- 7PPE classesHelmet, shield, safety vest, dust mask, eyewear, gloves, boots
- 2.3ktraining imagesPlus 652 validation and 327 test images
- YOLO-NASdetectorArchitecture found by neural architecture search
The challenge
Safety teams can’t watch every camera, and PPE violations are small objects — a missing glove, a mask under the chin — on people moving constantly through cluttered industrial scenes.
Our approach
We trained a YOLO-NAS detector on a Roboflow PPE dataset covering seven equipment classes, with held-out validation and test splits to measure generalization before any site deployment.
Workflow
How it works, step by step.
The pipeline in 4 stages, from data to monitor. Each stage has a clear input, a clear output and a reason to exist.
- 01
Data
Roboflow PPE detection dataset: 2.3k training, 652 validation and 327 test images.
- 02
Train
YOLO-NAS fine-tuned for the seven PPE classes.
- 03
Detect
Per-frame boxes and confidence scores for each item of equipment.
- 04
Monitor
Detections support compliance checks for safety teams at construction, manufacturing and mining sites.
Results
What it produced.
Stack & techniques
- YOLO-NAS
- Object detection
- Roboflow
- Python
Delivered by members of the Saerosense founding team. Client names are withheld unless public — references are available on request.
Related service
Start a project
Have a similar problem?
Send a few lines about your data and the outcome you want. We reply within two business days with a feasibility read and a proposed scoping plan.