PyTorch 系列教程之空間變換器網路

• 2019 年 10 月 6 日
• 筆記

在本教程中，您將學習如何使用稱為空間變換器網路的視覺注意機制來擴充您的網路。你可以在DeepMind paper 閱讀更多有關空間變換器網路的內容。

空間變換器網路是對任何空間變換的差異化關注的概括。空間變換器網路（簡稱STN）允許神經網路學習如何在輸入影像上執行空間變換， 以增強模型的幾何不變性。例如，它可以裁剪感興趣的區域，縮放並校正影像的方向。而這可能是一種有用的機制，因為CNN對於旋轉和 縮放以及更一般的仿射變換並不是不變的。

關於STN的最棒的事情之一是能夠簡單地將其插入任何現有的CNN，而且只需很少的修改。

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()   # 交互模式  

1.載入數據

在這篇文章中，我們嘗試了經典的 MNIST 數據集。使用標準卷積網路增強空間變換器網路。

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")    # 訓練數據集  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)  # 測試數據集  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)  

• 輸出結果

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

2.什麼是空間變換器網路？

空間變換器網路歸結為三個主要組成部分：

• 本地網路（Localisation Network）是常規CNN，其對變換參數進行回歸。不會從該數據集中明確地學習轉換，而是網路自動學習增強全局準確性的空間變換。
• 網格生成器( Grid Genator)在輸入影像中生成與輸出影像中的每個像素相對應的坐標網格。
• 取樣器（Sampler）使用變換的參數並將其應用於輸入影像。

注意： 我們使用最新版本的Pytorch，它應該包含affine_grid和grid_sample模組。

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)            # 空間變換器定位 - 網路          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)          )            # 3 * 2 affine矩陣的回歸量          self.fc_loc = nn.Sequential(              nn.Linear(10 * 3 * 3, 32),              nn.ReLU(True),              nn.Linear(32, 3 * 2)          )            # 使用身份轉換初始化權重/偏差          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))        # 空間變換器網路轉發功能      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)            # 執行一般的前進傳遞          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)  

3.訓練模型

訓練模型 現在我們使用 SGD（隨機梯度下降）演算法來訓練模型。網路正在以有監督的方式學習分類任務。同時，該模型以端到端的方式自動學習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()))  #  # 一種簡單的測試程式，用於測量STN在MNIST上的性能。.  #    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)                # 累加批量損失              test_loss += F.nll_loss(output, target, size_average=False).item()              # 獲取最大對數概率的索引              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)))  

4.可視化 STN 結果

現在，我們將檢查我們學習的視覺注意機制的結果。

我們定義了一個小輔助函數，以便在訓練時可視化變換。

def convert_http://pytorch.panchuang.net/FourSection/image_np(inp):      """Convert a Tensor to numpy http://pytorch.panchuang.net/FourSection/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    # 我們想要在訓練之後可視化空間變換器層的輸出  # 我們使用STN可視化一批輸入影像和相應的變換批次。  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_http://pytorch.panchuang.net/FourSection/image_np(              torchvision.utils.make_grid(input_tensor))            out_grid = convert_http://pytorch.panchuang.net/FourSection/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 http://pytorch.panchuang.net/FourSection/images')            axarr[1].imshow(out_grid)          axarr[1].set_title('Transformed http://pytorch.panchuang.net/FourSection/images')    for epoch in range(1, 20 + 1):      train(epoch)      test()    # 在某些輸入批處理上可視化STN轉換  visualize_stn()    plt.ioff()  plt.show()  

