Repository: yunjey/pytorch-tutorial Branch: master Commit: 0500d3df5a2a Files: 31 Total size: 80.7 KB Directory structure: gitextract_tc4rlnxl/ ├── .gitignore ├── LICENSE ├── README.md └── tutorials/ ├── 01-basics/ │ ├── feedforward_neural_network/ │ │ └── main.py │ ├── linear_regression/ │ │ └── main.py │ ├── logistic_regression/ │ │ └── main.py │ └── pytorch_basics/ │ └── main.py ├── 02-intermediate/ │ ├── bidirectional_recurrent_neural_network/ │ │ └── main.py │ ├── convolutional_neural_network/ │ │ └── main.py │ ├── deep_residual_network/ │ │ └── main.py │ ├── language_model/ │ │ ├── data_utils.py │ │ └── main.py │ └── recurrent_neural_network/ │ └── main.py ├── 03-advanced/ │ ├── generative_adversarial_network/ │ │ └── main.py │ ├── image_captioning/ │ │ ├── README.md │ │ ├── build_vocab.py │ │ ├── data_loader.py │ │ ├── download.sh │ │ ├── model.py │ │ ├── requirements.txt │ │ ├── resize.py │ │ ├── sample.py │ │ └── train.py │ ├── neural_style_transfer/ │ │ ├── README.md │ │ ├── main.py │ │ └── requirements.txt │ └── variational_autoencoder/ │ └── main.py └── 04-utils/ └── tensorboard/ ├── README.md ├── logger.py ├── main.py └── requirements.txt ================================================ FILE CONTENTS ================================================ ================================================ FILE: .gitignore ================================================ *.pkl *.zip data/ .ipynb_checkpoints ================================================ FILE: LICENSE ================================================ MIT License Copyright (c) 2017 Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. ================================================ FILE: README.md ================================================

-------------------------------------------------------------------------------- This repository provides tutorial code for deep learning researchers to learn [PyTorch](https://github.com/pytorch/pytorch). In the tutorial, most of the models were implemented with less than 30 lines of code. Before starting this tutorial, it is recommended to finish [Official Pytorch Tutorial](http://pytorch.org/tutorials/beginner/deep_learning_60min_blitz.html).
## Table of Contents #### 1. Basics * [PyTorch Basics](https://github.com/yunjey/pytorch-tutorial/tree/master/tutorials/01-basics/pytorch_basics/main.py) * [Linear Regression](https://github.com/yunjey/pytorch-tutorial/tree/master/tutorials/01-basics/linear_regression/main.py#L22-L23) * [Logistic Regression](https://github.com/yunjey/pytorch-tutorial/tree/master/tutorials/01-basics/logistic_regression/main.py#L33-L34) * [Feedforward Neural Network](https://github.com/yunjey/pytorch-tutorial/tree/master/tutorials/01-basics/feedforward_neural_network/main.py#L37-L49) #### 2. Intermediate * [Convolutional Neural Network](https://github.com/yunjey/pytorch-tutorial/tree/master/tutorials/02-intermediate/convolutional_neural_network/main.py#L35-L56) * [Deep Residual Network](https://github.com/yunjey/pytorch-tutorial/tree/master/tutorials/02-intermediate/deep_residual_network/main.py#L76-L113) * [Recurrent Neural Network](https://github.com/yunjey/pytorch-tutorial/tree/master/tutorials/02-intermediate/recurrent_neural_network/main.py#L39-L58) * [Bidirectional Recurrent Neural Network](https://github.com/yunjey/pytorch-tutorial/tree/master/tutorials/02-intermediate/bidirectional_recurrent_neural_network/main.py#L39-L58) * [Language Model (RNN-LM)](https://github.com/yunjey/pytorch-tutorial/tree/master/tutorials/02-intermediate/language_model/main.py#L30-L50) #### 3. Advanced * [Generative Adversarial Networks](https://github.com/yunjey/pytorch-tutorial/blob/master/tutorials/03-advanced/generative_adversarial_network/main.py#L41-L57) * [Variational Auto-Encoder](https://github.com/yunjey/pytorch-tutorial/blob/master/tutorials/03-advanced/variational_autoencoder/main.py#L38-L65) * [Neural Style Transfer](https://github.com/yunjey/pytorch-tutorial/tree/master/tutorials/03-advanced/neural_style_transfer) * [Image Captioning (CNN-RNN)](https://github.com/yunjey/pytorch-tutorial/tree/master/tutorials/03-advanced/image_captioning) #### 4. Utilities * [TensorBoard in PyTorch](https://github.com/yunjey/pytorch-tutorial/tree/master/tutorials/04-utils/tensorboard)
## Getting Started ```bash $ git clone https://github.com/yunjey/pytorch-tutorial.git $ cd pytorch-tutorial/tutorials/PATH_TO_PROJECT $ python main.py ```
## Dependencies * [Python 2.7 or 3.5+](https://www.continuum.io/downloads) * [PyTorch 0.4.0+](http://pytorch.org/) ================================================ FILE: tutorials/01-basics/feedforward_neural_network/main.py ================================================ import torch import torch.nn as nn import torchvision import torchvision.transforms as transforms # Device configuration device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') # Hyper-parameters input_size = 784 hidden_size = 500 num_classes = 10 num_epochs = 5 batch_size = 100 learning_rate = 0.001 # MNIST dataset train_dataset = torchvision.datasets.MNIST(root='../../data', train=True, transform=transforms.ToTensor(), download=True) test_dataset = torchvision.datasets.MNIST(root='../../data', train=False, transform=transforms.ToTensor()) # Data loader train_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=batch_size, shuffle=True) test_loader = torch.utils.data.DataLoader(dataset=test_dataset, batch_size=batch_size, shuffle=False) # Fully connected neural network with one hidden layer class NeuralNet(nn.Module): def __init__(self, input_size, hidden_size, num_classes): super(NeuralNet, self).__init__() self.fc1 = nn.Linear(input_size, hidden_size) self.relu = nn.ReLU() self.fc2 = nn.Linear(hidden_size, num_classes) def forward(self, x): out = self.fc1(x) out = self.relu(out) out = self.fc2(out) return out model = NeuralNet(input_size, hidden_size, num_classes).to(device) # Loss and optimizer criterion = nn.CrossEntropyLoss() optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate) # Train the model total_step = len(train_loader) for epoch in range(num_epochs): for i, (images, labels) in enumerate(train_loader): # Move tensors to the configured device images = images.reshape(-1, 28*28).to(device) labels = labels.to(device) # Forward pass outputs = model(images) loss = criterion(outputs, labels) # Backward and optimize optimizer.zero_grad() loss.backward() optimizer.step() if (i+1) % 100 == 0: print ('Epoch [{}/{}], Step [{}/{}], Loss: {:.4f}' .format(epoch+1, num_epochs, i+1, total_step, loss.item())) # Test the model # In test phase, we don't need to compute gradients (for memory efficiency) with torch.no_grad(): correct = 0 total = 0 for images, labels in test_loader: images = images.reshape(-1, 28*28).to(device) labels = labels.to(device) outputs = model(images) _, predicted = torch.max(outputs.data, 1) total += labels.size(0) correct += (predicted == labels).sum().item() print('Accuracy of the network on the 10000 test images: {} %'.format(100 * correct / total)) # Save the model checkpoint torch.save(model.state_dict(), 'model.ckpt') ================================================ FILE: tutorials/01-basics/linear_regression/main.py ================================================ import torch import torch.nn as nn import numpy as np import matplotlib.pyplot as plt # Hyper-parameters input_size = 1 output_size = 1 num_epochs = 60 learning_rate = 0.001 # Toy dataset x_train = np.array([[3.3], [4.4], [5.5], [6.71], [6.93], [4.168], [9.779], [6.182], [7.59], [2.167], [7.042], [10.791], [5.313], [7.997], [3.1]], dtype=np.float32) y_train = np.array([[1.7], [2.76], [2.09], [3.19], [1.694], [1.573], [3.366], [2.596], [2.53], [1.221], [2.827], [3.465], [1.65], [2.904], [1.3]], dtype=np.float32) # Linear regression model model = nn.Linear(input_size, output_size) # Loss and optimizer criterion = nn.MSELoss() optimizer = torch.optim.SGD(model.parameters(), lr=learning_rate) # Train the model for epoch in range(num_epochs): # Convert numpy arrays to torch tensors inputs = torch.from_numpy(x_train) targets = torch.from_numpy(y_train) # Forward pass outputs = model(inputs) loss = criterion(outputs, targets) # Backward and optimize optimizer.zero_grad() loss.backward() optimizer.step() if (epoch+1) % 5 == 0: print ('Epoch [{}/{}], Loss: {:.4f}'.format(epoch+1, num_epochs, loss.item())) # Plot the graph predicted = model(torch.from_numpy(x_train)).detach().numpy() plt.plot(x_train, y_train, 'ro', label='Original data') plt.plot(x_train, predicted, label='Fitted line') plt.legend() plt.show() # Save the model checkpoint torch.save(model.state_dict(), 'model.ckpt') ================================================ FILE: tutorials/01-basics/logistic_regression/main.py ================================================ import torch import torch.nn as nn import torchvision import torchvision.transforms as transforms # Hyper-parameters input_size = 28 * 28 # 784 num_classes = 10 num_epochs = 5 batch_size = 100 learning_rate = 0.001 # MNIST dataset (images and labels) train_dataset = torchvision.datasets.MNIST(root='../../data', train=True, transform=transforms.ToTensor(), download=True) test_dataset = torchvision.datasets.MNIST(root='../../data', train=False, transform=transforms.ToTensor()) # Data loader (input pipeline) train_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=batch_size, shuffle=True) test_loader = torch.utils.data.DataLoader(dataset=test_dataset, batch_size=batch_size, shuffle=False) # Logistic regression model model = nn.Linear(input_size, num_classes) # Loss and optimizer # nn.CrossEntropyLoss() computes softmax internally criterion = nn.CrossEntropyLoss() optimizer = torch.optim.SGD(model.parameters(), lr=learning_rate) # Train the model total_step = len(train_loader) for epoch in range(num_epochs): for i, (images, labels) in enumerate(train_loader): # Reshape images to (batch_size, input_size) images = images.reshape(-1, input_size) # Forward pass outputs = model(images) loss = criterion(outputs, labels) # Backward and optimize optimizer.zero_grad() loss.backward() optimizer.step() if (i+1) % 100 == 0: print ('Epoch [{}/{}], Step [{}/{}], Loss: {:.4f}' .format(epoch+1, num_epochs, i+1, total_step, loss.item())) # Test the model # In test phase, we don't need to compute gradients (for memory efficiency) with torch.no_grad(): correct = 0 total = 0 for images, labels in test_loader: images = images.reshape(-1, input_size) outputs = model(images) _, predicted = torch.max(outputs.data, 1) total += labels.size(0) correct += (predicted == labels).sum() print('Accuracy of the model on the 10000 test images: {} %'.format(100 * correct / total)) # Save the model checkpoint torch.save(model.state_dict(), 'model.ckpt') ================================================ FILE: tutorials/01-basics/pytorch_basics/main.py ================================================ import torch import torchvision import torch.nn as nn import numpy as np import torchvision.transforms as transforms # ================================================================== # # Table of Contents # # ================================================================== # # 1. Basic autograd example 1 (Line 25 to 39) # 2. Basic autograd example 2 (Line 46 to 83) # 3. Loading data from numpy (Line 90 to 97) # 4. Input pipline (Line 104 to 129) # 5. Input pipline for custom dataset (Line 136 to 156) # 6. Pretrained model (Line 163 to 176) # 7. Save and load model (Line 183 to 189) # ================================================================== # # 1. Basic autograd example 1 # # ================================================================== # # Create tensors. x = torch.tensor(1., requires_grad=True) w = torch.tensor(2., requires_grad=True) b = torch.tensor(3., requires_grad=True) # Build a computational graph. y = w * x + b # y = 2 * x + 3 # Compute gradients. y.backward() # Print out the gradients. print(x.grad) # x.grad = 2 print(w.grad) # w.grad = 1 print(b.grad) # b.grad = 1 # ================================================================== # # 2. Basic autograd example 2 # # ================================================================== # # Create tensors of shape (10, 3) and (10, 2). x = torch.randn(10, 3) y = torch.randn(10, 2) # Build a fully connected layer. linear = nn.Linear(3, 2) print ('w: ', linear.weight) print ('b: ', linear.bias) # Build loss function and optimizer. criterion = nn.MSELoss() optimizer = torch.optim.SGD(linear.parameters(), lr=0.01) # Forward pass. pred = linear(x) # Compute loss. loss = criterion(pred, y) print('loss: ', loss.item()) # Backward pass. loss.backward() # Print out the gradients. print ('dL/dw: ', linear.weight.grad) print ('dL/db: ', linear.bias.grad) # 1-step gradient descent. optimizer.step() # You can also perform gradient descent at the low level. # linear.weight.data.sub_(0.01 * linear.weight.grad.data) # linear.bias.data.sub_(0.01 * linear.bias.grad.data) # Print out the loss after 1-step gradient descent. pred = linear(x) loss = criterion(pred, y) print('loss after 1 step optimization: ', loss.item()) # ================================================================== # # 3. Loading data from numpy # # ================================================================== # # Create a numpy array. x = np.array([[1, 2], [3, 4]]) # Convert the numpy array to a torch tensor. y = torch.from_numpy(x) # Convert the torch tensor to a numpy array. z = y.numpy() # ================================================================== # # 4. Input pipeline # # ================================================================== # # Download and construct CIFAR-10 dataset. train_dataset = torchvision.datasets.CIFAR10(root='../../data/', train=True, transform=transforms.ToTensor(), download=True) # Fetch one data pair (read data from disk). image, label = train_dataset[0] print (image.size()) print (label) # Data loader (this provides queues and threads in a very simple way). train_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=64, shuffle=True) # When iteration starts, queue and thread start to load data from files. data_iter = iter(train_loader) # Mini-batch images and labels. images, labels = data_iter.next() # Actual usage of the data loader is as below. for images, labels in train_loader: # Training code should be written here. pass # ================================================================== # # 5. Input pipeline for custom dataset # # ================================================================== # # You should build your custom dataset as below. class CustomDataset(torch.utils.data.Dataset): def __init__(self): # TODO # 1. Initialize file paths or a list of file names. pass def __getitem__(self, index): # TODO # 1. Read one data from file (e.g. using numpy.fromfile, PIL.Image.open). # 2. Preprocess the data (e.g. torchvision.Transform). # 3. Return a data pair (e.g. image and label). pass def __len__(self): # You should change 0 to the total size of your dataset. return 0 # You can then use the prebuilt data loader. custom_dataset = CustomDataset() train_loader = torch.utils.data.DataLoader(dataset=custom_dataset, batch_size=64, shuffle=True) # ================================================================== # # 6. Pretrained model # # ================================================================== # # Download and load the pretrained ResNet-18. resnet = torchvision.models.resnet18(pretrained=True) # If you want to finetune only the top layer of the model, set as below. for param in resnet.parameters(): param.requires_grad = False # Replace the top layer for finetuning. resnet.fc = nn.Linear(resnet.fc.in_features, 100) # 100 is an example. # Forward pass. images = torch.randn(64, 3, 224, 224) outputs = resnet(images) print (outputs.size()) # (64, 100) # ================================================================== # # 7. Save and load the model # # ================================================================== # # Save and load the entire model. torch.save(resnet, 'model.ckpt') model = torch.load('model.ckpt') # Save and load only the model parameters (recommended). torch.save(resnet.state_dict(), 'params.ckpt') resnet.load_state_dict(torch.load('params.ckpt')) ================================================ FILE: tutorials/02-intermediate/bidirectional_recurrent_neural_network/main.py ================================================ import torch import torch.nn as nn import torchvision import torchvision.transforms as transforms # Device configuration device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') # Hyper-parameters sequence_length = 28 input_size = 28 hidden_size = 128 num_layers = 2 num_classes = 10 batch_size = 100 num_epochs = 2 learning_rate = 0.003 # MNIST dataset train_dataset = torchvision.datasets.MNIST(root='../../data/', train=True, transform=transforms.ToTensor(), download=True) test_dataset = torchvision.datasets.MNIST(root='../../data/', train=False, transform=transforms.ToTensor()) # Data loader train_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=batch_size, shuffle=True) test_loader = torch.utils.data.DataLoader(dataset=test_dataset, batch_size=batch_size, shuffle=False) # Bidirectional recurrent neural network (many-to-one) class BiRNN(nn.Module): def __init__(self, input_size, hidden_size, num_layers, num_classes): super(BiRNN, self).