Computer vision

Topics covered

• CNN
• Classification
• Object detection
• Segmentation
• Embeddings
• Vision Transformers
• Multimodal vision

Lessons in this unit

1. Convolution — a magnifying glass that sweeps across the image 80 min
   The operation that lets a network "see" edges, textures and shapes, computed by hand and then with NumPy.
2. Convolutional networks — stacking ever wider views 70 min
   Convolutions, activations and pooling chained into a complete CNN; receptive field, parameter counting and data augmentation.
3. Detection and segmentation — what's there and where it is 80 min
   Boxes, IoU, non-maximum suppression and average precision; the pieces any detector is built and evaluated with.
4. Vision Transformers — an image is worth 16 × 16 words 75 min
   Cutting the image into patches, turning them into tokens and letting attention do the rest; shapes, parameters and cost.

Prerequisites

Before this unit it helps to have done:

• Engineering with PyTorch (Course 3 · Unit 2)