict[test_label[j].item()] for j in range(4)))
    # imshow(utils.make_grid(test_image))

   	model_name = "vgg16"
    net = vgg(model_name=model_name, num_classes=5, init_weights=True)  # 实例化网络(5分类)
    net.to(device)
    loss_function = nn.CrossEntropyLoss()
    optimizer = optim.Adam(net.parameters(), lr=0.0001)

    epochs = 10
    save_path = "./VGGNet.pth"
    best_accuracy = 0.0
    train_steps = len(train_loader)
    for epoch in range(epochs):
        net.train()
        running_loss = 0.0
        train_bar = tqdm(train_loader, file=sys.stdout)
        for step, data in enumerate(train_bar):
            images, labels = data
            optimizer.zero_grad()
            outputs = net(images.to(device))
            loss = loss_function(outputs, labels.to(device))
            loss.backward()
            optimizer.step()
            running_loss += loss.item()
            train_bar.desc = "train epoch [{}/{}] loss:{:.3f}".format(epoch + 1,
                                                                      epochs,
                                                                      loss
                                                                      )
        # 验证
        net.eval()
        acc = 0.0
        with torch.no_grad():
            val_bar = tqdm(val_loader, file=sys.stdout)
            for val_data in val_bar:
                val_images, val_labels = val_data
                outputs = net(val_images.to(device))
                predict_y = torch.max(outputs, dim=1)[1]
                acc += torch.eq(predict_y, val_labels.to(device)).sum().item()
        val_accuracy = acc / val_num
        print("[epoch %d ] train_loss: %3f    val_accurancy: %3f" %
              (epoch + 1, running_loss / train_steps, val_accuracy))
        if val_accuracy > best_accuracy:
            best_accuracy = val_accuracy
            torch.save(net.state_dict(), save_path)
    print("Finshed Training.")

VGG 通过 GPU 训练对显存要求比较高。

3.3.5 结果预测


"""
预测
"""

import os
import json
import torch
from PIL import Image
from torchvision import transforms
import matplotlib.pyplot as plt
from model import vgg


def main():
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

    data_transform = transforms.Compose([
        transforms.Resize((224, 224)),
        transforms.ToTensor(),
        transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
    ])

      image_path = "./daisy01.jpg"
    img = Image.open(image_path)
    plt.imshow(img)
    img = data_transform(img)   # [N, C H, W]
    img = torch.unsqueeze(img, dim=0)   # 维度扩展
    # print(f"img={img}")
    json_path = "./calss_indices.json"
    with open(json_path, 'r') as f:
        class_indict = json.load(f)

    # model = AlexNet(num_classes=5).to(device)   # GPU
    model = vgg(model_name="vgg16", num_classes=5)  # CPU
    weights_path = "./VGGNet.pth"
    model.load_state_dict(torch.load(weights_path))
    model.eval()    # 关闭 Dorpout
    with torch.no_grad():
        # output = torch.squeeze(model(img.to(device))).cpu()   #GPU
        output = torch.squeeze(model(img))      # 维度压缩
        predict = torch.softmax(output, dim=0)
        predict_cla = torch.argmax(predict).numpy()
        print_res = "class: {}  prob: {:.3}".format(class_indict[str(predict_cla)],
                                                    predict[predict_cla].numpy())
        plt.title(print_res)
        # for i in range(len(predict)):
        #     print("class: {}  prob: {:.3}".format(class_indict[str(predict_cla)],
        #                                             predict[predict_cla].numpy()))
        plt.show()