__init__() self.hidden_size = hidden_size self.num_layers = num_layers self.lstm = nn.LSTM(input_size, hidden_size, num_layers, batch_first=True, bidirectional=True) self.fc = nn.Linear(hidden_size*2, num_classes) # 2 for bidirection def forward(self, x): # Set initial states h0 = torch.zeros(self.num_layers*2, x.size(0), self.hidden_size).to(device) # 2 for bidirection c0 = torch.zeros(self.num_layers*2, x.size(0), self.hidden_size).to(device) # Forward propagate LSTM out, _ = self.lstm(x, (h0, c0)) # out: tensor of shape (batch_size, seq_length, hidden_size*2) # Decode the hidden state of the last time step out = self.fc(out[:, -1, :]) return out model = BiRNN(input_size, hidden_size, num_layers, num_classes).to(device) # Loss and optimizer criterion = nn.CrossEntropyLoss() optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate) # Train the model total_step = len(train_loader) for epoch in range(num_epochs): for i, (images, labels) in enumerate(train_loader): images = images.reshape(-1, sequence_length, input_size).to(device) labels = labels.to(device) # Forward pass outputs = model(images) loss = criterion(outputs, labels) # Backward and optimize optimizer.zero_grad() loss.backward() optimizer.step() if (i+1) % 100 == 0: print ('Epoch [{}/{}], Step [{}/{}], Loss: {:.4f}' .format(epoch+1, num_epochs, i+1, total_step, loss.item())) # Test the model with torch.no_grad(): correct = 0 total = 0 for images, labels in test_loader: images = images.reshape(-1, sequence_length, input_size).to(device) labels = labels.to(device) outputs = model(images) _, predicted = torch.max(outputs.data, 1) total += labels.size(0) correct += (predicted == labels).sum().item() print('Test Accuracy of the model on the 10000 test images: {} %'.format(100 * correct / total)) # Save the model checkpoint torch.save(model.state_dict(), 'model.ckpt') ================================================ FILE: tutorials/02-intermediate/convolutional_neural_network/main.py ================================================ import torch import torch.nn as nn import torchvision import torchvision.transforms as transforms # Device configuration device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu') # Hyper parameters num_epochs = 5 num_classes = 10 batch_size = 100 learning_rate = 0.001 # MNIST dataset train_dataset = torchvision.datasets.MNIST(root='../../data/', train=True, transform=transforms.ToTensor(), download=True) test_dataset = torchvision.datasets.MNIST(root='../../data/', train=False, transform=transforms.ToTensor()) # Data loader train_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=batch_size, shuffle=True) test_loader = torch.utils.data.DataLoader(dataset=test_dataset, batch_size=batch_size, shuffle=False) # Convolutional neural network (two convolutional layers) class ConvNet(nn.Module): def __init__(self, num_classes=10): super(ConvNet, self).__init__() self.layer1 = nn.Sequential( nn.Conv2d(1, 16, kernel_size=5, stride=1, padding=2), nn.BatchNorm2d(16), nn.ReLU(), nn.MaxPool2d(kernel_size=2, stride=2)) self.layer2 = nn.Sequential( nn.Conv2d(16, 32, kernel_size=5, stride=1, padding=2), nn.BatchNorm2d(32), nn.ReLU(), nn.MaxPool2d(kernel_size=2, stride=2)) self.fc = nn.Linear(7*7*32, num_classes) def forward(self, x): out = self.layer1(x) out = self.layer2(out) out = out.reshape(out.size(0), -1) out = self.fc(out) return out model = ConvNet(num_classes).to(device) # Loss and optimizer criterion = nn.CrossEntropyLoss() optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate) # Train the model total_step = len(train_loader) for epoch in range(num_epochs): for i, (images, labels) in enumerate(train_loader): images = images.to(device) labels = labels.to(device) # Forward pass outputs = model(images) loss = criterion(outputs, labels) # Backward and optimize optimizer.zero_grad() loss.backward() optimizer.step() if (i+1) % 100 == 0: print ('Epoch [{}/{}], Step [{}/{}], Loss: {:.4f}' .format(epoch+1, num_epochs, i+1, total_step, loss.item())) # Test the model model.eval() # eval mode (batchnorm uses moving mean/variance instead of mini-batch mean/variance) with torch.no_grad(): correct = 0 total = 0 for images, labels in test_loader: images = images.to(device) labels = labels.to(device) outputs = model(images) _, predicted = torch.max(outputs.data, 1) total += labels.size(0) correct += (predicted == labels).sum().item() print('Test Accuracy of the model on the 10000 test images: {} %'.format(100 * correct / total)) # Save the model checkpoint torch.save(model.state_dict(), 'model.ckpt') ================================================ FILE: tutorials/02-intermediate/deep_residual_network/main.py ================================================ # ---------------------------------------------------------------------------- # # An implementation of https://arxiv.org/pdf/1512.03385.pdf # # See section 4.2 for the model architecture on CIFAR-10 # # Some part of the code was referenced from below # # https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py # # ---------------------------------------------------------------------------- # import torch import torch.nn as nn import torchvision import torchvision.transforms as transforms # Device configuration device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') # Hyper-parameters num_epochs = 80 batch_size = 100 learning_rate = 0.001 # Image preprocessing modules transform = transforms.Compose([ transforms.Pad(4), transforms.RandomHorizontalFlip(), transforms.RandomCrop(32), transforms.ToTensor()]) # CIFAR-10 dataset train_dataset = torchvision.datasets.CIFAR10(root='../../data/', train=True, transform=transform, download=True) test_dataset = torchvision.datasets.CIFAR10(root='../../data/', train=False, transform=transforms.ToTensor()) # Data loader train_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=batch_size, shuffle=True) test_loader = torch.utils.data.DataLoader(dataset=test_dataset, batch_size=batch_size, shuffle=False) # 3x3 convolution def conv3x3(in_channels, out_channels, stride=1): return nn.Conv2d(in_channels, out_channels, kernel_size=3, stride=stride, padding=1, bias=False) # Residual block class ResidualBlock(nn.Module): def __init__(self, in_channels, out_channels, stride=1, downsample=None): super(ResidualBlock, self).__init__() self.conv1 = conv3x3(in_channels, out_channels, stride) self.bn1 = nn.BatchNorm2d(out_channels) self.relu = nn.ReLU(inplace=True) self.conv2 = conv3x3(out_channels, out_channels) self.bn2 = nn.BatchNorm2d(out_channels) self.downsample = downsample def forward(self, x): residual = x out = self.conv1(x) out = self.bn1(out) out = self.relu(out) out = self.conv2(out) out = self.bn2(out) if self.downsample: residual = self.downsample(x) out += residual out = self.relu(out) return out # ResNet class ResNet(nn.Module): def __init__(self, block, layers, num_classes=10): super(ResNet, self).__init__() self.in_channels = 16 self.conv = conv3x3(3, 16) self.bn = nn.BatchNorm2d(16) self.relu = nn.ReLU(inplace=True) self.layer1 = self.make_layer(block, 16, layers[0]) self.layer2 = self.make_layer(block, 32, layers[1], 2) self.layer3 = self.make_layer(block, 64, layers[2], 2) self.avg_pool = nn.AvgPool2d(8) self.fc = nn.Linear(64, num_classes) def make_layer(self, block, out_channels, blocks, stride=1): downsample = None if (stride != 1) or (self.in_channels != out_channels): downsample = nn.Sequential( conv3x3(self.in_channels, out_channels, stride=stride), nn.BatchNorm2d(out_channels)) layers = [] layers.append(block(self.in_channels, out_channels, stride, downsample)) self.in_channels = out_channels for i in range(1, blocks): layers.append(block(out_channels, out_channels)) return nn.Sequential(*layers) def forward(self, x): out = self.conv(x) out = self.bn(out) out = self.relu(out) out = self.layer1(out) out = self.layer2(out) out = self.layer3(out) out = self.avg_pool(out) out = out.view(out.size(0), -1) out = self.fc(out) return out model = ResNet(ResidualBlock, [2, 2, 2]).to(device) # Loss and optimizer criterion = nn.CrossEntropyLoss() optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate) # For updating learning rate def update_lr(optimizer, lr): for param_group in optimizer.param_groups: param_group['lr'] = lr # Train the model total_step = len(train_loader) curr_lr = learning_rate for epoch in range(num_epochs): for i, (images, labels) in enumerate(train_loader): images = images.to(device) labels = labels.to(device) # Forward pass outputs = model(images) loss = criterion(outputs, labels) # Backward and optimize optimizer.zero_grad() loss.backward() optimizer.step() if (i+1) % 100 == 0: print ("Epoch [{}/{}], Step [{}/{}] Loss: {:.4f}" .format(epoch+1, num_epochs, i+1, total_step, loss.item())) # Decay learning rate if (epoch+1) % 20 == 0: curr_lr /= 3 update_lr(optimizer, curr_lr) # Test the model model.eval() with torch.no_grad(): correct = 0 total = 0 for images, labels in test_loader: images = images.to(device) labels = labels.to(device) outputs = model(images) _, predicted = torch.max(outputs.data, 1) total += labels.size(0) correct += (predicted == labels).sum().item() print('Accuracy of the model on the test images: {} %'.format(100 * correct / total)) # Save the model checkpoint torch.save(model.state_dict(), 'resnet.ckpt') ================================================ FILE: tutorials/02-intermediate/language_model/data_utils.py ================================================ import torch import os class Dictionary(object): def __init__(self): self.word2idx = {} self.idx2word = {} self.idx = 0 def add_word(self, word): if not word in self.word2idx: self.word2idx[word] = self.idx self.idx2word[self.idx] = word self.idx += 1 def __len__(self): return len(self.word2idx) class Corpus(object): def __init__(self): self.dictionary = Dictionary() def get_data(self, path, batch_size=20): # Add words to the dictionary with open(path, 'r') as f: tokens = 0 for line in f: words = line.split() + [''] tokens += len(words) for word in words: self.dictionary.add_word(word) # Tokenize the file content ids = torch.LongTensor(tokens) token = 0 with open(path, 'r') as f: for line in f: words = line.split() + [''] for word in words: ids[token] = self.dictionary.word2idx[word] token += 1 num_batches = ids.size(0) // batch_size ids = ids[:num_batches*batch_size] return ids.view(batch_size, -1) ================================================ FILE: tutorials/02-intermediate/language_model/main.py ================================================ # Some part of the code was referenced from below. # https://github.com/pytorch/examples/tree/master/word_language_model import torch import torch.nn as nn import numpy as np from torch.nn.utils import clip_grad_norm_ from data_utils import Dictionary, Corpus # Device configuration device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') # Hyper-parameters embed_size = 128 hidden_size = 1024 num_layers = 1 num_epochs = 5 num_samples = 1000 # number of words to be sampled batch_size = 20 seq_length = 30 learning_rate = 0.002 # Load "Penn Treebank" dataset corpus = Corpus() ids = corpus.get_data('data/train.txt', batch_size) vocab_size = len(corpus.dictionary) num_batches = ids.size(1) // seq_length # RNN based language model class RNNLM(nn.Module): def __init__(self, vocab_size, embed_size, hidden_size, num_layers): super(RNNLM, self).