Smolry/Helmet-classifer

Helmet and No-Helmet Detection — YOLO11s

A YOLO11s object detection model trained for detecting helmets and no-helmet instances in images.

The training dataset was obtained from Roboflow and was originally created by another Roboflow user. The dataset was forked and used for training this model. Dataset attribution and licensing information are provided below.

Model Description

This model is a custom-trained Ultralytics YOLO11s object detection model.

The model predicts two classes:

Class ID	Class
0	helmet
1	no-helmet

The model accepts images at a nominal resolution of 640 × 640 pixels and produces bounding-box detections for the two classes.

Model Architecture

• Architecture: YOLO11s
• Task: Object Detection
• Parameters: 9,428,566
• Layers: 181
• GFLOPs: 21.6
• Input image size: 640 × 640
• Number of classes: 2

The checkpoint identifies the architecture as YOLO11s and reports 181 layers, 9,428,566 parameters, and 21.6 GFLOPs.

Intended Use

This model is intended for research, experimentation, and computer vision applications involving helmet compliance detection.

Potential applications include:

• Helmet detection in CCTV footage
• Road safety monitoring
• Traffic violation detection
• Industrial safety monitoring
• Motorcycle helmet compliance analysis
• Computer vision research

The model is particularly intended as a component of a larger computer vision pipeline rather than as a complete traffic-violation system.

For example:

    CCTV Image
         |
         v
    Helmet Detector
         |
         +---- helmet
         |
         +---- no-helmet
         |
         v
    Person / Vehicle Association
         |
         v
    Number Plate Detection
         |
         v
    Violation Processing

This model itself only performs helmet/no-helmet object detection.

Classes

The class mapping stored in the trained checkpoint is:

0: helmet
1: no-helmet

The model should therefore be interpreted using this class mapping when processing its predictions.

Training

The model was trained using a dataset exported from Roboflow.

The original dataset was not created by the author of this model. Instead, the dataset was forked from an existing Roboflow dataset and subsequently used for training.

The checkpoint stores the following training configuration.

Training Configuration

Parameter	Value
Task	Detection
Image size	640
Batch size	16
Epochs configured	100
Pretrained	True
Optimizer	Auto
Workers	8
AMP	True
Seed	0
Deterministic	True
Patience	10
Validation	True
Validation split	val

The checkpoint was produced using Ultralytics version 8.3.233.

The stored checkpoint metadata identifies the model as an ultralytics.nn.tasks.DetectionModel.

Data Augmentation

The stored training configuration includes the following augmentation settings:

Augmentation	Value
Mosaic	1.0
MixUp	0.12
Copy-Paste	0.05
Horizontal Flip	0.5
Scale	0.6
Rotation	4.0
Translation	0.1
Shear	1.0
Perspective	0.0004

These values are reported from the training configuration stored inside the model checkpoint.

Dataset

Original Dataset

The training dataset was obtained from Roboflow.

Original dataset:

Original dataset author:

Roboflow project:

Dataset version:
[v1 2026-01-25 2:39am]

The dataset was forked from the original Roboflow project and used as the basis for training this model.

Dataset Attribution

This model does not claim ownership of the original dataset.

The dataset and its annotations remain subject to the original dataset's license and attribution requirements.

Users of this model should consult the original dataset page and license before redistributing the dataset, annotations, or derived datasets.

Data Preprocessing

The model was trained using the YOLO-compatible dataset configuration exported from Roboflow.

The checkpoint references the following dataset configuration:

/content/Helmet-and-Non-Helmet-Detection--2/data.yaml

The original training environment was hosted in Google Colab / Google Drive according to paths recorded in the checkpoint.

Model Input

The model expects an image input and was trained using:

640 × 640

Ultralytics handles the necessary image preprocessing during normal inference.

Inference

Install Ultralytics:

pip install ultralytics

Load the model:

from ultralytics import YOLO

model = YOLO("best.pt")

results = model("image.jpg", imgsz=640)

for result in results:
    result.show()