注意
跳转至页面底部下载完整示例代码。
空间变换网络教程#
创建日期：2017年11月08日 | 最后更新：2024年01月19日 | 最后验证：2024年11月05日
作者: Ghassen HAMROUNI
在本教程中，您将学习如何使用一种称为空间变换网络（Spatial Transformer Networks）的视觉注意力机制来增强您的网络。您可以在 DeepMind 论文 中阅读有关空间变换网络的更多信息。
空间变换网络是可微注意力机制对任意空间变换的推广。空间变换网络（简称 STN）允许神经网络学习如何对输入图像执行空间变换，从而提高模型的几何不变性。例如，它可以裁剪感兴趣区域，缩放并校正图像的方向。这是一个非常有用的机制，因为卷积神经网络（CNN）本身对旋转、缩放及更通用的仿射变换并不具有不变性。
STN 的一大优点是能够以极少的修改轻松地将其嵌入到任何现有的 CNN 中。
# License: BSD
# Author: Ghassen Hamrouni
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import torchvision
from torchvision import datasets, transforms
import matplotlib.pyplot as plt
import numpy as np
plt.ion() # interactive mode
<contextlib.ExitStack object at 0x7f5f621dbd00>
加载数据#
在这篇文章中，我们将使用经典的 MNIST 数据集进行实验，利用一个通过空间变换网络增强的标准卷积神经网络。
from six.moves import urllib
opener = urllib.request.build_opener()
opener.addheaders = [('User-agent', 'Mozilla/5.0')]
urllib.request.install_opener(opener)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# Training dataset
train_loader = torch.utils.data.DataLoader(
datasets.MNIST(root='.', train=True, download=True,
transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307,), (0.3081,))
])), batch_size=64, shuffle=True, num_workers=4)
# Test dataset
test_loader = torch.utils.data.DataLoader(
datasets.MNIST(root='.', train=False, transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307,), (0.3081,))
])), batch_size=64, shuffle=True, num_workers=4)
0%| | 0.00/9.91M [00:00<?, ?B/s]
100%|██████████| 9.91M/9.91M [00:00<00:00, 124MB/s]
0%| | 0.00/28.9k [00:00<?, ?B/s]
100%|██████████| 28.9k/28.9k [00:00<00:00, 30.3MB/s]
0%| | 0.00/1.65M [00:00<?, ?B/s]
100%|██████████| 1.65M/1.65M [00:00<00:00, 54.9MB/s]
0%| | 0.00/4.54k [00:00<?, ?B/s]
100%|██████████| 4.54k/4.54k [00:00<00:00, 21.7MB/s]
解析空间变换网络#
空间变换网络归结为三个主要组件：
定位网络是一个常规的 CNN，用于回归变换参数。该变换并非从数据集中显式学习，而是网络自动学习能够提高全局准确率的空间变换。
网格生成器用于生成输入图像中与输出图像每个像素对应的坐标网格。
采样器使用变换的参数并将其应用于输入图像。
注意
我们需要包含 affine_grid 和 grid_sample 模块的最新版 PyTorch。
class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = nn.Conv2d(1, 10, kernel_size=5)
self.conv2 = nn.Conv2d(10, 20, kernel_size=5)
self.conv2_drop = nn.Dropout2d()
self.fc1 = nn.Linear(320, 50)
self.fc2 = nn.Linear(50, 10)
# Spatial transformer localization-network
self.localization = nn.Sequential(
nn.Conv2d(1, 8, kernel_size=7),
nn.MaxPool2d(2, stride=2),
nn.ReLU(True),
nn.Conv2d(8, 10, kernel_size=5),
nn.MaxPool2d(2, stride=2),
nn.ReLU(True)
)
# Regressor for the 3 * 2 affine matrix
self.fc_loc = nn.Sequential(
nn.Linear(10 * 3 * 3, 32),
nn.ReLU(True),
nn.Linear(32, 3 * 2)
)
# Initialize the weights/bias with identity transformation
self.fc_loc[2].weight.data.zero_()
self.fc_loc[2].bias.data.copy_(torch.tensor([1, 0, 0, 0, 1, 0], dtype=torch.float))
# Spatial transformer network forward function
def stn(self, x):
xs = self.localization(x)
xs = xs.view(-1, 10 * 3 * 3)
theta = self.fc_loc(xs)
theta = theta.view(-1, 2, 3)
grid = F.affine_grid(theta, x.size())
x = F.grid_sample(x, grid)
return x
def forward(self, x):
# transform the input
x = self.stn(x)
# Perform the usual forward pass
x = F.relu(F.max_pool2d(self.conv1(x), 2))
x = F.relu(F.max_pool2d(self.conv2_drop(self.conv2(x)), 2))
x = x.view(-1, 320)
x = F.relu(self.fc1(x))
x = F.dropout(x, training=self.training)
x = self.fc2(x)
return F.log_softmax(x, dim=1)
model = Net().to(device)
训练模型#
现在，让我们使用 SGD 算法来训练模型。网络以监督学习的方式学习分类任务。同时，模型以端到端的方式自动学习 STN。
optimizer = optim.SGD(model.parameters(), lr=0.01)
def train(epoch):
model.train()
for batch_idx, (data, target) in enumerate(train_loader):
data, target = data.to(device), target.to(device)
optimizer.zero_grad()
output = model(data)
loss = F.nll_loss(output, target)
loss.backward()
optimizer.step()
if batch_idx % 500 == 0:
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
epoch, batch_idx * len(data), len(train_loader.dataset),
100. * batch_idx / len(train_loader), loss.item()))
