• 教程 >
  • 空间 Transformer 网络
Shortcuts

空间 Transformer 网络

作者Ghassen HAMROUNI

../_images/FSeq.png

在本教程中,你将学习如何使用称为空间 transformer 网络的视觉注意机制来增强网络。 你可以在 DeepMind 论文 中阅读更多有关空间 transformer 网络的内容。

空间 transformer 网络是任何空间变换的可微分注意力的泛化。 空间 transformer 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,它可以裁剪感兴趣的区域、缩放和校正图像的方向。 这是一个有用的机制,因为 CNN 不会对旋转和缩放以及更一般的仿射变换保持不变。

关于 STN 最好的事情之一是能够简单地将其插入任何现有的 CNN,只需很少的修改。

# License: BSD
# Author: Ghassen Hamrouni

from __future__ import print_function
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

加载数据|

在这篇文章中,我们尝试经典的MNIST数据集。 使用标准卷积网络,并用空间 transformer 网络进行增强。

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)

输出:

Downloading http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz to ./MNIST/raw/train-images-idx3-ubyte.gz
Extracting ./MNIST/raw/train-images-idx3-ubyte.gz to ./MNIST/raw
Downloading http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz to ./MNIST/raw/train-labels-idx1-ubyte.gz
Extracting ./MNIST/raw/train-labels-idx1-ubyte.gz to ./MNIST/raw
Downloading http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz to ./MNIST/raw/t10k-images-idx3-ubyte.gz
Extracting ./MNIST/raw/t10k-images-idx3-ubyte.gz to ./MNIST/raw
Downloading http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz to ./MNIST/raw/t10k-labels-idx1-ubyte.gz
Extracting ./MNIST/raw/t10k-labels-idx1-ubyte.gz to ./MNIST/raw
Processing...
Done!

描述空间 transformer 网络

空间 transformer 网络可归结为三个主要组成部分:

  • 本地化网络是一个常规的CNN,它回归转换参数。 转换永远不会从此数据集中显式学习,而是网络自动学习增强全局精度的空间变换。

  • 网格生成器在输入图像中生成与输出图像中每个像素对应的坐标网格。

  • 采样器使用转换的参数并将其应用于输入图像。

../_images/stn-arch.png

注意

我们需要包含affine_grid和grid_sample模块的最新版本的 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。 网络正在以有监督的方式学习分类任务。 同时,模型正在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 STN the 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,我们将检查我们学习的视觉注意机制的结果。

我们定义一个小辅助函数,以便在训练时可视化转换。

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()
../_images/sphx_glr_spatial_transformer_tutorial_001.png

输出:

Train Epoch: 1 [0/60000 (0%)]   Loss: 2.350204
Train Epoch: 1 [32000/60000 (53%)]      Loss: 0.691158

Test set: Average loss: 0.2023, Accuracy: 9402/10000 (94%)

Train Epoch: 2 [0/60000 (0%)]   Loss: 0.463056
Train Epoch: 2 [32000/60000 (53%)]      Loss: 0.313888

Test set: Average loss: 0.1185, Accuracy: 9630/10000 (96%)

Train Epoch: 3 [0/60000 (0%)]   Loss: 0.279039
Train Epoch: 3 [32000/60000 (53%)]      Loss: 0.332558

Test set: Average loss: 0.0930, Accuracy: 9699/10000 (97%)

Train Epoch: 4 [0/60000 (0%)]   Loss: 0.317962
Train Epoch: 4 [32000/60000 (53%)]      Loss: 0.157762

Test set: Average loss: 0.0760, Accuracy: 9775/10000 (98%)

Train Epoch: 5 [0/60000 (0%)]   Loss: 0.271342
Train Epoch: 5 [32000/60000 (53%)]      Loss: 0.145738

Test set: Average loss: 0.1344, Accuracy: 9569/10000 (96%)

Train Epoch: 6 [0/60000 (0%)]   Loss: 0.522058
Train Epoch: 6 [32000/60000 (53%)]      Loss: 0.369435

Test set: Average loss: 0.0611, Accuracy: 9809/10000 (98%)

Train Epoch: 7 [0/60000 (0%)]   Loss: 0.264292
Train Epoch: 7 [32000/60000 (53%)]      Loss: 0.076945

Test set: Average loss: 0.0587, Accuracy: 9819/10000 (98%)

Train Epoch: 8 [0/60000 (0%)]   Loss: 0.105667
Train Epoch: 8 [32000/60000 (53%)]      Loss: 0.327709

Test set: Average loss: 0.0533, Accuracy: 9829/10000 (98%)

Train Epoch: 9 [0/60000 (0%)]   Loss: 0.064149
Train Epoch: 9 [32000/60000 (53%)]      Loss: 0.169669

Test set: Average loss: 0.0604, Accuracy: 9807/10000 (98%)

Train Epoch: 10 [0/60000 (0%)]  Loss: 0.200340
Train Epoch: 10 [32000/60000 (53%)]     Loss: 0.073255

Test set: Average loss: 0.0549, Accuracy: 9832/10000 (98%)

Train Epoch: 11 [0/60000 (0%)]  Loss: 0.191361
Train Epoch: 11 [32000/60000 (53%)]     Loss: 0.020362

Test set: Average loss: 0.0455, Accuracy: 9864/10000 (99%)

Train Epoch: 12 [0/60000 (0%)]  Loss: 0.081118
Train Epoch: 12 [32000/60000 (53%)]     Loss: 0.106922

Test set: Average loss: 0.0490, Accuracy: 9857/10000 (99%)

Train Epoch: 13 [0/60000 (0%)]  Loss: 0.050231
Train Epoch: 13 [32000/60000 (53%)]     Loss: 0.154120

Test set: Average loss: 0.0415, Accuracy: 9876/10000 (99%)

Train Epoch: 14 [0/60000 (0%)]  Loss: 0.154015
Train Epoch: 14 [32000/60000 (53%)]     Loss: 0.022742

Test set: Average loss: 0.0392, Accuracy: 9874/10000 (99%)

Train Epoch: 15 [0/60000 (0%)]  Loss: 0.289922
Train Epoch: 15 [32000/60000 (53%)]     Loss: 0.049584

Test set: Average loss: 0.0383, Accuracy: 9884/10000 (99%)

Train Epoch: 16 [0/60000 (0%)]  Loss: 0.101700
Train Epoch: 16 [32000/60000 (53%)]     Loss: 0.100438

Test set: Average loss: 0.1181, Accuracy: 9658/10000 (97%)

Train Epoch: 17 [0/60000 (0%)]  Loss: 0.476445
Train Epoch: 17 [32000/60000 (53%)]     Loss: 0.089125

Test set: Average loss: 0.0375, Accuracy: 9892/10000 (99%)

Train Epoch: 18 [0/60000 (0%)]  Loss: 0.059651
Train Epoch: 18 [32000/60000 (53%)]     Loss: 0.118905

Test set: Average loss: 0.0384, Accuracy: 9877/10000 (99%)

Train Epoch: 19 [0/60000 (0%)]  Loss: 0.199291
Train Epoch: 19 [32000/60000 (53%)]     Loss: 0.095119

Test set: Average loss: 0.0358, Accuracy: 9887/10000 (99%)

Train Epoch: 20 [0/60000 (0%)]  Loss: 0.221594
Train Epoch: 20 [32000/60000 (53%)]     Loss: 0.019931

Test set: Average loss: 0.0386, Accuracy: 9877/10000 (99%)

脚本总运行时间: (17分24.132秒)

由狮身人面像库生成的画廊