注意
跳转到末尾 以下载完整示例代码。
空间变换网络教程#
创建日期：2017年11月08日 | 最后更新：2024年1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 的论文 中阅读更多关于空间变换网络的内容。
空间变换网络是将可微注意力机制推广到任何空间变换的一种通用形式。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，以增强模型的几何不变性。例如，它可以裁剪感兴趣区域、缩放并修正图像的方向。这是一个非常有用的机制，因为 CNN 对旋转、缩放以及更一般的仿射变换并不具有不变性。
STN 的最大优点之一是能够以极小的修改将其简单地插入到任何现有的 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 0x7fba9fe315a0>
加载数据#
在本篇内容中，我们使用经典的 MNIST 数据集进行实验。通过一个由空间变换网络增强的标准卷积网络来实现。
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, 101MB/s]
0%| | 0.00/28.9k [00:00<?, ?B/s]
100%|██████████| 28.9k/28.9k [00:00<00:00, 30.1MB/s]
0%| | 0.00/1.65M [00:00<?, ?B/s]
100%|██████████| 1.65M/1.65M [00:00<00:00, 212MB/s]
0%| | 0.00/4.54k [00:00<?, ?B/s]
100%|██████████| 4.54k/4.54k [00:00<00:00, 28.6MB/s]
空间变换网络描述#
空间变换网络可简化为三个主要组件：
定位网络（Localization network）是一个常规的 CNN，用于回归变换参数。变换参数并非直接从数据集中显式学习，相反，网络会自动学习能够提高全局准确率的空间变换。
网格生成器（Grid generator）在输入图像中生成一个坐标网格，该网格与输出图像的每个像素相对应。
采样器（Sampler）使用变换参数并将其应用于输入图像。
注意
我们需要最新版本的 PyTorch，其中包含 affine_grid 和 grid_sample 模块。
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。网络以监督方式学习分类任务。与此同时，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。
我们定义了一个简单的辅助函数，以便在训练期间可视化变换过程。
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: 2.327547
Train Epoch: 1 [32000/60000 (53%)] Loss: 0.776089
/usr/local/lib/python3.10/dist-packages/torch/nn/functional.py:3182: 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: 0.2329, Accuracy: 9280/10000 (93%)
Train Epoch: 2 [0/60000 (0%)] Loss: 0.355935
Train Epoch: 2 [32000/60000 (53%)] Loss: 0.314704
Test set: Average loss: 0.1247, Accuracy: 9601/10000 (96%)
Train Epoch: 3 [0/60000 (0%)] Loss: 0.247603
Train Epoch: 3 [32000/60000 (53%)] Loss: 0.258551
Test set: Average loss: 0.0998, Accuracy: 9698/10000 (97%)
Train Epoch: 4 [0/60000 (0%)] Loss: 0.185652
Train Epoch: 4 [32000/60000 (53%)] Loss: 0.164821
Test set: Average loss: 0.0997, Accuracy: 9697/10000 (97%)
Train Epoch: 5 [0/60000 (0%)] Loss: 0.267513
Train Epoch: 5 [32000/60000 (53%)] Loss: 0.278723
Test set: Average loss: 0.0954, Accuracy: 9706/10000 (97%)
Train Epoch: 6 [0/60000 (0%)] Loss: 0.154455
Train Epoch: 6 [32000/60000 (53%)] Loss: 0.141601
Test set: Average loss: 0.1243, Accuracy: 9621/10000 (96%)
Train Epoch: 7 [0/60000 (0%)] Loss: 0.166564
Train Epoch: 7 [32000/60000 (53%)] Loss: 0.185901
Test set: Average loss: 0.0678, Accuracy: 9787/10000 (98%)
Train Epoch: 8 [0/60000 (0%)] Loss: 0.098844
Train Epoch: 8 [32000/60000 (53%)] Loss: 0.107987
Test set: Average loss: 0.0699, Accuracy: 9786/10000 (98%)
Train Epoch: 9 [0/60000 (0%)] Loss: 0.077916
Train Epoch: 9 [32000/60000 (53%)] Loss: 0.146027
Test set: Average loss: 0.0627, Accuracy: 9807/10000 (98%)
Train Epoch: 10 [0/60000 (0%)] Loss: 0.078330
Train Epoch: 10 [32000/60000 (53%)] Loss: 0.297468
Test set: Average loss: 0.0883, Accuracy: 9736/10000 (97%)
Train Epoch: 11 [0/60000 (0%)] Loss: 0.121364
Train Epoch: 11 [32000/60000 (53%)] Loss: 0.311983
Test set: Average loss: 0.0578, Accuracy: 9832/10000 (98%)
Train Epoch: 12 [0/60000 (0%)] Loss: 0.082238
Train Epoch: 12 [32000/60000 (53%)] Loss: 0.069036
Test set: Average loss: 0.2101, Accuracy: 9376/10000 (94%)
Train Epoch: 13 [0/60000 (0%)] Loss: 0.373635
Train Epoch: 13 [32000/60000 (53%)] Loss: 0.077407
Test set: Average loss: 0.0568, Accuracy: 9841/10000 (98%)
Train Epoch: 14 [0/60000 (0%)] Loss: 0.068573
Train Epoch: 14 [32000/60000 (53%)] Loss: 0.186205
Test set: Average loss: 0.0569, Accuracy: 9841/10000 (98%)
Train Epoch: 15 [0/60000 (0%)] Loss: 0.062182
Train Epoch: 15 [32000/60000 (53%)] Loss: 0.253495
Test set: Average loss: 0.0573, Accuracy: 9828/10000 (98%)
Train Epoch: 16 [0/60000 (0%)] Loss: 0.039149
Train Epoch: 16 [32000/60000 (53%)] Loss: 0.052378
Test set: Average loss: 0.0852, Accuracy: 9758/10000 (98%)
Train Epoch: 17 [0/60000 (0%)] Loss: 0.265677
Train Epoch: 17 [32000/60000 (53%)] Loss: 0.101247
Test set: Average loss: 0.0563, Accuracy: 9823/10000 (98%)
Train Epoch: 18 [0/60000 (0%)] Loss: 0.072422
Train Epoch: 18 [32000/60000 (53%)] Loss: 0.163925
Test set: Average loss: 0.0481, Accuracy: 9861/10000 (99%)
Train Epoch: 19 [0/60000 (0%)] Loss: 0.076223
Train Epoch: 19 [32000/60000 (53%)] Loss: 0.294367
Test set: Average loss: 0.0527, Accuracy: 9850/10000 (98%)
Train Epoch: 20 [0/60000 (0%)] Loss: 0.026749
Train Epoch: 20 [32000/60000 (53%)] Loss: 0.075701
Test set: Average loss: 0.0455, Accuracy: 9863/10000 (99%)
脚本总运行时间： (1 分钟 35.579 秒)