Skip to main content
Foundation pathwaybeginner ~3 hours 10 lessons Final exam · Certificate

From the first image to the first model

Computer Vision Foundations

Build the vocabulary, intuition, and first hands-on reflexes for modern computer vision — tasks, datasets, metrics, and what they mean for a real product.

By Ultralytics Academy

Begin course Sign in to save progress
Ultralytics YOLO object detection examples on a city street
What you'll learn
Frame a computer vision problem, choose the right task and labels, evaluate a model honestly, and know what "good" looks like before you train.

  • Match a product need to the right vision task — classification, detection, segmentation, pose, or OBB.

  • Decide between bounding boxes and segmentation masks based on shape, cost, and downstream use.

  • Plan datasets and annotation guidelines that match deployment conditions.

  • Read precision, recall, and mAP together with examples to find the next improvement.

  • Recognize the difference between training, validation, and test data — and why people get this wrong.

What you'll build

  • A one-page task spec for a vision project: what the model predicts, what it ignores, what shape the output takes.

  • A dataset plan that lists splits, edge cases, label rules, and review criteria.

  • A small evaluation worksheet that converts metrics into a decision: ship, iterate, or rethink.

Prerequisites

  • Basic Python familiarity is helpful but not required — this course is concept-first.

  • Comfort installing software locally if you want to follow the optional ultralytics install in lesson 10.

Course content

4 modules · 10 lessons

Module 1

Frame the Problem
1

Frame the Problem

Turn a fuzzy product idea into a single sentence the model can answer.

Module 2

Pick a Task
2

The Five Vision Tasks

Match an output sentence to classification, detection, segmentation, pose, or oriented bounding boxes.

3

Understand Object Detection

Boxes, classes, and confidence — what detection actually returns and what it does not.

4

Boxes vs Masks: When You Actually Need Segmentation

Move from rectangles to pixels — and only when the shape genuinely matters.

5

Pose Estimation and OBB

Two specialized tasks and the kinds of problems where they shine.

Module 3

Datasets and Labels
6

Datasets that Reflect Reality

The model only knows what's in the dataset. Build the dataset like the model depends on it.

7

Annotation Quality

Half a million boxes are worse than fifty thousand consistent ones.

Module 4

Read the Metrics
8

Splits that Tell the Truth

Why training, validation, and test splits exist — and the leak that ruins half of them.

9

Reading Detection Metrics Honestly

Precision, recall, mAP — what they hide and how to look at them together.

10

From Lab to Production

The smallest end-to-end loop: install, predict, measure, decide.