• 輸出結果

Train Epoch: 1 [0/60000 (0%)]   Loss: 2.336866  Train Epoch: 1 [32000/60000 (53%)]      Loss: 0.841600    Test set: Average loss: 0.2624, Accuracy: 9212/10000 (92%)    Train Epoch: 2 [0/60000 (0%)]   Loss: 0.527656  Train Epoch: 2 [32000/60000 (53%)]      Loss: 0.428908    Test set: Average loss: 0.1176, Accuracy: 9632/10000 (96%)    Train Epoch: 3 [0/60000 (0%)]   Loss: 0.305364  Train Epoch: 3 [32000/60000 (53%)]      Loss: 0.263615    Test set: Average loss: 0.1099, Accuracy: 9677/10000 (97%)    Train Epoch: 4 [0/60000 (0%)]   Loss: 0.169776  Train Epoch: 4 [32000/60000 (53%)]      Loss: 0.408683    Test set: Average loss: 0.0861, Accuracy: 9734/10000 (97%)    Train Epoch: 5 [0/60000 (0%)]   Loss: 0.286635  Train Epoch: 5 [32000/60000 (53%)]      Loss: 0.122162    Test set: Average loss: 0.0817, Accuracy: 9743/10000 (97%)    Train Epoch: 6 [0/60000 (0%)]   Loss: 0.331074  Train Epoch: 6 [32000/60000 (53%)]      Loss: 0.126413    Test set: Average loss: 0.0589, Accuracy: 9822/10000 (98%)    Train Epoch: 7 [0/60000 (0%)]   Loss: 0.109780  Train Epoch: 7 [32000/60000 (53%)]      Loss: 0.172199    Test set: Average loss: 0.0629, Accuracy: 9814/10000 (98%)    Train Epoch: 8 [0/60000 (0%)]   Loss: 0.078934  Train Epoch: 8 [32000/60000 (53%)]      Loss: 0.156452    Test set: Average loss: 0.0563, Accuracy: 9839/10000 (98%)    Train Epoch: 9 [0/60000 (0%)]   Loss: 0.063500  Train Epoch: 9 [32000/60000 (53%)]      Loss: 0.186023    Test set: Average loss: 0.0713, Accuracy: 9799/10000 (98%)    Train Epoch: 10 [0/60000 (0%)]  Loss: 0.199808  Train Epoch: 10 [32000/60000 (53%)]     Loss: 0.083502    Test set: Average loss: 0.0528, Accuracy: 9850/10000 (98%)    Train Epoch: 11 [0/60000 (0%)]  Loss: 0.092909  Train Epoch: 11 [32000/60000 (53%)]     Loss: 0.204410    Test set: Average loss: 0.0471, Accuracy: 9857/10000 (99%)    Train Epoch: 12 [0/60000 (0%)]  Loss: 0.078322  Train Epoch: 12 [32000/60000 (53%)]     Loss: 0.041391    Test set: Average loss: 0.0634, Accuracy: 9796/10000 (98%)    Train Epoch: 13 [0/60000 (0%)]  Loss: 0.061228  Train Epoch: 13 [32000/60000 (53%)]     Loss: 0.137952    Test set: Average loss: 0.0654, Accuracy: 9802/10000 (98%)    Train Epoch: 14 [0/60000 (0%)]  Loss: 0.068635  Train Epoch: 14 [32000/60000 (53%)]     Loss: 0.084583    Test set: Average loss: 0.0515, Accuracy: 9853/10000 (99%)    Train Epoch: 15 [0/60000 (0%)]  Loss: 0.263158  Train Epoch: 15 [32000/60000 (53%)]     Loss: 0.127036    Test set: Average loss: 0.0493, Accuracy: 9851/10000 (99%)    Train Epoch: 16 [0/60000 (0%)]  Loss: 0.083642  Train Epoch: 16 [32000/60000 (53%)]     Loss: 0.028274    Test set: Average loss: 0.0461, Accuracy: 9867/10000 (99%)    Train Epoch: 17 [0/60000 (0%)]  Loss: 0.076734  Train Epoch: 17 [32000/60000 (53%)]     Loss: 0.034796    Test set: Average loss: 0.0409, Accuracy: 9864/10000 (99%)    Train Epoch: 18 [0/60000 (0%)]  Loss: 0.122501  Train Epoch: 18 [32000/60000 (53%)]     Loss: 0.152187    Test set: Average loss: 0.0474, Accuracy: 9860/10000 (99%)    Train Epoch: 19 [0/60000 (0%)]  Loss: 0.050512  Train Epoch: 19 [32000/60000 (53%)]     Loss: 0.270055    Test set: Average loss: 0.0416, Accuracy: 9878/10000 (99%)    Train Epoch: 20 [0/60000 (0%)]  Loss: 0.073357  Train Epoch: 20 [32000/60000 (53%)]     Loss: 0.017542    Test set: Average loss: 0.0713, Accuracy: 9816/10000 (98%)