Module 1
Teach a Machine to See
A 16-day journey from "a photo is just numbers" all the way to training AIs, detecting objects, and building real projects like an engineer. Two parts — play them in order, about an hour each.
How to get the most out of this module
This module isn't just for fun — it's to make you a real builder. Do these and you'll be able to build vision apps yourself (or tell an AI assistant exactly what to build):
🔬 Part 1 · How Machines See
The core skills — from pixels all the way to YOLO.
How a Computer Sees
Zoom inside a photo, mix light, split colors, and do real image math. Discover the big secret: pictures are just grids of numbers.
Find the Edges
Slide a tiny "magic window" over the numbers to find edges and shapes — the #1 trick every vision AI uses. Invent your own photo filters!
From Pixels to a Guess
You draw, and a real AI recognizer makes its best guess — with confidence bars. Then try to fool it and see it admit "I'm not sure!"
Train Your Own AI
Become the teacher: train your very own AI with your webcam to recognize your moves — then control a game with it.
Good Data vs Bad Data
Break an AI on purpose with biased or messy data, then fix it. Learn the golden rule: an AI is only as smart as its examples.
Scoring an AI
True/false positives, the confusion matrix, precision, recall & F1 — learn how scientists grade an AI, as a slider game.
What AND Where
Object detection: drawing boxes around things, and the "overlap score" (IoU) that grades them — like self-driving cars.
The 3 Flavors of Vision
Classification vs detection vs segmentation — the three big ways an AI can "see," side by side.
How YOLO Works
The famous "You Only Look Once" detector — see how it splits an image into a grid and spots everything in one glance.
Ship It!
The boss project: put everything together into your own working vision app, and graduate Module 1.
🛠️ Part 2 · Build Real Projects & Make Smart Choices
Think like an engineer — split data, beat overfitting, and pick the right model for real jobs.
Don't Grade on Homework
Train / validation / test split — why testing on practice data is cheating, and how to get an honest score.
Memorizer vs Learner
Overfitting vs underfitting — spot a model that memorized instead of learned by the practice-vs-exam gap.
Make Your Data Tougher
Data augmentation — beat "distribution shift" by practicing with real-world mess (tilt, dark, blur).
The Real World Is Messy
Failure analysis — play detective, diagnose why the AI got each case wrong, and pick the fix.
Pick the Right Tool
Real case studies — classify, detect, segment, track, or just a rule? Which model for which job, and good vs bad.
Fast, Accurate, Private
Speed/accuracy/size, on-device vs cloud, and being a responsible builder — the Model Shop finale.
Real tools & open-source models
You know the ideas AND the pro words now. Here are the real, free tools engineers use — for hobby projects and professional apps.
🧰 Beginner-friendly (start here)
- Teachable Machine — train a webcam model, no code beginner
- Hugging Face — find open models & demos beginner
- MediaPipe — ready-made vision tasks beginner
- TensorFlow.js — ML in the browser beginner
🛠️ Level up (pro tools)
- Ultralytics YOLO — real-time detection & segmentation pro
- transformers.js — pretrained vision in the browser beginner
- OpenCV — image processing & camera tools pro
- CVAT — label your own image datasets pro