__init__() self.embed = nn.Embedding(vocab_size, embed_size) self.lstm = nn.LSTM(embed_size, hidden_size, num_layers, batch_first=True) self.linear = nn.Linear(hidden_size, vocab_size) def forward(self, x, h): # Embed word ids to vectors x = self.embed(x) # Forward propagate LSTM out, (h, c) = self.lstm(x, h) # Reshape output to (batch_size*sequence_length, hidden_size) out = out.reshape(out.size(0)*out.size(1), out.size(2)) # Decode hidden states of all time steps out = self.linear(out) return out, (h, c) model = RNNLM(vocab_size, embed_size, hidden_size, num_layers).to(device) # Loss and optimizer criterion = nn.CrossEntropyLoss() optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate) # Truncated backpropagation def detach(states): return [state.detach() for state in states] # Train the model for epoch in range(num_epochs): # Set initial hidden and cell states states = (torch.zeros(num_layers, batch_size, hidden_size).to(device), torch.zeros(num_layers, batch_size, hidden_size).to(device)) for i in range(0, ids.size(1) - seq_length, seq_length): # Get mini-batch inputs and targets inputs = ids[:, i:i+seq_length].to(device) targets = ids[:, (i+1):(i+1)+seq_length].to(device) # Forward pass states = detach(states) outputs, states = model(inputs, states) loss = criterion(outputs, targets.reshape(-1)) # Backward and optimize optimizer.zero_grad() loss.backward() clip_grad_norm_(model.parameters(), 0.5) optimizer.step() step = (i+1) // seq_length if step % 100 == 0: print ('Epoch [{}/{}], Step[{}/{}], Loss: {:.4f}, Perplexity: {:5.2f}' .format(epoch+1, num_epochs, step, num_batches, loss.item(), np.exp(loss.item()))) # Test the model with torch.no_grad(): with open('sample.txt', 'w') as f: # Set intial hidden ane cell states state = (torch.zeros(num_layers, 1, hidden_size).to(device), torch.zeros(num_layers, 1, hidden_size).to(device)) # Select one word id randomly prob = torch.ones(vocab_size) input = torch.multinomial(prob, num_samples=1).unsqueeze(1).to(device) for i in range(num_samples): # Forward propagate RNN output, state = model(input, state) # Sample a word id prob = output.exp() word_id = torch.multinomial(prob, num_samples=1).item() # Fill input with sampled word id for the next time step input.fill_(word_id) # File write word = corpus.dictionary.idx2word[word_id] word = '\n' if word == '' else word + ' ' f.write(word) if (i+1) % 100 == 0: print('Sampled [{}/{}] words and save to {}'.format(i+1, num_samples, 'sample.txt')) # Save the model checkpoints torch.save(model.state_dict(), 'model.ckpt') ================================================ FILE: tutorials/02-intermediate/recurrent_neural_network/main.py ================================================ import torch import torch.nn as nn import torchvision import torchvision.transforms as transforms # Device configuration device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') # Hyper-parameters sequence_length = 28 input_size = 28 hidden_size = 128 num_layers = 2 num_classes = 10 batch_size = 100 num_epochs = 2 learning_rate = 0.01 # MNIST dataset train_dataset = torchvision.datasets.MNIST(root='../../data/', train=True, transform=transforms.ToTensor(), download=True) test_dataset = torchvision.datasets.MNIST(root='../../data/', train=False, transform=transforms.ToTensor()) # Data loader train_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=batch_size, shuffle=True) test_loader = torch.utils.data.DataLoader(dataset=test_dataset, batch_size=batch_size, shuffle=False) # Recurrent neural network (many-to-one) class RNN(nn.Module): def __init__(self, input_size, hidden_size, num_layers, num_classes): super(RNN, self).__init__() self.hidden_size = hidden_size self.num_layers = num_layers self.lstm = nn.LSTM(input_size, hidden_size, num_layers, batch_first=True) self.fc = nn.Linear(hidden_size, num_classes) def forward(self, x): # Set initial hidden and cell states h0 = torch.zeros(self.num_layers, x.size(0), self.hidden_size).to(device) c0 = torch.zeros(self.num_layers, x.size(0), self.hidden_size).to(device) # Forward propagate LSTM out, _ = self.lstm(x, (h0, c0)) # out: tensor of shape (batch_size, seq_length, hidden_size) # Decode the hidden state of the last time step out = self.fc(out[:, -1, :]) return out model = RNN(input_size, hidden_size, num_layers, num_classes).to(device) # Loss and optimizer criterion = nn.CrossEntropyLoss() optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate) # Train the model total_step = len(train_loader) for epoch in range(num_epochs): for i, (images, labels) in enumerate(train_loader): images = images.reshape(-1, sequence_length, input_size).to(device) labels = labels.to(device) # Forward pass outputs = model(images) loss = criterion(outputs, labels) # Backward and optimize optimizer.zero_grad() loss.backward() optimizer.step() if (i+1) % 100 == 0: print ('Epoch [{}/{}], Step [{}/{}], Loss: {:.4f}' .format(epoch+1, num_epochs, i+1, total_step, loss.item())) # Test the model model.eval() with torch.no_grad(): correct = 0 total = 0 for images, labels in test_loader: images = images.reshape(-1, sequence_length, input_size).to(device) labels = labels.to(device) outputs = model(images) _, predicted = torch.max(outputs.data, 1) total += labels.size(0) correct += (predicted == labels).sum().item() print('Test Accuracy of the model on the 10000 test images: {} %'.format(100 * correct / total)) # Save the model checkpoint torch.save(model.state_dict(), 'model.ckpt') ================================================ FILE: tutorials/03-advanced/generative_adversarial_network/main.py ================================================ import os import torch import torchvision import torch.nn as nn from torchvision import transforms from torchvision.utils import save_image # Device configuration device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') # Hyper-parameters latent_size = 64 hidden_size = 256 image_size = 784 num_epochs = 200 batch_size = 100 sample_dir = 'samples' # Create a directory if not exists if not os.path.exists(sample_dir): os.makedirs(sample_dir) # Image processing # transform = transforms.Compose([ # transforms.ToTensor(), # transforms.Normalize(mean=(0.5, 0.5, 0.5), # 3 for RGB channels # std=(0.5, 0.5, 0.5))]) transform = transforms.Compose([ transforms.ToTensor(), transforms.Normalize(mean=[0.5], # 1 for greyscale channels std=[0.5])]) # MNIST dataset mnist = torchvision.datasets.MNIST(root='../../data/', train=True, transform=transform, download=True) # Data loader data_loader = torch.utils.data.DataLoader(dataset=mnist, batch_size=batch_size, shuffle=True) # Discriminator D = nn.Sequential( nn.Linear(image_size, hidden_size), nn.LeakyReLU(0.2), nn.Linear(hidden_size, hidden_size), nn.LeakyReLU(0.2), nn.Linear(hidden_size, 1), nn.Sigmoid()) # Generator G = nn.Sequential( nn.Linear(latent_size, hidden_size), nn.ReLU(), nn.Linear(hidden_size, hidden_size), nn.ReLU(), nn.Linear(hidden_size, image_size), nn.Tanh()) # Device setting D = D.to(device) G = G.to(device) # Binary cross entropy loss and optimizer criterion = nn.BCELoss() d_optimizer = torch.optim.Adam(D.parameters(), lr=0.0002) g_optimizer = torch.optim.Adam(G.parameters(), lr=0.0002) def denorm(x): out = (x + 1) / 2 return out.clamp(0, 1) def reset_grad(): d_optimizer.zero_grad() g_optimizer.zero_grad() # Start training total_step = len(data_loader) for epoch in range(num_epochs): for i, (images, _) in enumerate(data_loader): images = images.reshape(batch_size, -1).to(device) # Create the labels which are later used as input for the BCE loss real_labels = torch.ones(batch_size, 1).to(device) fake_labels = torch.zeros(batch_size, 1).to(device) # ================================================================== # # Train the discriminator # # ================================================================== # # Compute BCE_Loss using real images where BCE_Loss(x, y): - y * log(D(x)) - (1-y) * log(1 - D(x)) # Second term of the loss is always zero since real_labels == 1 outputs = D(images) d_loss_real = criterion(outputs, real_labels) real_score = outputs # Compute BCELoss using fake images # First term of the loss is always zero since fake_labels == 0 z = torch.randn(batch_size, latent_size).to(device) fake_images = G(z) outputs = D(fake_images) d_loss_fake = criterion(outputs, fake_labels) fake_score = outputs # Backprop and optimize d_loss = d_loss_real + d_loss_fake reset_grad() d_loss.backward() d_optimizer.step() # ================================================================== # # Train the generator # # ================================================================== # # Compute loss with fake images z = torch.randn(batch_size, latent_size).to(device) fake_images = G(z) outputs = D(fake_images) # We train G to maximize log(D(G(z)) instead of minimizing log(1-D(G(z))) # For the reason, see the last paragraph of section 3. https://arxiv.org/pdf/1406.2661.pdf g_loss = criterion(outputs, real_labels) # Backprop and optimize reset_grad() g_loss.backward() g_optimizer.step() if (i+1) % 200 == 0: print('Epoch [{}/{}], Step [{}/{}], d_loss: {:.4f}, g_loss: {:.4f}, D(x): {:.2f}, D(G(z)): {:.2f}' .format(epoch, num_epochs, i+1, total_step, d_loss.item(), g_loss.item(), real_score.mean().item(), fake_score.mean().item())) # Save real images if (epoch+1) == 1: images = images.reshape(images.size(0), 1, 28, 28) save_image(denorm(images), os.path.join(sample_dir, 'real_images.png')) # Save sampled images fake_images = fake_images.reshape(fake_images.size(0), 1, 28, 28) save_image(denorm(fake_images), os.path.join(sample_dir, 'fake_images-{}.png'.format(epoch+1))) # Save the model checkpoints torch.save(G.state_dict(), 'G.ckpt') torch.save(D.state_dict(), 'D.ckpt') ================================================ FILE: tutorials/03-advanced/image_captioning/README.md ================================================ # Image Captioning The goal of image captioning is to convert a given input image into a natural language description. The encoder-decoder framework is widely used for this task. The image encoder is a convolutional neural network (CNN). In this tutorial, we used [resnet-152](https://arxiv.org/abs/1512.03385) model pretrained on the [ILSVRC-2012-CLS](http://www.image-net.org/challenges/LSVRC/2012/) image classification dataset. The decoder is a long short-term memory (LSTM) network. ![alt text](png/model.png) #### Training phase For the encoder part, the pretrained CNN extracts the feature vector from a given input image. The feature vector is linearly transformed to have the same dimension as the input dimension of the LSTM network. For the decoder part, source and target texts are predefined. For example, if the image description is **"Giraffes standing next to each other"**, the source sequence is a list containing **['\', 'Giraffes', 'standing', 'next', 'to', 'each', 'other']** and the target sequence is a list containing **['Giraffes', 'standing', 'next', 'to', 'each', 'other', '\']**. Using these source and target sequences and the feature vector, the LSTM decoder is trained as a language model conditioned on the feature vector. #### Test phase In the test phase, the encoder part is almost same as the training phase. The only difference is that batchnorm layer uses moving average and variance instead of mini-batch statistics. This can be easily implemented using [encoder.eval()](https://github.com/yunjey/pytorch-tutorial/blob/master/tutorials/03-advanced/image_captioning/sample.py#L37). For the decoder part, there is a significant difference between the training phase and the test phase. In the test phase, the LSTM decoder can't see the image description. To deal with this problem, the LSTM decoder feeds back the previosly generated word to the next input. This can be implemented using a [for-loop](https://github.com/yunjey/pytorch-tutorial/blob/master/tutorials/03-advanced/image_captioning/model.py#L48). ## Usage #### 1. Clone the repositories ```bash git clone https://github.com/pdollar/coco.git cd coco/PythonAPI/ make python setup.py build python setup.py install cd ../../ git clone https://github.com/yunjey/pytorch-tutorial.git cd pytorch-tutorial/tutorials/03-advanced/image_captioning/ ``` #### 2. Download the dataset ```bash pip install -r requirements.txt chmod +x download.sh ./download.sh ``` #### 3. Preprocessing ```bash python build_vocab.py python resize.py ``` #### 4. Train the model ```bash python train.py ``` #### 5. Test the model ```bash python sample.py --image='png/example.png' ```
## Pretrained model If you do not want to train the model from scratch, you can use a pretrained model. You can download the pretrained model [here](https://www.dropbox.com/s/ne0ixz5d58ccbbz/pretrained_model.zip?dl=0) and the vocabulary file [here](https://www.dropbox.com/s/26adb7y9m98uisa/vocap.zip?dl=0). You should extract pretrained_model.zip to `./models/` and vocab.pkl to `./data/` using `unzip` command. ================================================ FILE: tutorials/03-advanced/image_captioning/build_vocab.py ================================================ import nltk import pickle import argparse from collections import Counter from pycocotools.coco import COCO class Vocabulary(object): """Simple vocabulary wrapper.""" def __init__(self): self.word2idx = {} self.idx2word = {} self.idx = 0 def add_word(self, word): if not word in self.word2idx: self.word2idx[word] = self.idx self.idx2word[self.idx] = word self.idx += 1 def __call__(self, word): if not word in self.word2idx: return self.word2idx[''] return self.word2idx[word] def __len__(self): return len(self.word2idx) def build_vocab(json, threshold): """Build a simple vocabulary wrapper.""" coco = COCO(json) counter = Counter() ids = coco.anns.keys() for i, id in enumerate(ids): caption = str(coco.anns[id]['caption']) tokens = nltk.tokenize.word_tokenize(caption.lower()) counter.update(tokens) if (i+1) % 1000 == 0: print("[{}/{}] Tokenized the captions.".format(i+1, len(ids))) # If the word frequency is less than 'threshold', then the word is discarded. words = [word for word, cnt in counter.items() if cnt >= threshold] # Create a vocab wrapper and add some special tokens. vocab = Vocabulary() vocab.add_word('') vocab.add_word('') vocab.add_word('') vocab.add_word('') # Add the words to the vocabulary. for i, word in enumerate(words): vocab.add_word(word) return vocab def main(args): vocab = build_vocab(json=args.caption_path, threshold=args.threshold) vocab_path = args.vocab_path with open(vocab_path, 'wb') as f: pickle.dump(vocab, f) print("Total vocabulary size: {}".format(len(vocab))) print("Saved the vocabulary wrapper to '{}'".format(vocab_path)) if __name__ == '__main__': parser = argparse.ArgumentParser() parser.add_argument('--caption_path', type=str, default='data/annotations/captions_train2014.json', help='path for train annotation file') parser.add_argument('--vocab_path', type=str, default='./data/vocab.pkl', help='path for saving vocabulary wrapper') parser.add_argument('--threshold', type=int, default=4, help='minimum word count threshold') args = parser.parse_args() main(args) ================================================ FILE: tutorials/03-advanced/image_captioning/data_loader.py ================================================ import torch import torchvision.transforms as transforms import torch.utils.data as data import os import pickle import numpy as np import nltk from PIL import Image from build_vocab import Vocabulary from pycocotools.coco import COCO class CocoDataset(data.Dataset): """COCO Custom Dataset compatible with torch.utils.data.DataLoader.""" def __init__(self, root, json, vocab, transform=None): """Set the path for images, captions and vocabulary wrapper. Args: root: image directory. json: coco annotation file path. vocab: vocabulary wrapper. transform: image transformer. """ self.root = root self.coco = COCO(json) self.ids = list(self.coco.anns.keys()) self.vocab = vocab self.transform = transform def __getitem__(self, index): """Returns one data pair (image and caption).""" coco = self.coco vocab = self.vocab ann_id = self.ids[index] caption = coco.anns[ann_id]['caption'] img_id = coco.anns[ann_id]['image_id'] path = coco.loadImgs(img_id)[0]['file_name'] image = Image.open(os.path.join(self.root, path)).convert('RGB') if self.transform is not None: image = self.transform(image) # Convert caption (string) to word ids. tokens = nltk.tokenize.word_tokenize(str(caption).lower()) caption = [] caption.append(vocab('')) caption.extend([vocab(token) for token in tokens]) caption.append(vocab('')) target = torch.Tensor(caption) return image, target def __len__(self): return len(self.ids) def collate_fn(data): """Creates mini-batch tensors from the list of tuples (image, caption). We should build custom collate_fn rather than using default collate_fn, because merging caption (including padding) is not supported in default. Args: data: list of tuple (image, caption). - image: torch tensor of shape (3, 256, 256). - caption: torch tensor of shape (?); variable length. Returns: images: torch tensor of shape (batch_size, 3, 256, 256). targets: torch tensor of shape (batch_size, padded_length). lengths: list; valid length for each padded caption. """ # Sort a data