.. _sec_image_augmentation:
图像增广
========
:numref:`sec_alexnet`\ 提到过大型数据集是成功应用深度神经网络的先决条件。
图像增广在对训练图像进行一系列的随机变化之后，生成相似但不同的训练样本，从而扩大了训练集的规模。
此外，应用图像增广的原因是，随机改变训练样本可以减少模型对某些属性的依赖，从而提高模型的泛化能力。
例如，我们可以以不同的方式裁剪图像，使感兴趣的对象出现在不同的位置，减少模型对于对象出现位置的依赖。
我们还可以调整亮度、颜色等因素来降低模型对颜色的敏感度。
可以说，图像增广技术对于AlexNet的成功是必不可少的。本节将讨论这项广泛应用于计算机视觉的技术。
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
%matplotlib inline
import mindcv
import mindspore
from mindspore import nn
from d2l import mindspore as d2l
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
%matplotlib inline
import torch
import torchvision
from torch import nn
from d2l import torch as d2l
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
d2l.set_figsize()
img = d2l.Image.open('../img/cat1.jpg')
d2l.plt.imshow(img);
.. figure:: output_image-augmentation_7d0887_12_0.svg
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
d2l.set_figsize()
img = d2l.Image.open('../img/cat1.jpg')
d2l.plt.imshow(img);
.. figure:: output_image-augmentation_7d0887_15_0.svg
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
def apply(img, aug, num_rows=2, num_cols=4, scale=1.5):
if isinstance(aug(img), tuple):
# add [0] to tuple of the return of mindspore.Compose()
Y = [aug(img)[0] for _ in range(num_rows * num_cols)]
else:
Y = [aug(img) for _ in range(num_rows * num_cols)]
d2l.show_images(Y, num_rows, num_cols, scale=scale)
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
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)
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
apply(img, mindspore.dataset.vision.RandomHorizontalFlip())
.. figure:: output_image-augmentation_7d0887_30_0.svg
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
apply(img, torchvision.transforms.RandomHorizontalFlip())
.. figure:: output_image-augmentation_7d0887_33_0.svg
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
apply(img, mindspore.dataset.vision.RandomVerticalFlip())
.. figure:: output_image-augmentation_7d0887_39_0.svg
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
apply(img, torchvision.transforms.RandomVerticalFlip())
.. figure:: output_image-augmentation_7d0887_42_0.svg
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
shape_aug = mindspore.dataset.vision.RandomResizedCrop(
(200, 200), scale=(0.1, 1), ratio=(0.5, 2))
apply(img, shape_aug)
.. figure:: output_image-augmentation_7d0887_48_0.svg
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
shape_aug = torchvision.transforms.RandomResizedCrop(
(200, 200), scale=(0.1, 1), ratio=(0.5, 2))
apply(img, shape_aug)
.. figure:: output_image-augmentation_7d0887_51_0.svg
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
apply(img, mindspore.dataset.vision.RandomColorAdjust(
brightness=0.5, contrast=0, saturation=0, hue=0))
.. figure:: output_image-augmentation_7d0887_57_0.svg
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
apply(img, torchvision.transforms.ColorJitter(
brightness=0.5, contrast=0, saturation=0, hue=0))
.. figure:: output_image-augmentation_7d0887_60_0.svg
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
apply(img, mindspore.dataset.vision.RandomColorAdjust(
brightness=0, contrast=0, saturation=0, hue=0.5))
.. figure:: output_image-augmentation_7d0887_66_0.svg
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
apply(img, torchvision.transforms.ColorJitter(
brightness=0, contrast=0, saturation=0, hue=0.5))
.. figure:: output_image-augmentation_7d0887_69_0.svg
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
color_aug = mindspore.dataset.vision.RandomColorAdjust(
brightness=0.5, contrast=0.5, saturation=0.5, hue=0.5)
apply(img, color_aug)
.. figure:: output_image-augmentation_7d0887_75_0.svg
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
color_aug = torchvision.transforms.ColorJitter(
brightness=0.5, contrast=0.5, saturation=0.5, hue=0.5)
apply(img, color_aug)