#
# A simple test procedure to measure the STN performances on MNIST.
#
def test():
with torch.no_grad():
model.eval()
test_loss = 0
correct = 0
for data, target in test_loader:
data, target = data.to(device), target.to(device)
output = model(data)
# sum up batch loss
test_loss += F.nll_loss(output, target, size_average=False).item()
# get the index of the max log-probability
pred = output.max(1, keepdim=True)[1]
correct += pred.eq(target.view_as(pred)).sum().item()
test_loss /= len(test_loader.dataset)
print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'
.format(test_loss, correct, len(test_loader.dataset),
100. * correct / len(test_loader.dataset)))
可视化 STN 结果#
现在，我们将检查学习到的视觉注意力机制的结果。
我们定义了一个小的辅助函数，以便在训练过程中可视化变换情况。
def convert_image_np(inp):
"""Convert a Tensor to numpy image."""
inp = inp.numpy().transpose((1, 2, 0))
mean = np.array([0.485, 0.456, 0.406])
std = np.array([0.229, 0.224, 0.225])
inp = std * inp + mean
inp = np.clip(inp, 0, 1)
return inp
# We want to visualize the output of the spatial transformers layer
# after the training, we visualize a batch of input images and
# the corresponding transformed batch using STN.
def visualize_stn():
with torch.no_grad():
# Get a batch of training data
data = next(iter(test_loader))[0].to(device)
input_tensor = data.cpu()
transformed_input_tensor = model.stn(data).cpu()
in_grid = convert_image_np(
torchvision.utils.make_grid(input_tensor))
out_grid = convert_image_np(
torchvision.utils.make_grid(transformed_input_tensor))
# Plot the results side-by-side
f, axarr = plt.subplots(1, 2)
axarr[0].imshow(in_grid)
axarr[0].set_title('Dataset Images')
axarr[1].imshow(out_grid)
axarr[1].set_title('Transformed Images')
for epoch in range(1, 20 + 1):
train(epoch)
test()
# Visualize the STN transformation on some input batch
visualize_stn()
plt.ioff()
plt.show()
/var/lib/workspace/intermediate_source/spatial_transformer_tutorial.py:130: UserWarning: Default grid_sample and affine_grid behavior has changed to align_corners=False since 1.3.0. Please specify align_corners=True if the old behavior is desired. See the documentation of grid_sample for details.
grid = F.affine_grid(theta, x.size())
/var/lib/workspace/intermediate_source/spatial_transformer_tutorial.py:131: UserWarning: Default grid_sample and affine_grid behavior has changed to align_corners=False since 1.3.0. Please specify align_corners=True if the old behavior is desired. See the documentation of grid_sample for details.
x = F.grid_sample(x, grid)
Train Epoch: 1 [0/60000 (0%)] Loss: 2.354045
Train Epoch: 1 [32000/60000 (53%)] Loss: 0.829746
/var/lib/ci-user/.local/lib/python3.10/site-packages/torch/nn/functional.py:3230: UserWarning: size_average and reduce args will be deprecated, please use reduction='sum' instead.
