Download the Tutorial Inputs
Download the Helmet Dataset in YOLO26 format and its matching trained Model.
The tutorial uses a Dataset and a trained Model. Download both before creating your Project. These files are inputs to Knotest; they are not themselves a Knotest Project.
Download the Dataset
Download Helmet Dataset version 2 from Roboflow Universe.
- On the Roboflow download page, choose YOLO26 as the export format.
- Choose the ZIP download option and save the archive to your computer.
- Extract the archive without changing its internal folder structure.
Knotest imports this export as YOLO Object Detection with axis-aligned
bounding boxes. The Dataset uses class 0 for helmet. Keep data.yaml and
the image and label folders together so Knotest can identify the splits and
annotations.
The Dataset is attributed to Yusin, Helmet Dataset version 2, on Roboflow Universe and is published under CC BY 4.0. Retain the Dataset's README and attribution when reusing or distributing it.
Download the Model
Download the trained Helmet YOLO26 Model
The tutorial Model is helmet-object-detection-best.pt, an approximately 5.4 MB
Object Detection Model pretrained by the Knotest team using the Ultralytics
YOLO26 architecture for the matching helmet class. It is a Knotest-trained
Model, not an official Ultralytics weight. It has been validated with the
documented Knotest Prediction Set and Evaluation workflow.
Use these Model settings when importing it:
| Setting | Value |
|---|---|
| Backend / task | YOLO / Object Detection |
| Class order | 0: helmet |
| Input dimensions | 640 × 640 |
| Minimum object confidence | 0.25 |
| Duplicate-box overlap (NMS IoU) | 0.5 |
| Class-agnostic NMS | Off |
| Maximum detections | 300 |
The Model is derived from Ultralytics YOLO26. Ultralytics provides YOLO26 as free, open-source software under the AGPL-3.0 license and also offers an Enterprise license for commercial use. Review the Ultralytics licensing terms and use the license appropriate for your intended deployment.
Do not replace this file with the general-purpose yolo26n.pt COCO weights.
Those weights do not include a helmet class and therefore do not match this
Dataset or the tutorial results.
Check the downloads
Before continuing, confirm that:
- the Dataset archive extracts without errors;
- the extracted folder contains
data.yamlplus matching image and label folders; - the Model filename is
helmet-object-detection-best.pt; and - the Dataset and Model both use class
0: helmet.
Continue to Create a Project.