.. figure:: output_image-augmentation_7d0887_78_0.svg
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
augs = mindspore.dataset.transforms.Compose([
mindspore.dataset.vision.RandomHorizontalFlip(), color_aug, shape_aug])
apply(img, augs)
.. figure:: output_image-augmentation_7d0887_84_0.svg
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
augs = torchvision.transforms.Compose([
torchvision.transforms.RandomHorizontalFlip(), color_aug, shape_aug])
apply(img, augs)
.. figure:: output_image-augmentation_7d0887_87_0.svg
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
all_images = mindcv.create_dataset('cifar10',root="../data", split='train',
download=True).create_tuple_iterator()
d2l.show_images([next(all_images)[0] for i in range(32)], 4, 8, scale=0.8);
.. figure:: output_image-augmentation_7d0887_93_0.svg
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
all_images = torchvision.datasets.CIFAR10(train=True, root="../data",
download=True)
d2l.show_images([all_images[i][0] for i in range(32)], 4, 8, scale=0.8);
.. figure:: output_image-augmentation_7d0887_96_0.svg
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
train_augs = mindspore.dataset.transforms.Compose([
mindspore.dataset.vision.RandomHorizontalFlip(),
mindspore.dataset.vision.ToTensor()])
test_augs = mindspore.dataset.transforms.Compose([
mindspore.dataset.vision.ToTensor()])
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
train_augs = torchvision.transforms.Compose([
torchvision.transforms.RandomHorizontalFlip(),
torchvision.transforms.ToTensor()])
test_augs = torchvision.transforms.Compose([
torchvision.transforms.ToTensor()])
接下来，我们定义一个辅助函数，以便于读取图像和应用图像增广。PyTorch数据集提供的\ ``transform``\ 参数应用图像增广来转化图像。有关\ ``DataLoader``\ 的详细介绍，请参阅
:numref:`sec_fashion_mnist`\ 。
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
def load_cifar10(is_train, augs, batch_size):
usage = 'train' if is_train else 'test'
target_trans = mindspore.dataset.transforms.TypeCast(mindspore.dtype.int32)
dataset = mindcv.create_dataset('cifar10',root="../data", split=usage, shuffle=is_train, download=True)
dataset = dataset.map(augs, input_columns='image', num_parallel_workers=d2l.get_dataloader_workers())
dataset = dataset.map(target_trans, input_columns='label', num_parallel_workers=d2l.get_dataloader_workers())
dataset = dataset.batch(batch_size)
return dataset
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
def load_cifar10(is_train, augs, batch_size):
dataset = torchvision.datasets.CIFAR10(root="../data", train=is_train,
transform=augs, download=True)
dataloader = torch.utils.data.DataLoader(dataset, batch_size=batch_size,
shuffle=is_train, num_workers=d2l.get_dataloader_workers())
return dataloader
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
#@save
def train_batch_ch13(train_step, X, y):
"""用GPU进行小批量训练"""
l, pred = train_step(X, y)
train_loss_sum = l.sum()
train_acc_sum = d2l.accuracy(pred, y)
return train_loss_sum, train_acc_sum
#@save
def train_ch13(net, train_iter, test_iter, loss, trainer, num_epochs):
"""用GPU进行模型训练"""
timer, num_batches = d2l.Timer(), train_iter.get_dataset_size()
animator = d2l.Animator(xlabel='epoch', xlim=[1, num_epochs], ylim=[0, 1],
legend=['train loss', 'train acc', 'test acc'])
def forward_fn(inputs, targets):
logits = net(inputs)
l = loss(logits, targets)
return l, logits
grad_fn = mindspore.value_and_grad(forward_fn, None, trainer.parameters, has_aux=True)
def train_step(inputs, targets):
(l, logits), grads = grad_fn(inputs, targets)
trainer(grads)
return l, logits
for epoch in range(num_epochs):
# 4个维度：储存训练损失，训练准确度，实例数，特点数
metric = d2l.Accumulator(4)
net.set_train()
for i, (features, labels) in enumerate(train_iter):
timer.start()
l, acc = train_batch_ch13(
train_step, features, labels) # , loss, trainer, devices
metric.add(l, acc, labels.shape[0], labels.numel())
timer.stop()
