参考视频:36 数据增广【动手学深度学习v2】_哔哩哔哩_bilibili
一、数据增广
通过对数据变形(翻转、切割、颜色变换等)获取多样性的数据,增强泛化能力。
数据增广后训练的模型通常训练正确率低于测试正确率。
代码
import torchvision
from d2l import torch as d2l
d2l.set_figsize()
img = d2l.Image.open("/home/wujiqing/PythonProject/Helloworld/image/dog.jpg")
"""
d2l.plt.imshow(img);
d2l.plt.show()
"""
#定义辅助函数apply。 此函数在输入图像img上多次运行图像增广方法aug并显示所有结果
def apply(img, aug, num_rows=2, num_cols=4, scale=1.5):
Y = [aug(img) for _ in range(num_rows * num_cols)]
d2l.show_images(Y, num_rows, num_cols, scale=scale)
d2l.plt.show()
50%的几率使图像向左或向右翻转
apply(img, torchvision.transforms.RandomHorizontalFlip())
50%的几率向上或向下翻转
apply(img, torchvision.transforms.RandomVerticalFlip())
随机裁剪一个面积为原始面积10%到100%的区域,该区域的宽高比从0.5~2之间随机取值。 然后,区域的宽度和高度都被缩放到200像素
shape_aug = torchvision.transforms.RandomResizedCrop((200, 200), scale=(0.1, 1), ratio=(0.5, 2))
apply(img, shape_aug)
更改图像的亮度(brightness)、对比度(contrast)、饱和度(saturation)和色调(hue)
color_aug = torchvision.transforms.ColorJitter(
brightness=0.5, contrast=0.5, saturation=0.5, hue=0.5)
apply(img, color_aug)
结合多种图像增广方法
augs = torchvision.transforms.Compose([
torchvision.transforms.RandomHorizontalFlip(), color_aug, shape_aug])
apply(img, augs)
二、微调
利用预训练好的模型进行训练
当目标数据集比源数据集小得多时,微调有助于提高模型的泛化能力
特点:
- 更强正则化
- 更小学习率
- 更少迭代次数
- 重用分类器权重
- 低层、特征层参数通用
热狗识别代码
import os
import torch
import torchvision
from torch import nn
from d2l import torch as d2l
#获取数据集
d2l.DATA_HUB['hotdog'] = (d2l.DATA_URL + 'hotdog.zip',
'fba480ffa8aa7e0febbb511d181409f899b9baa5')
data_dir = d2l.download_extract('hotdog')
train_imgs = torchvision.datasets.ImageFolder(os.path.join(data_dir, 'train'))
test_imgs = torchvision.datasets.ImageFolder(os.path.join(data_dir, 'test'))
hotdogs = [train_imgs[i][0] for i in range(8)]
not_hotdogs = [train_imgs[-i - 1][0] for i in range(8)]
d2l.show_images(hotdogs + not_hotdogs, 2, 8, scale=1.4)
d2l.plt.show()
# 使用RGB通道的均值和标准差,以标准化每个通道
normalize = torchvision.transforms.Normalize(
[0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
train_augs = torchvision.transforms.Compose([
torchvision.transforms.RandomResizedCrop(224),
torchvision.transforms.RandomHorizontalFlip(),
torchvision.transforms.ToTensor(),
normalize])
test_augs = torchvision.transforms.Compose([
torchvision.transforms.Resize([256, 256]),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
normalize])
#构建一个新的神经网络作为目标模型, 它的定义方式与预训练源模型的定义方式相同,只是最终层中的输出数量被设置为目标数据集中的类数
finetune_net = torchvision.models.resnet18(pretrained=True)
finetune_net.fc = nn.Linear(finetune_net.fc.in_features, 2)
#成员变量output的参数是随机初始化的
nn.init.xavier_uniform_(finetune_net.fc.weight)
#微调模型
def train_fine_tuning(net, learning_rate, batch_size=128, num_epochs=5,param_group=True):
train_iter = torch.utils.data.DataLoader(torchvision.datasets.ImageFolder(
os.path.join(data_dir, 'train'), transform=train_augs),
batch_size=batch_size, shuffle=True)
test_iter = torch.utils.data.DataLoader(torchvision.datasets.ImageFolder(
os.path.join(data_dir, 'test'), transform=test_augs),
batch_size=batch_size)
devices = d2l.try_all_gpus()
loss = nn.CrossEntropyLoss(reduction="none")
if param_group:
#quchubushiquanlianjiedecen
params_1x = [param for name, param in net.named_parameters()
if name not in ["fc.weight", "fc.bias"]]
trainer = torch.optim.SGD([{'params': params_1x},
{'params': net.fc.parameters(),
'lr': learning_rate * 10}],
lr=learning_rate, weight_decay=0.001)
else:
trainer = torch.optim.SGD(net.parameters(), lr=learning_rate,
weight_decay=0.001)
d2l.train_ch13(net, train_iter, test_iter, loss, trainer, num_epochs,
devices)
train_fine_tuning(finetune_net, 5e-5)
结果
loss 0.247, train acc 0.910, test acc 0.926
363.4 examples/sec on [device(type='cuda', index=0)]