01
Any camera you already own
RTSP, USB, GigE Vision, thermal and depth. Most sites go live on the hardware already mounted on the wall.
VRTLST Vision
One model that understands every camera you own
Vision is the engine underneath Inspect, Guard and Watch. One model stack that handles detection, segmentation, tracking and classification — trained on industrial footage rather than internet photos, so it arrives already knowing what a pallet, a weld and a hard hat look like.
person · 0.98
forklift · 0.94
pallet · 0.91
hard-hat · 0.88
What it does
The core computer-vision platform
02
One model, four tasks
Detection, segmentation, tracking and classification come from the same deployment. No separate pipeline per question.
03
Few-shot training
The base models are pre-trained on industrial footage, so a new class is a fine-tune from about fifty labelled frames.
How it works
From a camera to a decision
Detection, segmentation, tracking and classification from a single deployment — no per-camera retraining.
- Works with the RTSP cameras already on the wall
- Fine-tune on 50 labelled frames, not 50,000
- Runs cloud, on-prem or fully offline
- Connect a stream
- Point Vision at an existing camera URL. It handles decoding, framing and calibration.
- Label a handful of frames
- Draw on fifty examples of the thing you care about. Training runs in under an hour.
- Deploy where it belongs
- The same model artefact runs in your cloud, on a plant server, or on a VRTLST Edge device.
- Correct and improve
- Operator corrections queue up as labelled data for the next training round.
4 vision tasks from one model
~50 frames to teach a new class
38 ms median inference at the edge
0 frames used to train shared models
Specifications
The technical detail
What your integrator will ask for on the first call.
- Inputs
- RTSP · USB · GigE Vision · thermal · depth
- Tasks
- detection · segmentation · tracking · classification
- Resolution
- up to 4096 × 2160
- Frame rate
- 60 fps sustained per stream
- Training set
- ~50 labelled frames per new class
- Deployment
- cloud · on-prem · edge · air-gapped
Getting started
How a Vision deployment goes
The same shape every time, whichever product you start with.
- Week 0
One hour of footage
Send us recorded video from a single line. We come back with what the model found in it — before anyone signs anything.
- Week 1–2
One line, one model
We stand up a pilot on the line that hurts most, using the cameras already mounted. Operators start correcting it on day one.
- Week 3–4
Wired into the line
The verdict moves from a dashboard to the PLC, the MES or the alerting channel your shift leads actually watch.
- Month 2
Second line, second site
The model and its thresholds travel. Command stages the rollout so a bad version never reaches more than one line.
- Ongoing
It gets better
Corrections accumulate, drift gets flagged, and new defect classes are a fine-tune rather than a project.
Does it work with our existing cameras?
Almost always. Vision consumes standard RTSP, USB, GigE Vision, thermal and depth streams. If a camera can produce a stream your VMS can read, Vision can read it too.
What if our lighting is inconsistent?
The base models are trained on industrial footage, which is by definition badly lit. Where lighting genuinely defeats a model we will say so during the pilot rather than after the rollout.
Can we run our own models on it?
Yes. Vision will serve an ONNX model you bring, with the same deployment, versioning and rollback machinery Command gives our own models.
How is a new class added later?
Label about fifty frames, fine-tune, stage it to one camera in Command, then promote it. No re-architecture and no downtime on the rest of the estate.
Keep exploring
The rest of the stack
Inspect
Catch the defect before it leaves the line
Sub-millimetre surface inspection at line speed, with a decision the PLC can act on.
Explore Inspect Guard
Every zone watched. Every rule enforced.
Draw a boundary once. Guard enforces it on every shift, on every camera, without a human staring at a wall of screens.
Explore Guard Watch
Know what's happening the moment it happens
Continuous monitoring across every line and every site, with alerts that reach the right person in seconds.
Explore Watch