Train your first YOLO model

Take Ultralytics YOLO from zero to a model trained on your own data — installation, the YOLO label format, training, evaluation, video inference, and export. Assumes a dataset that's already been planned, collected, labeled, and split (see *Building High-Performance YOLO Datasets* for that work).

By Ultralytics Academy

Begin course Sign in to save progress

What you'll learn

Train, validate, and export an Ultralytics YOLO model on a custom dataset, and run it on video — all from the command line and Python.

Install Ultralytics and run Ultralytics YOLO predictions in Python and on the CLI.
Pick the right model size for your latency, accuracy, and hardware budget.
Convert a custom dataset into the Ultralytics YOLO label format and write a data.yaml.
Train, validate, and resume training with sensible defaults.
Export to ONNX, TensorRT, or QNN for deployment.

What you'll build

A small custom dataset in Ultralytics YOLO format with a data.yaml.
A trained best.pt checkpoint that beats the pretrained baseline on your data.
A short video inference script that overlays detections.
An exported model file (ONNX or engine) ready for your runtime.

Prerequisites

Comfortable running Python and shell commands.
An NVIDIA GPU with CUDA available will make training faster but is not required — Apple Silicon and CPU work too.
Recommended: complete Computer Vision Foundations and Building High-Performance YOLO Datasets first.

Course content

5 modules · 10 lessons

Module 1

Setup

Install Ultralytics

Get a working environment in one command — and verify GPU access.

Module 2

Predict

Predict on an Image

Run a pretrained Ultralytics YOLO model and read the result object — boxes, confidences, classes.

Choose an Ultralytics YOLO Model Size

Nano, small, medium, large, x-large — what each one costs and where each one wins.

Module 3

Data

Prepare a Custom Dataset

Convert images and labels into the Ultralytics YOLO format — folders, files, normalized coordinates.

Write a Dataset YAML

The single config file that connects your folders to YOLO's training loop.

Module 4

Train

Train Your First Model

One Python call, sensible defaults, and a checkpoint at the end.

Validate and Diagnose

Beyond the single mAP number — per-class AP, confusion matrices, and what they tell you to do next.

Resume and Iterate

Continue from where training stopped, fine-tune best.pt, and run experiments cheaply.

Module 5

Ship

Run Inference on Video

From single-image predict to real-time video — and what changes when frames are continuous.

Export for Deployment

ONNX, TensorRT, CoreML, OpenVINO, QNN — pick the format your runtime actually wants.