论文中不平衡因子为100,基准网络为 ResNet-32 的的情况下,使用sigmoid损失函数时,分类错误率为29.55%;使用类平衡损失函数时,分类错误率为:25.43%.准确率提升约 4 个百分点。
本博客中使用的基准网络为 Res-Net18 实验结果,使用sigmoid损失函数时,分类错误率为26.64%;使用类平衡损失函数时,分类错误率为:25.85%.准确率提升约 1 个百分点。表明类平衡损失函数对长尾数据集分类准确度有所提升,但是由于基准网络不同,效果没有论文中那么明显。
以下为训练代码:
第三步,训练
cifartrain.py
import torch
import torch.nn as nn
import torch.optim as optim
import torchvision
import torchvision.transforms as transforms
import argparse
from resnet18 import ResNet18
import os
from class_balanced_loss import CB_loss
import numpy as np
import torch.nn.functional as F
from loadcifar import Cifar10_Dataset
from torch.optim import lr_scheduler
from sigmoidCE import sigmoidlose
# 定义是否使用GPU
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# 参数设置,使得我们能够手动输入命令行参数,就是让风格变得和Linux命令行差不多
parser = argparse.ArgumentParser(description='PyTorch CIFAR10 Training')
parser.add_argument('--outf', default='./model/', help='folder to output images and model checkpoints') #输出结果保存路径
args = parser.parse_args()
# 超参数设置
EPOCH = 200 #遍历数据集次数
pre_epoch = 0 # 定义已经遍历数据集的次数
BATCH_SIZE = 128 #批处理尺寸(batch_size)
LR = 0.1 #学习率
# 准备数据集并预处理
transform_train = transforms.Compose([
transforms.RandomCrop(32, padding=4), #先四周填充0,在吧图像随机裁剪成32*32
transforms.RandomHorizontalFlip(), #图像一半的概率翻转,一半的概率不翻转
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)), #R,G,B每层的归一化用到的均值和方差
])
transform_test = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
])
def target_transform(label):
label = np.array(label)
target = torch.from_numpy(label).long()
return target
# trainset = torchvision.datasets.CIFAR10(root='./data', train=True, download=True, transform=transform_train) #训练数据集
# trainloader = torch.utils.data.DataLoader(trainset, batch_size=BATCH_SIZE, shuffle=True, num_workers=2) #生成一个个batch进行批训练,组成batch的时候顺序打乱取
trainset = Cifar10_Dataset(True, transform=transform_train, target_transform=target_transform )#训练数据集
trainloader = torch.utils.data.DataLoader(trainset, batch_size=BATCH_SIZE, shuffle=True, num_workers=2) #生成一个个batch进行批训练,组成batch的时候顺序打乱取
print('size of train_data:{}'.format(trainset.__len__()))
testset = Cifar10_Dataset(False, transform=transform_test, target_transform=target_transform)
testloader = torch.utils.data.DataLoader(testset, batch_size=100, shuffle=False, num_workers=2)
print('size of testset_data:{}'.format(testset.__len__()))
# Cifar-10的标签
classes = ('plane', 'car', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck')
# 模型定义-ResNet
net = ResNet18().to(device)
# 定义损失函数和优化方式
optimizer = optim.SGD(net.parameters(), lr=LR, momentum=0.9, weight_decay=5e-4) #优化方式为mini-batch momentum-SGD,并采用L2正则化(权重衰减)
scheduler = lr_scheduler.MultiStepLR(optimizer, [160, 180], 0.1)
# 训练
if __name__ == "__main__":
num = []
for i in range(10):
num.append(int(np.floor(5000 * ((1 / 100) ** (1 / 9)) ** (i))))
num = np.array(num)
# num = torch.from_numpy(num)
# num = num.to(device)
if not os.path.exists(args.outf):
os.makedirs(args.outf)
best_acc = 0 #2 初始化best test accuracy
print("Start Training, Resnet-18!") # 定义遍历数据集的次数
with open("acc.txt", "w") as f:
with open("log.txt", "w")as f2:
for epoch in range(pre_epoch, EPOCH):
scheduler.step()
print('\nEpoch: %d' % (epoch + 1))
net.train()
sum_loss = 0.0
correct = 0.0
total = 0.0
for i, data in enumerate(trainloader, 0):
# 准备数据
length = len(trainloader)
inputs, labels = data
inputs, labels = inputs.to(device), labels.to(device)
optimizer.zero_grad()
# forward + backward
outputs = net(inputs)
loss = CB_loss(labels = labels, logits = outputs,
samples_per_cls = num, no_of_classes = 10,
loss_type = "sigmoid", beta = 0.9999, gamma=2)
# loss = sigmoidlose(labels, outputs)
loss.backward()
optimizer.step()
# 每训练1个batch打印一次loss和准确率
sum_loss += loss.item()
_, predicted = torch.max(outputs.data, 1)
total += labels.size(0)
correct += predicted.eq(labels.data).cpu().sum()
print('[epoch:%d, iter:%d] Loss: %.03f | Acc: %.3f%% | Lr: %.03f'
% (epoch + 1, (i + 1 + epoch * length), sum_loss / (i + 1),
100. * float(correct) / total, optimizer.state_dict()['param_groups'][0]['lr']))
f2.write('%03d %05d |Loss: %.03f | Acc: %.3f%% | Lr: %.03f'
% (epoch + 1, (i + 1 + epoch * length), sum_loss / (i + 1),
100. * float(correct) / total, optimizer.state_dict()['param_groups'][0]['lr']))
f2.write('\n')
f2.flush()
# 每训练完一个epoch测试一下准确率
print("Waiting Test!")
with torch.no_grad():
correct = 0
total = 0
for data in testloader:
net.eval()
images, labels = data
images, labels = images.to(device), labels.to(device)
outputs = net(images)
# 取得分最高的那个类 (outputs.data的索引号)
_, predicted = torch.max(outputs.data, 1)
total += labels.size(0)
correct += (predicted == labels).sum()
print('测试分类准确率为:%.3f%%' % (100 * float(correct) / total))
acc = 100. * float(correct) / total
# 将每次测试结果实时写入acc.txt文件中
print('Saving model......')
torch.save(net.state_dict(), '%s/net_%03d.pth' % (args.outf, epoch + 1))
f.write("EPOCH=%03d,Accuracy= %.3f%%| Lr: %.03f" % (epoch + 1, acc,
optimizer.state_dict()['param_groups'][0]['lr']))
f.write('\n')
f.flush()
# 记录最佳测试分类准确率并写入best_acc.txt文件中
if acc > best_acc:
f3 = open("best_acc.txt", "w")
f3.write("EPOCH=%d,best_acc= %.3f%%" % (epoch + 1, acc))
f3.close()
best_acc = acc
print("Training Finished, TotalEPOCH=%d" % EPOCH)