list by caption length (descending order). data.sort(key=lambda x: len(x[1]), reverse=True) images, captions = zip(*data) # Merge images (from tuple of 3D tensor to 4D tensor). images = torch.stack(images, 0) # Merge captions (from tuple of 1D tensor to 2D tensor). lengths = [len(cap) for cap in captions] targets = torch.zeros(len(captions), max(lengths)).long() for i, cap in enumerate(captions): end = lengths[i] targets[i, :end] = cap[:end] return images, targets, lengths def get_loader(root, json, vocab, transform, batch_size, shuffle, num_workers): """Returns torch.utils.data.DataLoader for custom coco dataset.""" # COCO caption dataset coco = CocoDataset(root=root, json=json, vocab=vocab, transform=transform) # Data loader for COCO dataset # This will return (images, captions, lengths) for each iteration. # images: a tensor of shape (batch_size, 3, 224, 224). # captions: a tensor of shape (batch_size, padded_length). # lengths: a list indicating valid length for each caption. length is (batch_size). data_loader = torch.utils.data.DataLoader(dataset=coco, batch_size=batch_size, shuffle=shuffle, num_workers=num_workers, collate_fn=collate_fn) return data_loader ================================================ FILE: tutorials/03-advanced/image_captioning/download.sh ================================================ mkdir data wget http://msvocds.blob.core.windows.net/annotations-1-0-3/captions_train-val2014.zip -P ./data/ wget http://images.cocodataset.org/zips/train2014.zip -P ./data/ wget http://images.cocodataset.org/zips/val2014.zip -P ./data/ unzip ./data/captions_train-val2014.zip -d ./data/ rm ./data/captions_train-val2014.zip unzip ./data/train2014.zip -d ./data/ rm ./data/train2014.zip unzip ./data/val2014.zip -d ./data/ rm ./data/val2014.zip ================================================ FILE: tutorials/03-advanced/image_captioning/model.py ================================================ import torch import torch.nn as nn import torchvision.models as models from torch.nn.utils.rnn import pack_padded_sequence class EncoderCNN(nn.Module): def __init__(self, embed_size): """Load the pretrained ResNet-152 and replace top fc layer.""" super(EncoderCNN, self).__init__() resnet = models.resnet152(pretrained=True) modules = list(resnet.children())[:-1] # delete the last fc layer. self.resnet = nn.Sequential(*modules) self.linear = nn.Linear(resnet.fc.in_features, embed_size) self.bn = nn.BatchNorm1d(embed_size, momentum=0.01) def forward(self, images): """Extract feature vectors from input images.""" with torch.no_grad(): features = self.resnet(images) features = features.reshape(features.size(0), -1) features = self.bn(self.linear(features)) return features class DecoderRNN(nn.Module): def __init__(self, embed_size, hidden_size, vocab_size, num_layers, max_seq_length=20): """Set the hyper-parameters and build the layers.""" super(DecoderRNN, self).__init__() self.embed = nn.Embedding(vocab_size, embed_size) self.lstm = nn.LSTM(embed_size, hidden_size, num_layers, batch_first=True) self.linear = nn.Linear(hidden_size, vocab_size) self.max_seg_length = max_seq_length def forward(self, features, captions, lengths): """Decode image feature vectors and generates captions.""" embeddings = self.embed(captions) embeddings = torch.cat((features.unsqueeze(1), embeddings), 1) packed = pack_padded_sequence(embeddings, lengths, batch_first=True) hiddens, _ = self.lstm(packed) outputs = self.linear(hiddens[0]) return outputs def sample(self, features, states=None): """Generate captions for given image features using greedy search.""" sampled_ids = [] inputs = features.unsqueeze(1) for i in range(self.max_seg_length): hiddens, states = self.lstm(inputs, states) # hiddens: (batch_size, 1, hidden_size) outputs = self.linear(hiddens.squeeze(1)) # outputs: (batch_size, vocab_size) _, predicted = outputs.max(1) # predicted: (batch_size) sampled_ids.append(predicted) inputs = self.embed(predicted) # inputs: (batch_size, embed_size) inputs = inputs.unsqueeze(1) # inputs: (batch_size, 1, embed_size) sampled_ids = torch.stack(sampled_ids, 1) # sampled_ids: (batch_size, max_seq_length) return sampled_ids ================================================ FILE: tutorials/03-advanced/image_captioning/requirements.txt ================================================ matplotlib nltk numpy Pillow argparse ================================================ FILE: tutorials/03-advanced/image_captioning/resize.py ================================================ import argparse import os from PIL import Image def resize_image(image, size): """Resize an image to the given size.""" return image.resize(size, Image.ANTIALIAS) def resize_images(image_dir, output_dir, size): """Resize the images in 'image_dir' and save into 'output_dir'.""" if not os.path.exists(output_dir): os.makedirs(output_dir) images = os.listdir(image_dir) num_images = len(images) for i, image in enumerate(images): with open(os.path.join(image_dir, image), 'r+b') as f: with Image.open(f) as img: img = resize_image(img, size) img.save(os.path.join(output_dir, image), img.format) if (i+1) % 100 == 0: print ("[{}/{}] Resized the images and saved into '{}'." .format(i+1, num_images, output_dir)) def main(args): image_dir = args.image_dir output_dir = args.output_dir image_size = [args.image_size, args.image_size] resize_images(image_dir, output_dir, image_size) if __name__ == '__main__': parser = argparse.ArgumentParser() parser.add_argument('--image_dir', type=str, default='./data/train2014/', help='directory for train images') parser.add_argument('--output_dir', type=str, default='./data/resized2014/', help='directory for saving resized images') parser.add_argument('--image_size', type=int, default=256, help='size for image after processing') args = parser.parse_args() main(args) ================================================ FILE: tutorials/03-advanced/image_captioning/sample.py ================================================ import torch import matplotlib.pyplot as plt import numpy as np import argparse import pickle import os from torchvision import transforms from build_vocab import Vocabulary from model import EncoderCNN, DecoderRNN from PIL import Image # Device configuration device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') def load_image(image_path, transform=None): image = Image.open(image_path).convert('RGB') image = image.resize([224, 224], Image.LANCZOS) if transform is not None: image = transform(image).unsqueeze(0) return image def main(args): # Image preprocessing transform = transforms.Compose([ transforms.ToTensor(), transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))]) # Load vocabulary wrapper with open(args.vocab_path, 'rb') as f: vocab = pickle.load(f) # Build models encoder = EncoderCNN(args.embed_size).eval() # eval mode (batchnorm uses moving mean/variance) decoder = DecoderRNN(args.embed_size, args.hidden_size, len(vocab), args.num_layers) encoder = encoder.to(device) decoder = decoder.to(device) # Load the trained model parameters encoder.load_state_dict(torch.load(args.encoder_path)) decoder.load_state_dict(torch.load(args.decoder_path)) # Prepare an image image = load_image(args.image, transform) image_tensor = image.to(device) # Generate an caption from the image feature = encoder(image_tensor) sampled_ids = decoder.sample(feature) sampled_ids = sampled_ids[0].cpu().numpy() # (1, max_seq_length) -> (max_seq_length) # Convert word_ids to words sampled_caption = [] for word_id in sampled_ids: word = vocab.idx2word[word_id] sampled_caption.append(word) if word == '': break sentence = ' '.join(sampled_caption) # Print out the image and the generated caption print (sentence) image = Image.open(args.image) plt.imshow(np.asarray(image)) if __name__ == '__main__': parser = argparse.ArgumentParser() parser.add_argument('--image', type=str, required=True, help='input image for generating caption') parser.add_argument('--encoder_path', type=str, default='models/encoder-5-3000.pkl', help='path for trained encoder') parser.add_argument('--decoder_path', type=str, default='models/decoder-5-3000.pkl', help='path for trained decoder') parser.add_argument('--vocab_path', type=str, default='data/vocab.pkl', help='path for vocabulary wrapper') # Model parameters (should be same as paramters in train.py) parser.add_argument('--embed_size', type=int , default=256, help='dimension of word embedding vectors') parser.add_argument('--hidden_size', type=int , default=512, help='dimension of lstm hidden states') parser.add_argument('--num_layers', type=int , default=1, help='number of layers in lstm') args = parser.parse_args() main(args) ================================================ FILE: tutorials/03-advanced/image_captioning/train.py ================================================ import argparse import torch import torch.nn as nn import numpy as np import os import pickle from data_loader import get_loader from build_vocab import Vocabulary from model import EncoderCNN, DecoderRNN from torch.nn.utils.rnn import pack_padded_sequence from torchvision import transforms # Device configuration device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') def main(args): # Create