Custom AI models

Something specific to watch for? We train QVision for it.

Beyond the built-in detections, QVision can learn what matters on your site: your products, your equipment, your hazards. Every custom model follows the same path, from a visit to your site to monitoring after it goes live.

Custom modelGloves on Packing Line 24 / 6
  1. Site inspectionDone
  2. Data collection and staged scenesDone
  3. AI trainingDone
  4. TestingIn progress
  5. DeploymentNext
  6. On-site monitoringNext
Use cases

What teams ask us for.

A few of the requests that go through this path. After the site visit, we tell you what's realistic and how we'll measure it.

  • Your own defects

    Flaws specific to your product, your packaging or your line.

  • More protective equipment

    Gloves, goggles, harnesses or ear protection, where your rules require them.

  • Vehicles and machines

    Forklifts, loaders, tankers or a particular type of truck.

  • Things out of place

    Blocked exits, pallets left in walkways, tools left on a line.

  • Doors and gates left open

    A gate, a hatch or a fire door that should be closed.

  • Counting

    Bags, pallets or vehicles passing a point on your site.

The path

From your site to a live model, in six steps.

The same path for every custom model, run by the QVision team with yours.

  1. 1

    Site inspection

    We visit the site, look at the cameras, the angles and the light, and agree with your teams on exactly what the model must catch and what counts as success.

    You getA written scope and the success criteria

  2. 2

    Data collection and staged scenes

    We record footage from your own cameras, in the conditions they work in. Rare events don't wait for a camera, so we stage them safely with your teams until the model has enough examples.

    You getFootage from your cameras, covering every situation

  3. 3

    AI training

    Our team labels the footage and trains a model for your use case, then measures it on footage it has never seen.

    You getA trained model and its first results

  4. 4

    Testing

    The model runs on your live cameras without triggering anything, and on recordings from your site. We review the results with you against the criteria agreed at the visit.

    You getTest results you sign off

  5. 5

    Deployment

    Once you sign it off, the model goes to your edge boxes and becomes a detection like any other, with zones, rules, incidents and analytics.

    You getThe model live on your sites

  6. 6

    On-site monitoring

    After go-live, we keep watching its results on site. When something changes, a new product, a new season, a moved camera, we collect fresh footage and retrain.

    You getRegular reviews and updates

Testing

Proven before it's switched on.

  • Measured on unseen footageResults come from recordings the model never trained on.
  • Run quietly on live camerasIt detects on your site without sounding, texting or opening anything.
  • Signed off by your teamsIt goes live only once it meets the criteria agreed at the visit.
After go-live

Watched long after it goes live.

  • Results reviewed on siteWe follow its detections and false alarms with your teams.
  • Retrained when your site changesNew products, new lighting or a moved camera mean fresh footage and a new version.
  • Updates from your hubA new version reaches every edge box that uses it.
In QVision

Works like every QVision detection.

  • Zones and schedules

    Draw where it applies and choose when, camera by camera.

  • Rules and responses

    Sirens, barriers, line stops, incidents, SMS, email and webhooks.

  • Incidents with evidence

    The snapshot with its box and the recorded clip.

  • Analytics

    By zone, camera, weekday and hour, like every other detection.

Questions

What can a custom model detect?

Most things a person could reliably spot on your camera picture: an object, a piece of equipment, a state or a defect. After the site visit, we tell you what's realistic and how we'll measure it.

How much footage do you need?

It depends on the use case and on how often the situation happens. That's why we stage the rare scenes with your teams instead of waiting for them.

Does our footage leave the site?

Training footage is collected with your agreement and used to build your model. Once deployed, the model runs on your edge boxes inside your network, like every QVision detection.

What happens when our site changes?

On-site monitoring catches it. We collect new footage, retrain the model and roll the update out from your hub.

See QVision on a site like yours.

A walkthrough on footage close to your sites, with answers for your operations, security and IT teams.

Book a demo