600字范文 > 「深度学习一遍过」必修17：基于Pytorch细粒度分类实战

「深度学习一遍过」必修17：基于Pytorch细粒度分类实战

时间：2023-03-03 17:59:16

本专栏用于记录关于深度学习的笔记，不光方便自己复习与查阅，同时也希望能给您解决一些关于深度学习的相关问题，并提供一些微不足道的人工神经网络模型设计思路。
专栏地址：「深度学习一遍过」必修篇

1 实战内容简介

2 数据集读取

2.1 dataset

2.2 dataloader

3 模型搭建

3.1 基准模型

3.2 与基准模型相对应的双线性模型

4 性能差异比较

4.1 tensorboard查看测试集准确率差异

4.2 耗时比较（单位：秒）

1 实战内容简介

数据集：CUB-200，共200类不同种类的鸟。

第步：计算每一个通道对应的特征，，表示第行列激活值最大，表示图像个数。第步：聚类初始化，使用等聚类方法对特征进行聚类，得到个部件第步：得到初始的个部件，为每一个部件引入全连接层，预测维向量，对应每一个通道有多大概率属于该部件，初始标签为第步聚类结果第步：得到个维向量，对特征图进行加权求和，得到个部件的注意力第步：将虚拟部件对应区域特征进行池化得到特征向量，进行分类

2 数据集读取

2.1 dataset

import osimport random# 定义一个列表，用于txt文件内存放路径及标签数据data_list = []# 初始化类别标签class_label = -1# 加载dataset图片数据dataset_path = './data/CUB_200_/images/'# 遍历文件，依次将文件名存入上述定义列表当中for root,_,filenames in os.walk(dataset_path):for i in filenames:data = root+"/"+i+"\t"+str(class_label)+"\n"print(data)data_list.append(data) # 依次添加，不清空class_label += 1# 打乱txt文件中的数据，保证下面分类进行测试集与训练集每个标签都有涉及random.shuffle(data_list)# 定义训练文本数据列表train_list = []# 将打乱后的总数据列表中的80%的数据用于训练集for i in range(int(len(data_list) * 0.8)):train_list.append(data_list[i])# 创建并以“写”方式打开train.txtwith open('train.txt', 'w', encoding='UTF-8') as f:for train_img in train_list:f.write(str(train_img)) # 将训练数据集数据写入train.txt# 定义测试文本数据列表eval_list = []# 将打乱后的总数据列表中的20%的数据用于训练集for i in range(int(len(data_list) * 0.8),len(data_list)):eval_list.append(data_list[i])# 创建并以“写”方式打开eval.txtwith open('eval.txt', 'w', encoding='UTF-8') as f:for eval_img in eval_list:f.write(eval_img) # 将测试数据集数据写入eval.txt

