Skip to content

Introduction

In this course block and its subsequent chapters we will demonstrate how to build your own YOLO model in practice. In the end, you will be able to build a "ready-to-go" model that can detect objects in images, videos, and live streams.

Along the way we will explore the four major steps of building a YOLO model:

graph LR
  A[Data Acquisition] --> B[Annotation];
  B --> C[Training];
  C --> D[Inference];
  click A "../acquisition" _self
  click B "../annotation" _self
  click C "../training" _self
  click D "../inference" _self
  classDef active fill:#950f42

Let's get started! 🚀

In addition to the theoretical foundation, we will look at the following chapters using a practical example. This will enable us to better understand and apply the theoretical concepts.

Prerequisites

0. 🏆 What's our goal?

We will start with defining a goal for our practical example.

Build a YOLO model that can detect and classify different types of banknotes (5€ and 10€).

In order to build this kind of model, we will first need to find and generate suitable dataset.

1. 📁 Project structure

Start with creating a new folder for our project:

📁 yolo_training/

2. 💻 Virtual environment

Create a virtual environment. Now, you should have the following structure:

📁 yolo_training/
├── 📁 .venv/

Be sure to activate the environment!

3. 📦 Install packages

Install the necessary packages in this specific order: label-studio, ImageEngine, opencv-python, ultralytics. This time, the sequence of installation is important to due some dependencies.