model directory if not os.path.exists(args.model_path): os.makedirs(args.model_path) # Image preprocessing, normalization for the pretrained resnet transform = transforms.Compose([ transforms.RandomCrop(args.crop_size), transforms.RandomHorizontalFlip(), transforms.ToTensor(), transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))]) # Load vocabulary wrapper with open(args.vocab_path, 'rb') as f: vocab = pickle.load(f) # Build data loader data_loader = get_loader(args.image_dir, args.caption_path, vocab, transform, args.batch_size, shuffle=True, num_workers=args.num_workers) # Build the models encoder = EncoderCNN(args.embed_size).to(device) decoder = DecoderRNN(args.embed_size, args.hidden_size, len(vocab), args.num_layers).to(device) # Loss and optimizer criterion = nn.CrossEntropyLoss() params = list(decoder.parameters()) + list(encoder.linear.parameters()) + list(encoder.bn.parameters()) optimizer = torch.optim.Adam(params, lr=args.learning_rate) # Train the models total_step = len(data_loader) for epoch in range(args.num_epochs): for i, (images, captions, lengths) in enumerate(data_loader): # Set mini-batch dataset images = images.to(device) captions = captions.to(device) targets = pack_padded_sequence(captions, lengths, batch_first=True)[0] # Forward, backward and optimize features = encoder(images) outputs = decoder(features, captions, lengths) loss = criterion(outputs, targets) decoder.zero_grad() encoder.zero_grad() loss.backward() optimizer.step() # Print log info if i % args.log_step == 0: print('Epoch [{}/{}], Step [{}/{}], Loss: {:.4f}, Perplexity: {:5.4f}' .format(epoch, args.num_epochs, i, total_step, loss.item(), np.exp(loss.item()))) # Save the model checkpoints if (i+1) % args.save_step == 0: torch.save(decoder.state_dict(), os.path.join( args.model_path, 'decoder-{}-{}.ckpt'.format(epoch+1, i+1))) torch.save(encoder.state_dict(), os.path.join( args.model_path, 'encoder-{}-{}.ckpt'.format(epoch+1, i+1))) if __name__ == '__main__': parser = argparse.ArgumentParser() parser.add_argument('--model_path', type=str, default='models/' , help='path for saving trained models') parser.add_argument('--crop_size', type=int, default=224 , help='size for randomly cropping images') parser.add_argument('--vocab_path', type=str, default='data/vocab.pkl', help='path for vocabulary wrapper') parser.add_argument('--image_dir', type=str, default='data/resized2014', help='directory for resized images') parser.add_argument('--caption_path', type=str, default='data/annotations/captions_train2014.json', help='path for train annotation json file') parser.add_argument('--log_step', type=int , default=10, help='step size for prining log info') parser.add_argument('--save_step', type=int , default=1000, help='step size for saving trained models') # Model parameters parser.add_argument('--embed_size', type=int , default=256, help='dimension of word embedding vectors') parser.add_argument('--hidden_size', type=int , default=512, help='dimension of lstm hidden states') parser.add_argument('--num_layers', type=int , default=1, help='number of layers in lstm') parser.add_argument('--num_epochs', type=int, default=5) parser.add_argument('--batch_size', type=int, default=128) parser.add_argument('--num_workers', type=int, default=2) parser.add_argument('--learning_rate', type=float, default=0.001) args = parser.parse_args() print(args) main(args) ================================================ FILE: tutorials/03-advanced/neural_style_transfer/README.md ================================================ # Neural Style Transfer [Neural style transfer](https://arxiv.org/abs/1508.06576) is an algorithm that combines the content of one image with the style of another image using CNN. Given a content image and a style image, the goal is to generate a target image that minimizes the content difference with the content image and the style difference with the style image.

#### Content loss To minimize the content difference, we forward propagate the content image and the target image to pretrained [VGGNet](https://arxiv.org/abs/1409.1556) respectively, and extract feature maps from multiple convolutional layers. Then, the target image is updated to minimize the [mean-squared error](https://github.com/yunjey/pytorch-tutorial/blob/master/tutorials/03-advanced/neural_style_transfer/main.py#L81-L82) between the feature maps of the content image and its feature maps. #### Style loss As in computing the content loss, we forward propagate the style image and the target image to the VGGNet and extract convolutional feature maps. To generate a texture that matches the style of the style image, we update the target image by minimizing the mean-squared error between the Gram matrix of the style image and the Gram matrix of the target image (feature correlation minimization). See [here](https://github.com/yunjey/pytorch-tutorial/blob/master/tutorials/03-advanced/neural_style_transfer/main.py#L84-L94) for how to compute the style loss.
## Usage ```bash $ pip install -r requirements.txt $ python main.py --content='png/content.png' --style='png/style.png' ```
## Results The following is the result of applying variaous styles of artwork to Anne Hathaway's photograph. ![alt text](png/neural_style.png) ================================================ FILE: tutorials/03-advanced/neural_style_transfer/main.py ================================================ from __future__ import division from torchvision import models from torchvision import transforms from PIL import Image import argparse import torch import torchvision import torch.nn as nn import numpy as np # Device configuration device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') def load_image(image_path, transform=None, max_size=None, shape=None): """Load an image and convert it to a torch tensor.""" image = Image.open(image_path) if max_size: scale = max_size / max(image.size) size = np.array(image.size) * scale image = image.resize(size.astype(int), Image.ANTIALIAS) if shape: image = image.resize(shape, Image.LANCZOS) if transform: image = transform(image).unsqueeze(0) return image.to(device) class VGGNet(nn.Module): def __init__(self): """Select conv1_1 ~ conv5_1 activation maps.""" super(VGGNet, self).__init__() self.select = ['0', '5', '10', '19', '28'] self.vgg = models.vgg19(pretrained=True).features def forward(self, x): """Extract multiple convolutional feature maps.""" features = [] for name, layer in self.vgg._modules.items(): x = layer(x) if name in self.select: features.append(x) return features def main(config): # Image preprocessing # VGGNet was trained on ImageNet where images are normalized by mean=[0.485, 0.456, 0.406] and std=[0.229, 0.224, 0.225]. # We use the same normalization statistics here. transform = transforms.Compose([ transforms.ToTensor(), transforms.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225))]) # Load content and style images # Make the style image same size as the content image content = load_image(config.content, transform, max_size=config.max_size) style = load_image(config.style, transform, shape=[content.size(2), content.size(3)]) # Initialize a target image with the content image target = content.clone().requires_grad_(True) optimizer = torch.optim.Adam([target], lr=config.lr, betas=[0.5, 0.999]) vgg = VGGNet().to(device).eval() for step in range(config.total_step): # Extract multiple(5) conv feature vectors target_features = vgg(target) content_features = vgg(content) style_features = vgg(style) style_loss = 0 content_loss = 0 for f1, f2, f3 in zip(target_features, content_features, style_features): # Compute content loss with target and content images content_loss += torch.mean((f1 - f2)**2) # Reshape convolutional feature maps _, c, h, w = f1.size() f1 = f1.view(c, h * w) f3 = f3.view(c, h * w) # Compute gram matrix f1 = torch.mm(f1, f1.t()) f3 = torch.mm(f3, f3.t()) # Compute style loss with target and style images style_loss += torch.mean((f1 - f3)**2) / (c * h * w) # Compute total loss, backprop and optimize loss = content_loss + config.style_weight * style_loss optimizer.zero_grad() loss.backward() optimizer.step() if (step+1) % config.log_step == 0: print ('Step [{}/{}], Content Loss: {:.4f}, Style Loss: {:.4f}' .format(step+1, config.total_step, content_loss.item(), style_loss.item())) if (step+1) % config.sample_step == 0: # Save the generated image denorm = transforms.Normalize((-2.12, -2.04, -1.80), (4.37, 4.46, 4.44)) img = target.clone().squeeze() img = denorm(img).clamp_(0, 1) torchvision.utils.save_image(img, 'output-{}.png'.format(step+1)) if __name__ == "__main__": parser = argparse.ArgumentParser() parser.add_argument('--content', type=str, default='png/content.png') parser.add_argument('--style', type=str, default='png/style.png') parser.add_argument('--max_size', type=int, default=400) parser.add_argument('--total_step', type=int, default=2000) parser.add_argument('--log_step', type=int, default=10) parser.add_argument('--sample_step', type=int, default=500) parser.add_argument('--style_weight', type=float, default=100) parser.add_argument('--lr', type=float, default=0.003) config = parser.parse_args() print(config) main(config) ================================================ FILE: tutorials/03-advanced/neural_style_transfer/requirements.txt ================================================ argparse torch torchvision Pillow ================================================ FILE: tutorials/03-advanced/variational_autoencoder/main.py ================================================ import os import torch import torch.nn as nn import torch.nn.functional as F import torchvision from torchvision import transforms from torchvision.utils import save_image # Device configuration device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') # Create a directory if not exists sample_dir = 'samples' if not os.path.exists(sample_dir): os.makedirs(sample_dir) # Hyper-parameters image_size = 784 h_dim = 400 z_dim = 20 num_epochs = 15 batch_size = 128 learning_rate = 1e-3 # MNIST dataset dataset = torchvision.datasets.MNIST(root='../../data', train=True, transform=transforms.ToTensor(), download=True) # Data loader data_loader = torch.utils.data.DataLoader(dataset=dataset, batch_size=batch_size, shuffle=True) # VAE model class VAE(nn.Module): def __init__(self, image_size=784, h_dim=400, z_dim=20): super(VAE, self).