Skip to main content

Computer Vision

Eyes on the line, 24 hours a day.

Vision models for quality control, counting, safety and plant monitoring: they answer questions like "how many people are on the shop floor?" or "is this part compliant?" in real time, on on-premise GPUs or on the machine itself.

In practice

Quality inspection

Surface, assembly or shape defects detected in line with convolutional networks and segmentation; thresholds tuned to your tolerance.

People and safety

Presence counting, danger zones, PPE: monitoring that helps safety while respecting privacy (anonymization, no biometric recognition).

Processes and plants

Reading gauges, fill levels, positions, OCR on labels and documents: the camera becomes a smart sensor.

Deploy where needed

On-premise GPUs with NVIDIA Triton for multiple parallel streams, or on-board inference on edge devices; integration with PLCs, MES and alarm systems.

Stack and technologies

  • OpenCV
  • PyTorch
  • NVIDIA Triton · TensorRT
  • ONNX
  • Python

Project

Delivered

On-premise vision, next to the LLMs

In the on-premise installation delivered for a local company (HPE system with four NVIDIA Tesla GPUs), the vision models run next to the language models, served by the same NVIDIA Triton: images and video never leave the company.

Deployment
on-premise · HPE 4× GPU
Serving
NVIDIA Triton
Data
images and video on site
Alongside
on-premise LLMs

How we work

Four steps, always the same

  1. 01

    Image collection and definition of the criteria (what is a defect, what is a risk)

  2. 02

    Annotation, training and evaluation on real cases

  3. 03

    Deployment on GPU or edge, integration with line and alarms

  4. 04

    Performance monitoring and retraining with new cases

Frequently asked questions

How many images are needed?

For a prototype a few hundred examples per class are often enough thanks to transfer learning; the dataset then grows with the real cases collected in production.

Do you recognize people?

We count and localize, we do not identify: no biometric recognition, images anonymized where possible, in line with GDPR and the EU AI Act.

Get started

Ready to transform your business?

Contact us for a free consultation and discover how we can help you reach your goals.