if (i + 1) % (num_batches // 5) == 0 or i == num_batches - 1:
animator.add(epoch + (i + 1) / num_batches,
(metric[0] / metric[2], metric[1] / metric[3],
None))
test_acc = d2l.evaluate_accuracy_gpu(net, test_iter)
animator.add(epoch + 1, (None, None, test_acc))
print(f'loss {metric[0] / metric[2]:.3f}, train acc '
f'{metric[1] / metric[3]:.3f}, test acc {test_acc:.3f}')
print(f'{metric[2] * num_epochs / timer.sum():.1f} examples/sec')
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
#@save
def train_batch_ch13(net, X, y, loss, trainer, devices):
"""用GPU进行小批量训练"""
if isinstance(X, list):
# 微调BERT中所需
X = [x.to(devices[0]) for x in X]
else:
X = X.to(devices[0])
y = y.to(devices[0])
net.train()
trainer.zero_grad()
pred = net(X)
l = loss(pred, y)
l.sum().backward()
trainer.step()
train_loss_sum = l.sum()
train_acc_sum = d2l.accuracy(pred, y)
return train_loss_sum, train_acc_sum
#@save
def train_ch13(net, train_iter, test_iter, loss, trainer, num_epochs,
devices=d2l.try_all_gpus()):
"""用GPU进行模型训练"""
timer, num_batches = d2l.Timer(), len(train_iter)
animator = d2l.Animator(xlabel='epoch', xlim=[1, num_epochs], ylim=[0, 1],
legend=['train loss', 'train acc', 'test acc'])
net = nn.DataParallel(net, device_ids=devices).to(devices[0])
for epoch in range(num_epochs):
# 4个维度：储存训练损失，训练准确度，实例数，特点数
metric = d2l.Accumulator(4)
for i, (features, labels) in enumerate(train_iter):
timer.start()
l, acc = train_batch_ch13(
net, features, labels, loss, trainer, devices)
metric.add(l, acc, labels.shape[0], labels.numel())
timer.stop()
if (i + 1) % (num_batches // 5) == 0 or i == num_batches - 1:
animator.add(epoch + (i + 1) / num_batches,
(metric[0] / metric[2], metric[1] / metric[3],
None))
test_acc = d2l.evaluate_accuracy_gpu(net, test_iter)
animator.add(epoch + 1, (None, None, test_acc))
print(f'loss {metric[0] / metric[2]:.3f}, train acc '
f'{metric[1] / metric[3]:.3f}, test acc {test_acc:.3f}')
print(f'{metric[2] * num_epochs / timer.sum():.1f} examples/sec on '
f'{str(devices)}')
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
batch_size, net = 256, d2l.resnet18(10, 3)
def init_weights(m):
if type(m) in [nn.Dense, nn.Conv2d]:
m.weight.set_data(mindspore.common.initializer.initializer('xavier_uniform', m.weight.shape))
net.apply(init_weights)
def train_with_data_aug(train_augs, test_augs, net, lr=0.001):
train_iter = load_cifar10(True, train_augs, batch_size)
test_iter = load_cifar10(False, test_augs, batch_size)
loss = nn.CrossEntropyLoss(reduction="none")
trainer = nn.Adam(net.trainable_params(), learning_rate=lr)
train_ch13(net, train_iter, test_iter, loss, trainer, 10)
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
batch_size, devices, net = 256, d2l.try_all_gpus(), d2l.resnet18(10, 3)
def init_weights(m):
if type(m) in [nn.Linear, nn.Conv2d]:
nn.init.xavier_uniform_(m.weight)
net.apply(init_weights)
def train_with_data_aug(train_augs, test_augs, net, lr=0.001):
train_iter = load_cifar10(True, train_augs, batch_size)
test_iter = load_cifar10(False, test_augs, batch_size)
loss = nn.CrossEntropyLoss(reduction="none")
trainer = torch.optim.Adam(net.parameters(), lr=lr)
train_ch13(net, train_iter, test_iter, loss, trainer, 10, devices)
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
train_with_data_aug(train_augs, test_augs, net)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
loss 0.170, train acc 0.940, test acc 0.842
5951.8 examples/sec
.. figure:: output_image-augmentation_7d0887_139_1.svg
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
train_with_data_aug(train_augs, test_augs, net)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
loss 0.168, train acc 0.942, test acc 0.844
8022.7 examples/sec on [device(type='cuda', index=0)]
.. figure:: output_image-augmentation_7d0887_142_1.svg
.. raw:: html
.. raw:: html
.. raw:: html
`讨论 `__
.. raw:: html
.. raw:: html
`讨论 `__
.. raw:: html
.. raw:: html