Training Trains Internship
Student login
← PyTorch learning path
PyTorch · Module 7 of 10

Computer vision with PyTorch

Build image models with torchvision.

📖 Key ideas

  1. Use nn.Conv2d, nn.MaxPool2d, nn.BatchNorm2d.
  2. Pretrained models: torchvision.models.resnet50(weights="DEFAULT").
  3. Replace the final layer for your classes.
  4. Augment with transforms (RandomResizedCrop, RandomHorizontalFlip).
  5. Detection/segmentation models also in torchvision.
💡 Real-life example: Using a pretrained ResNet is borrowing an expert's eyes and teaching them your new labels.
nn.Conv2d2D convolution layer
ResNetResidual CNN family
weights="DEFAULT"Loads best pretrained weights
Final layer replacementAdapting to new classes