2.2 dataloader

import torchfrom PIL import Imageimport torchvision.transforms as transformsfrom PIL import ImageFileImageFile.LOAD_TRUNCATED_IMAGES = Truefrom torch.utils.data import Dataset# 定义一个dataloader用于等会调用class Data_Loader(Dataset):def __init__(self, txt_path, train_flag=True):self.imgs_info = self.get_images(txt_path)self.train_flag = train_flagself.targetsize = 224# 把图片压缩成224X224# 训练集的处理方法self.train_tf = pose([transforms.Resize(self.targetsize), # 压缩图片transforms.RandomHorizontalFlip(),# 随机水平反转transforms.RandomVerticalFlip(), # 随机垂直反转图片transforms.ToTensor(), # 把图片转变为Tensor()格式，pytorch才能读写])# 验证集（测试集）的处理方法self.val_tf = pose([transforms.Resize(self.targetsize),transforms.ToTensor(),])# 通过读取txt文档内容，返回文档中的每一条信息def get_images(self, txt_path):with open(txt_path, 'r', encoding='utf-8') as f:imgs_info = f.readlines()imgs_info = list(map(lambda x:x.strip().split('\t'), imgs_info))return imgs_infodef padding_black(self, img):w, h = img.sizescale = 224. / max(w, h)img_fg = img.resize([int(x) for x in [w * scale, h * scale]])size_fg = img_fg.sizesize_bg = 224img_bg = Image.new("RGB", (size_bg, size_bg))img_bg.paste(img_fg, ((size_bg - size_fg[0]) // 2,(size_bg - size_fg[1]) // 2))img = img_bgreturn img# 我们在遍历数据集中返回的每一条数据def __getitem__(self, index):img_path, label = self.imgs_info[index]# 读取每一条数据，得到图片路径和标签值img = Image.open(img_path) # 利用 Pillow打开图片img = img.convert('RGB')# 将图片转变为RGB格式img = self.padding_black(img)if self.train_flag:# 对训练集和测试集分别处理img = self.train_tf(img)else:img = self.val_tf(img)label = int(label)return img, label # 返回图片和其标签值# 我们在遍历数据集时，遍历多少，返回的是数据集的长度def __len__(self):return len(self.imgs_info)if __name__ == "__main__":train_dataset = Data_Loader("eval.txt", True)print("数据个数：", len(train_dataset))train_loader = torch.utils.data.DataLoader(dataset=train_dataset,batch_size=10,shuffle=True)test_dataset = Data_Loader("eval.txt", False)print("数据个数：", len(test_dataset))test_loader = torch.utils.data.DataLoader(dataset=test_dataset,batch_size=50,shuffle=True)for image, label in test_loader:print(image.shape)print(label)

3 模型搭建

3.1 基准模型

import torch.optim as optimimport torchimport torch.nn as nnfrom torch.utils.data import DataLoaderfrom dataloader import Data_Loaderfrom torchvision.models import resnet50from tensorboardX import SummaryWriterimport time# 创建一个基准模型类class Baisc_Net(nn.Module):def __init__(self):super(Baisc_Net, self).__init__()self.features = nn.Sequential(resnet50().conv1,resnet50().bn1,resnet50().relu,resnet50().maxpool,resnet50().layer1,resnet50().layer2,resnet50().layer3,resnet50().layer4)self.classifiers = nn.Linear(100352, 5)def forward(self, x):x = self.features(x)x = x.view(-1, 100352)x = self.classifiers(x)return x# 定义一个函数，在训练集上进行，打印输出loss值与acc值，并写入tensorboard中def baisic_net_train(train_loader, model, criterion, optimizer, epoch):model.train()# 初始化正确率running_corrects = 0.0for i, (input, target) in enumerate(train_loader):input = input.cuda()target = target.cuda()output = model(input)# 将输入输入到模型中，产生一个输出loss = criterion(output, target) # 计算实际输出与目标输出之间的差距，将值传入loss变量中_, preds = torch.max(output.data, 1)# 按行输出该维度预测概率最大的那一个标签running_corrects += torch.sum(preds == target).item() # 计算该轮次中预测正确的标签数总和optimizer.zero_grad() # 梯度清零loss.backward()# 反向传播optimizer.step() # 迭代更新if i % 20 == 0:print("Training loss = ", loss.item()) # 每轮中的20次输出一次lossepoch_acc = running_corrects / dataset_sizesprint("Training Accuracy = ", epoch_acc)# 输出每轮的准确率writer.add_scalar('contrast figure basic net', epoch_acc, global_step=epoch)# 将准确率写入到tensorboard中if __name__ == "__main__":train_dir_list = 'train.txt'valid_dir_list = 'eval.txt'batch_size = 1# 看显存epochs = 50 # 总共训练多少回合# 加载数据集# 自创了一个data_loader，调用时需要从dataloader.py中进行读取# train_data是我们的训练集train_data = Data_Loader(train_dir_list, train_flag=True)valid_data = Data_Loader(valid_dir_list, train_flag=False)dataset_sizes = len(train_data) # 查看训练数据集数量print(dataset_sizes)# 用dataloader加载dataset# 数据集的读写方式，num_workers控制多线程读写数据集，pin_memory是内存上锁，batch_size是多少个数据并行读取，shuffle是每次读写重新打乱数据集train_loader = DataLoader(dataset=train_data, num_workers=0, pin_memory=True, batch_size=batch_size, shuffle=True)valid_loader = DataLoader(dataset=valid_data, num_workers=0, pin_memory=True, batch_size=batch_size)# 定义网络model = Baisc_Net()print(model)model = model.cuda()# 使用交叉熵损失函数criterion = nn.CrossEntropyLoss()# 利用SGD优化算法optimizer = optim.SGD(model.parameters(), lr=0.01)# 将tensorboard文件写入runs文件夹中writer = SummaryWriter('./runs')# 定义一个开始时间，用于查看整个模型训练耗时start_time = time.time()# 开始训练for epoch in range(epochs):print("********************* Epoch ", epoch, " ************************")baisic_net_train(train_loader, model, criterion, optimizer, epoch) # 调用前面定义的训练方法epoch = epoch + 1# 定义一个结束时间end_time = time.time()# 用开始时间-结束时间=总耗时time = end_time - start_timeprint(time)# 关闭tensorboard写入writer.close()

