Train your first YOLO model

Take Ultralytics YOLO26 from a fresh install to a custom-trained, validated, and exported detector. You'll predict on images, lay out a real dataset, train and diagnose a pill detector, run models on video with tracking, and export to ONNX with proof that it matches PyTorch. Every step runs in Python and on the CLI, on a laptop CPU, with the real output shown. Planning, collecting, and labeling the dataset itself is covered in Building High-Performance YOLO Datasets.

By Ultralytics Academy

Begin course Sign in to save progress

What you'll learn

Train, validate, and export an Ultralytics YOLO26 model on a custom dataset, run it on video, and know how to diagnose each step when it goes wrong — from both Python and the command line.

Install Ultralytics, confirm which device it will use, and run YOLO26 predictions in Python and on the CLI.
Choose a model size from a measured latency budget instead of a leaderboard.
Lay out a dataset in the Ultralytics YOLO format, convert labels, and write a data.yaml that resolves from anywhere.
Train, validate, resume, and warm-start runs, and turn validation artifacts into a fix list.
Stream and track video, then export to ONNX, TensorRT, OpenVINO, and more, and prove the export matches PyTorch.

What you'll build

A YOLO-format pill dataset with a data.yaml that Ultralytics reads with zero corrupt labels.
A trained best.pt detector, with its validation numbers next to the pretrained baseline.
A streaming video script that counts unique objects with tracking and checks it keeps up with the frame rate.
An ONNX export whose detections and mAP match the PyTorch model under matched settings.

Prerequisites

Python 3.10+ and comfortable running Python scripts and shell commands.
No GPU required: every example in this course runs on a laptop CPU in minutes. An NVIDIA GPU or Apple Silicon makes larger runs faster.
Recommended: complete Computer Vision Foundations and Building High-Performance YOLO Datasets first.

Course content

5 modules · 10 lessons

Module 1

Setup

Install Ultralytics

Install Ultralytics in a clean environment, read the yolo checks report to confirm your device, and prove the setup with one prediction.

Module 2

Predict

Predict on an Image

Run a pretrained YOLO26 model from Python, read classes, confidences, and pixel boxes from the results, and tune the arguments that fix most misses.

Choose an Ultralytics YOLO Model Size

Compare YOLO26 n, s, m, l, and x, benchmark them on your target hardware, and pick the smallest size that fits your latency budget.

Module 3

Data

Prepare a Custom Dataset

Lay out images and labels in the Ultralytics YOLO format, convert pixel boxes to normalized label lines, and catch broken labels before training.

Write a Dataset YAML

Write the data.yaml that points Ultralytics at your folders and names your classes, make its paths resolve from anywhere, and smoke-test it with yolo val.

Module 4

Train

Train Your First Model

Fine-tune YOLO26 on your dataset with one call, read the live training log line by line, and find best.pt and the plots that matter in the run folder.

Validate and Diagnose

Validate best.pt against a baseline, read the confusion matrix, per-class numbers, and prediction images, and turn them into a specific fix list.

Resume and Iterate

Resume an interrupted run exactly, warm-start a new round from best.pt with a learning rate that actually applies, and compare runs side by side.

Module 5

Ship

Run Inference on Video

Run YOLO26 on video files and live cameras with bounded memory, track objects to count each one once, and check the model keeps up with the frame rate.

Export for Deployment

Export best.pt to ONNX, OpenVINO, TensorRT, CoreML, LiteRT, or QNN, then prove the export matches the PyTorch model before you ship it.

Get Started

Train your first YOLO model

Course content