reduction = _Reduction.legacy_get_string(size_average, reduce)
Test set: Average loss: 0.2309, Accuracy: 9358/10000 (94%)
Train Epoch: 2 [0/60000 (0%)] Loss: 0.526761
Train Epoch: 2 [32000/60000 (53%)] Loss: 0.359206
Test set: Average loss: 0.4360, Accuracy: 8599/10000 (86%)
Train Epoch: 3 [0/60000 (0%)] Loss: 1.097828
Train Epoch: 3 [32000/60000 (53%)] Loss: 0.219073
Test set: Average loss: 0.1543, Accuracy: 9517/10000 (95%)
Train Epoch: 4 [0/60000 (0%)] Loss: 0.275019
Train Epoch: 4 [32000/60000 (53%)] Loss: 0.119273
Test set: Average loss: 0.1115, Accuracy: 9662/10000 (97%)
Train Epoch: 5 [0/60000 (0%)] Loss: 0.369317
Train Epoch: 5 [32000/60000 (53%)] Loss: 0.155932
Test set: Average loss: 0.1049, Accuracy: 9667/10000 (97%)
Train Epoch: 6 [0/60000 (0%)] Loss: 0.357618
Train Epoch: 6 [32000/60000 (53%)] Loss: 0.133235
Test set: Average loss: 0.0579, Accuracy: 9812/10000 (98%)
Train Epoch: 7 [0/60000 (0%)] Loss: 0.341314
Train Epoch: 7 [32000/60000 (53%)] Loss: 0.041031
Test set: Average loss: 0.0556, Accuracy: 9831/10000 (98%)
Train Epoch: 8 [0/60000 (0%)] Loss: 0.283948
Train Epoch: 8 [32000/60000 (53%)] Loss: 0.063095
Test set: Average loss: 0.0563, Accuracy: 9829/10000 (98%)
Train Epoch: 9 [0/60000 (0%)] Loss: 0.232695
Train Epoch: 9 [32000/60000 (53%)] Loss: 0.136149
Test set: Average loss: 0.0588, Accuracy: 9829/10000 (98%)
Train Epoch: 10 [0/60000 (0%)] Loss: 0.220676
Train Epoch: 10 [32000/60000 (53%)] Loss: 0.082914
Test set: Average loss: 0.0467, Accuracy: 9853/10000 (99%)
Train Epoch: 11 [0/60000 (0%)] Loss: 0.052282
Train Epoch: 11 [32000/60000 (53%)] Loss: 0.033466
Test set: Average loss: 0.0473, Accuracy: 9847/10000 (98%)
Train Epoch: 12 [0/60000 (0%)] Loss: 0.071206
Train Epoch: 12 [32000/60000 (53%)] Loss: 0.150549
Test set: Average loss: 0.0432, Accuracy: 9862/10000 (99%)
Train Epoch: 13 [0/60000 (0%)] Loss: 0.192705
Train Epoch: 13 [32000/60000 (53%)] Loss: 0.055879
Test set: Average loss: 0.0646, Accuracy: 9810/10000 (98%)
Train Epoch: 14 [0/60000 (0%)] Loss: 0.214826
Train Epoch: 14 [32000/60000 (53%)] Loss: 0.210343
Test set: Average loss: 0.0438, Accuracy: 9882/10000 (99%)
Train Epoch: 15 [0/60000 (0%)] Loss: 0.057981
Train Epoch: 15 [32000/60000 (53%)] Loss: 0.078952
Test set: Average loss: 0.0442, Accuracy: 9862/10000 (99%)
Train Epoch: 16 [0/60000 (0%)] Loss: 0.046081
Train Epoch: 16 [32000/60000 (53%)] Loss: 0.031271
Test set: Average loss: 0.0510, Accuracy: 9855/10000 (99%)
Train Epoch: 17 [0/60000 (0%)] Loss: 0.117700
Train Epoch: 17 [32000/60000 (53%)] Loss: 0.079447
Test set: Average loss: 0.0433, Accuracy: 9863/10000 (99%)
Train Epoch: 18 [0/60000 (0%)] Loss: 0.144576
Train Epoch: 18 [32000/60000 (53%)] Loss: 0.158445
Test set: Average loss: 0.0382, Accuracy: 9883/10000 (99%)
Train Epoch: 19 [0/60000 (0%)] Loss: 0.100756
Train Epoch: 19 [32000/60000 (53%)] Loss: 0.057655
Test set: Average loss: 0.0604, Accuracy: 9813/10000 (98%)
Train Epoch: 20 [0/60000 (0%)] Loss: 0.124034
Train Epoch: 20 [32000/60000 (53%)] Loss: 0.110330
Test set: Average loss: 0.0399, Accuracy: 9886/10000 (99%)
脚本总运行时间： (1 分钟 36.701 秒)