Handbook logo

Handbook

Quick tech ticks and comparisons

Computer Vision

Learn computer vision from the ground up: how images become numbers, what object detection and localisation mean, how bounding boxes are measured, and how modern detectors such as Faster R-CNN, YOLO, and SSD find objects in images.

Recommended Learning Path:

  1. Part 1: Image Processing Fundamentals — Pixels, colour models, image shapes, matrix operations, filters, convolution, and feature extraction — how a computer reads an image as numbers before it can understand it.
  2. Part 2: Introduction — What computer vision is, why it is hard, the standard pipeline, and how it relates to deep learning.
  3. Part 3: CNNs for Computer Vision — Learned filters, feature maps, pooling, the edges-to-objects hierarchy, and the idea of a pretrained CNN backbone — just enough to understand the detectors below.
  4. Part 4: Object Detection Basics — Bounding boxes, coordinate formats, IoU, precision/recall, mAP, and the two-stage vs one-stage distinction.
  5. Part 5: Loss in Object Localisation — L1, L2, Smooth L1, IoU, GIoU, DIoU, and CIoU losses with formulas and code.
  6. Part 6: Region-Based Detectors — R-CNN, Fast R-CNN, Faster R-CNN, the Region Proposal Network, and the limits of two-stage detectors.
  7. Part 7: Anchor Boxes — What anchors are, how they are placed and matched to ground truth, and their role in RPN, YOLO, and SSD.
  8. Part 8: One-Stage Detectors — YOLO and SSD — How YOLO and SSD work, how to decode YOLO output, and how non-maximum suppression produces final boxes.
  9. Part 9: YOLO Evolution and YOLO11 Demo — The YOLO family from v1 to YOLO11, and a practical detection demo with ultralytics.

Related Deep Learning topics (for the full mathematical depth):