YouTip

torchvision is an extension library in the PyTorch ecosystem specifically designed for computer vision tasks, providing the following core functions: 1. **Pre-trained Models**: Includes implementations of classic CNN architectures (such as ResNet, VGG, AlexNet, etc.) 2. **Dataset Tools**: Built-in common vision datasets (such as CIFAR10, MNIST, ImageNet, etc.) 3. **Image Transforms**: Provides various image preprocessing and data augmentation methods 4. **Utility Tools**: Includes video processing, image manipulation and other auxiliary functions # Install torchvision (usually installed together with PyTorch) pip install torch torchvision * * * ## Core Components Analysis ### 1. torchvision.models Provides pre-trained computer vision models, which can be directly used for transfer learning: ## Example import torchvision.models as models # Load pre-trained model resnet18 = models.resnet18(pretrained=True) alexnet = models.alexnet(pretrained=True) vgg16 = models.vgg16(pretrained=True) #### Common Model List: | Model Name | Use Case | Parameters | Top-1 Accuracy | | --- | --- | --- | --- | | ResNet | General image classification | 11M-60M | 69%-80% | | VGG | Feature extraction | 138M | 71.3% | | MobileNet | Mobile applications | 3.4M | 70.6% | | EfficientNet | Efficient models | 5M-66M | 77%-84% | * * * ### 2. torchvision.datasets Built-in common computer vision datasets, simplifying data loading process: ## Example from torchvision import datasets # Load CIFAR10 dataset train_data = datasets.CIFAR10( root='data', train=True, download=True, transform=transforms.ToTensor() ) # Load MNIST dataset test_data = datasets.MNIST( root='data', train=False, download=True ) #### Supported Dataset Types: ## Example graph TD A[torchvision.datasets] --> B A --> C A --> D B --> B1[CIFAR10/100] B --> B2[MNIST/FashionMNIST] B --> B3 C --> C1 C --> C2 D --> D1 * * * ### 3. torchvision.transforms Core tools for image preprocessing and data augmentation: ## Example from torchvision import transforms # Define image transformation pipeline transform = transforms.Compose([ transforms.Resize(256),# Resize transforms.CenterCrop(224),# Center crop transforms.ToTensor(),# Convert to tensor transforms.Normalize(# Normalize mean=[0.485,0.456,0.406], std=[0.229,0.224,0.225] ) ]) #### Common Transform Methods Classification: | Category | Method Example | Purpose | | --- | --- | --- | | Geometric transforms | RandomRotation, RandomResizedCrop | Increase position invariance | | Color transforms | ColorJitter, Grayscale | Enhance color robustness | | Blur/Noise | GaussianBlur, RandomErasing | Prevent overfitting | | Composite transforms | RandomApply, RandomChoice | Flexible combination strategy | * * * ## Practical Example: Image Classification Workflow ### 1. Data Preparation ## Example import torch from torchvision import datasets, transforms from torch.utils.data import DataLoader # Define data transformation train_transform = transforms.Compose([ transforms.RandomHorizontalFlip(), transforms.RandomRotation(10), transforms.ToTensor(), transforms.Normalize((0.5,),(0.5,)) ]) # Load dataset train_set = datasets.CIFAR10( root='./data', train=True, download=True, transform=train_transform ) # Create data loader train_loader = DataLoader( train_set, batch_size=32, shuffle=True ) ### 2. Model Training ## Example import torch.nn as nn import torch.optim as optim # Use pre-trained model model = models.resnet18(pretrained=True) # Modify the last layer (adapt to CIFAR10's 10 classes) num_ftrs = model.fc.in_features model.fc