Skip to content
learn
  1. Learn
  2. Machine Learning
  3. Computer Vision

ML-M09

Computer Vision

How machines see: images as tensors, convolutions and pooling, the landmark CNNs, transfer learning, detection, segmentation and vision transformers.

9 lessonsabout 110 min

Computer vision turned deep learning from a research topic into everyday technology: photo search, medical scans, factory inspection and self-driving perception all rest on the ideas in this module.

You will start with how an image becomes a tensor of numbers, then build convolutional networks from kernels, feature maps and pooling. You will trace the landmark architectures from LeNet to ResNet, adapt pretrained models with transfer learning, and use augmentation well. Then you will move beyond whole-image labels to object detection and segmentation, and finish with vision transformers, which treat an image as a sequence of patches.

Lessons

  1. ML-09.01 Images as tensors To a computer, an image is a grid of numbers: one value per pixel for grayscale, three per pixel for color. Beginner 8 min
  2. ML-09.02 Convolutions A convolution slides a small grid of weights, the kernel, across an image and computes a weighted sum at each position, producing a feature map that lights up where the pattern occurs. Beginner 15 min
  3. ML-09.03 Pooling and CNN architecture Pooling layers downsample feature maps, usually by taking the maximum in small windows, which reduces computation and makes features robust to small shifts. Beginner 12 min
  4. ML-09.04 Landmark CNNs: LeNet to ResNet LeNet showed convolutional networks reading digits; AlexNet won ImageNet in 2012 using GPUs and ReLU, starting the deep learning era; VGG went deeper with small kernels; Inception used parallel branches; ResNet added residual connections to train networks with over a hundred layers; EfficientNet scaled depth, width and resolution together. Intermediate 14 min
  5. ML-09.05 Transfer learning and fine-tuning Transfer learning starts from a model pretrained on a large dataset and adapts it to your task. Intermediate 12 min
  6. ML-09.06 Data augmentation Data augmentation creates new training examples by transforming existing ones in ways that do not change the label: flipping, cropping, rotating, adjusting color or mixing images. Beginner 8 min
  7. ML-09.07 Object detection Object detection finds what objects are in an image and where, drawing bounding boxes with class labels. Advanced 15 min
  8. ML-09.08 Image segmentation Segmentation labels every pixel. Advanced 12 min
  9. ML-09.09 Vision transformers Vision transformers split an image into small patches, turn each patch into a token embedding with position information, and process the sequence with a standard transformer. Advanced 12 min

Start with lesson 1