Skip to content

🌱 Beginner Tier: Images and Classical Vision

Levels 0–2 · Goal: understand images and master the classical toolkit · ~10–14 weeks at about an hour a day

You start here if you know some Python but have never worked with images seriously. By the end you'll understand images from the sensor up, process them fluently, and build classical vision systems that still power calibration, 3D reconstruction, AR, and tracking in modern products.

What you'll be able to do

  • Explain how a camera turns light into an array of numbers, and how color spaces, gamma, and compression change those numbers.
  • Manipulate images confidently with NumPy, OpenCV, and scikit-image, without falling into the BGR, overflow, or coordinate traps.
  • Enhance, filter, and threshold images, detect edges, and work in the frequency domain.
  • Segment and measure objects with morphology, contours, and watershed.
  • Detect and match features, estimate homographies robustly with RANSAC, and stitch panoramas.
  • Calibrate a camera, estimate pose, compute depth from stereo, and track motion with optical flow and Kalman filters.

The levels

  • Level 0: Images and tools

    What computer vision is, the toolkit, how digital images work, images as arrays, and image and video I/O.

    Start Level 0

  • Level 1: Image processing

    Histograms, convolution and filtering, edges, the frequency domain, morphology and segmentation, and geometric transforms.

    Start Level 1

  • Level 2: Classical vision

    Features, matching and RANSAC, cameras and calibration, stereo and multi-view geometry, motion and tracking, and classical recognition.

    Start Level 2

Tier checkpoint

You're ready for the Intermediate tier when you can do all of these without notes:

  • Explain why img1 + img2 on uint8 images gives wrong results, and fix it.
  • Implement 2D convolution in NumPy, and explain the difference between convolution and correlation.
  • Walk through the Canny edge detector step by step.
  • Derive the Harris corner response from the structure tensor.
  • Explain how many RANSAC iterations you need, and why.
  • Write down the pinhole projection equation, and name every parameter.
  • Convert a disparity map to depth, and explain why depth error grows with distance.

See the full roadmap for every concept in this tier.