Lab tutorial

Teaching Robots to See

An introduction to Convolutional Neural Networks, YOLO object detection, and a practical guide to training a detector that works on a real robot.

Parts 1–3

Slides

CNNs, YOLO, and a step-by-step training guide. Use the arrow keys to move through them, S for speaker notes, Esc for an overview.

Open slides Download PDF

Part 4

Training notebook

Train a YOLO11n detector on a 4-class wildlife dataset, read the results, look at predictions, and find the failures. Shown here with saved outputs.

View notebook Download .ipynb

What's covered

  1. What is a CNN? Images as numbers, filters and feature maps, layers, how learning works
  2. YOLO. Detection vs classification, grid prediction, backbone/neck/head, IoU, NMS, model sizes, transfer learning
  3. Training, practically. Collecting and labelling data, splits, training, reading curves and metrics, augmentation, troubleshooting
  4. Live notebook. Train a model end to end and inspect its predictions