3.2 与基准模型相对应的双线性模型

import torch.optim as optimimport torchimport timeimport torch.nn as nnfrom torch.utils.data import DataLoaderfrom dataloader import Data_Loaderfrom torchvision.models import resnet50from tensorboardX import SummaryWriter# 定义一个双线性模型类class Bilinear_form_Net(nn.Module):def __init__(self):super(Bilinear_form_Net, self).__init__()self.features = nn.Sequential(resnet50().conv1,resnet50().bn1,resnet50().relu,resnet50().maxpool,resnet50().layer1,resnet50().layer2,resnet50().layer3,resnet50().layer4)self.classifiers = nn.Linear(2048 ** 2, 5)def forward(self, x):x = self.features(x)batch_size = x.size(0)x = x.view(batch_size, 2048, x.size(2) ** 2)x = (torch.bmm(x, torch.transpose(x, 1, 2)) / 28 ** 2).view(batch_size, -1)x = self.classifiers(x)return x# 定义一个双线性模型测试集训练方法def bilinear_form_net_train(train_loader, model, criterion, optimizer, epoch):#, writer):model.train()running_corrects = 0.0for i, (input, target) in enumerate(train_loader):input = input.cuda()target = target.cuda()output = model(input)loss = criterion(output, target)_, preds = torch.max(output.data, 1)running_corrects += torch.sum(preds == target).item()optimizer.zero_grad()loss.backward()optimizer.step()if i % 20 == 0:print("Training loss = ", loss.item())epoch_acc = running_corrects / dataset_sizesprint("Training Accuracy = ", epoch_acc)writer.add_scalar('contrast figure bilinear form net', epoch_acc, global_step=epoch)if __name__ == "__main__":train_dir_list = 'train.txt'valid_dir_list = 'eval.txt'batch_size = 1epochs = 50# 加载数据train_data = Data_Loader(train_dir_list, train_flag=True)valid_data = Data_Loader(valid_dir_list, train_flag=False)dataset_sizes = len(train_data)print(dataset_sizes)# 用dataloader读取datasettrain_loader = DataLoader(dataset=train_data, num_workers=0, pin_memory=True, batch_size=batch_size, shuffle=True)valid_loader = DataLoader(dataset=valid_data, num_workers=0, pin_memory=True, batch_size=batch_size)# 实例化双线性模型model = Bilinear_form_Net()print(model)model = model.cuda()# 损失函数采用交叉熵损失criterion = nn.CrossEntropyLoss()# 优化器采用SGD优化器optimizer = optim.SGD(model.parameters(), lr=0.01)# tensorboard文件写入runs文件夹下的文件中writer = SummaryWriter('./runs')start_time = time.time()# 开始训练，迭代epoch次for epoch in range(epochs):print("********************* Epoch ", epoch, " ************************")bilinear_form_net_train(train_loader, model, criterion, optimizer, epoch)# 调用训练方法epoch = epoch + 1end_time = time.time()time = end_time - start_time# 总耗时print(time)# 关闭tensorboard写入writer.close()