__init__() self.fc1 = nn.Linear(image_size, h_dim) self.fc2 = nn.Linear(h_dim, z_dim) self.fc3 = nn.Linear(h_dim, z_dim) self.fc4 = nn.Linear(z_dim, h_dim) self.fc5 = nn.Linear(h_dim, image_size) def encode(self, x): h = F.relu(self.fc1(x)) return self.fc2(h), self.fc3(h) def reparameterize(self, mu, log_var): std = torch.exp(log_var/2) eps = torch.randn_like(std) return mu + eps * std def decode(self, z): h = F.relu(self.fc4(z)) return F.sigmoid(self.fc5(h)) def forward(self, x): mu, log_var = self.encode(x) z = self.reparameterize(mu, log_var) x_reconst = self.decode(z) return x_reconst, mu, log_var model = VAE().to(device) optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate) # Start training for epoch in range(num_epochs): for i, (x, _) in enumerate(data_loader): # Forward pass x = x.to(device).view(-1, image_size) x_reconst, mu, log_var = model(x) # Compute reconstruction loss and kl divergence # For KL divergence, see Appendix B in VAE paper or http://yunjey47.tistory.com/43 reconst_loss = F.binary_cross_entropy(x_reconst, x, size_average=False) kl_div = - 0.5 * torch.sum(1 + log_var - mu.pow(2) - log_var.exp()) # Backprop and optimize loss = reconst_loss + kl_div optimizer.zero_grad() loss.backward() optimizer.step() if (i+1) % 10 == 0: print ("Epoch[{}/{}], Step [{}/{}], Reconst Loss: {:.4f}, KL Div: {:.4f}" .format(epoch+1, num_epochs, i+1, len(data_loader), reconst_loss.item(), kl_div.item())) with torch.no_grad(): # Save the sampled images z = torch.randn(batch_size, z_dim).to(device) out = model.decode(z).view(-1, 1, 28, 28) save_image(out, os.path.join(sample_dir, 'sampled-{}.png'.format(epoch+1))) # Save the reconstructed images out, _, _ = model(x) x_concat = torch.cat([x.view(-1, 1, 28, 28), out.view(-1, 1, 28, 28)], dim=3) save_image(x_concat, os.path.join(sample_dir, 'reconst-{}.png'.format(epoch+1))) ================================================ FILE: tutorials/04-utils/tensorboard/README.md ================================================ # TensorBoard in PyTorch In this tutorial, we implement a MNIST classifier using a simple neural network and visualize the training process using [TensorBoard](https://www.tensorflow.org/get_started/summaries_and_tensorboard). In training phase, we plot the loss and accuracy functions through `scalar_summary` and visualize the training images through `image_summary`. In addition, we visualize the weight and gradient values of the parameters of the neural network using `histogram_summary`. PyTorch code for handling these summary functions can be found [here](https://github.com/yunjey/pytorch-tutorial/blob/master/tutorials/04-utils/tensorboard/main.py#L81-L97). ![alt text](gif/tensorboard.gif)
## Usage #### 1. Install the dependencies ```bash $ pip install -r requirements.txt ``` #### 2. Train the model ```bash $ python main.py ``` #### 3. Open the TensorBoard To run the TensorBoard, open a new terminal and run the command below. Then, open http://localhost:6006/ on your web browser. ```bash $ tensorboard --logdir='./logs' --port=6006 ``` ================================================ FILE: tutorials/04-utils/tensorboard/logger.py ================================================ # Code referenced from https://gist.github.com/gyglim/1f8dfb1b5c82627ae3efcfbbadb9f514 import tensorflow as tf import numpy as np import scipy.misc try: from StringIO import StringIO # Python 2.7 except ImportError: from io import BytesIO # Python 3.x class Logger(object): def __init__(self, log_dir): """Create a summary writer logging to log_dir.""" self.writer = tf.summary.FileWriter(log_dir) def scalar_summary(self, tag, value, step): """Log a scalar variable.""" summary = tf.Summary(value=[tf.Summary.Value(tag=tag, simple_value=value)]) self.writer.add_summary(summary, step) def image_summary(self, tag, images, step): """Log a list of images.""" img_summaries = [] for i, img in enumerate(images): # Write the image to a string try: s = StringIO() except: s = BytesIO() scipy.misc.toimage(img).save(s, format="png") # Create an Image object img_sum = tf.Summary.Image(encoded_image_string=s.getvalue(), height=img.shape[0], width=img.shape[1]) # Create a Summary value img_summaries.append(tf.Summary.Value(tag='%s/%d' % (tag, i), image=img_sum)) # Create and write Summary summary = tf.Summary(value=img_summaries) self.writer.add_summary(summary, step) def histo_summary(self, tag, values, step, bins=1000): """Log a histogram of the tensor of values.""" # Create a histogram using numpy counts, bin_edges = np.histogram(values, bins=bins) # Fill the fields of the histogram proto hist = tf.HistogramProto() hist.min = float(np.min(values)) hist.max = float(np.max(values)) hist.num = int(np.prod(values.shape)) hist.sum = float(np.sum(values)) hist.sum_squares = float(np.sum(values**2)) # Drop the start of the first bin bin_edges = bin_edges[1:] # Add bin edges and counts for edge in bin_edges: hist.bucket_limit.append(edge) for c in counts: hist.bucket.append(c) # Create and write Summary summary = tf.Summary(value=[tf.Summary.Value(tag=tag, histo=hist)]) self.writer.add_summary(summary, step) self.writer.flush() ================================================ FILE: tutorials/04-utils/tensorboard/main.py ================================================ import torch import torch.nn as nn import torchvision from torchvision import transforms from logger import Logger # Device configuration device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') # MNIST dataset dataset = torchvision.datasets.MNIST(root='../../data', train=True, transform=transforms.ToTensor(), download=True) # Data loader data_loader = torch.utils.data.DataLoader(dataset=dataset, batch_size=100, shuffle=True) # Fully connected neural network with one hidden layer class NeuralNet(nn.Module): def __init__(self, input_size=784, hidden_size=500, num_classes=10): super(NeuralNet, self).__init__() self.fc1 = nn.Linear(input_size, hidden_size) self.relu = nn.ReLU() self.fc2 = nn.Linear(hidden_size, num_classes) def forward(self, x): out = self.fc1(x) out = self.relu(out) out = self.fc2(out) return out model = NeuralNet().to(device) logger = Logger('./logs') # Loss and optimizer criterion = nn.CrossEntropyLoss() optimizer = torch.optim.Adam(model.parameters(), lr=0.00001) data_iter = iter(data_loader) iter_per_epoch = len(data_loader) total_step = 50000 # Start training for step in range(total_step): # Reset the data_iter if (step+1) % iter_per_epoch == 0: data_iter = iter(data_loader) # Fetch images and labels images, labels = next(data_iter) images, labels = images.view(images.size(0), -1).to(device), labels.to(device) # Forward pass outputs = model(images) loss = criterion(outputs, labels) # Backward and optimize optimizer.zero_grad() loss.backward() optimizer.step() # Compute accuracy _, argmax = torch.max(outputs, 1) accuracy = (labels == argmax.squeeze()).float().mean() if (step+1) % 100 == 0: print ('Step [{}/{}], Loss: {:.4f}, Acc: {:.2f}' .format(step+1, total_step, loss.item(), accuracy.item())) # ================================================================== # # Tensorboard Logging # # ================================================================== # # 1. Log scalar values (scalar summary) info = { 'loss': loss.item(), 'accuracy': accuracy.item() } for tag, value in info.items(): logger.scalar_summary(tag, value, step+1) # 2. Log values and gradients of the parameters (histogram summary) for tag, value in model.named_parameters(): tag = tag.replace('.', '/') logger.histo_summary(tag, value.data.cpu().numpy(), step+1) logger.histo_summary(tag+'/grad', value.grad.data.cpu().numpy(), step+1) # 3. Log training images (image summary) info = { 'images': images.view(-1, 28, 28)[:10].cpu().numpy() } for tag, images in info.items(): logger.image_summary(tag, images, step+1) ================================================ FILE: tutorials/04-utils/tensorboard/requirements.txt ================================================ tensorflow torch torchvision scipy numpy