125 Vision Models, Ready to Evaluate
Finding a vision model is easy. Working out which one runs on your hardware, in your format, under a licence you can actually use is the part that eats a month — and the Model Zoo is built to shorten it.
What's In the Library
125
Curated models
Vision models for evaluation across the common edge AI tasks — detection, classification, segmentation and reading.
5
Filter types
Narrow the library by domain, task, architecture, format and supported hardware until you are looking at real candidates.
4
Format groups
Check ONNX, PyTorch and TFLite availability before you invest a week discovering a model will not run on your target.
A Faster Starting Point for Model Evaluation
- 01
Browse
Filter by domain, task, architecture, format or supported hardware. Go from a library to a shortlist in a couple of clicks.
- 02
Test
On supported model pages, upload an image and run a quick test in the browser. See how a model behaves on your imagery before you download anything.
- 03
Download
Review architecture, input size, licence and available formats, then take the model away for proper evaluation.
Five Ways to Narrow the Library
Most model libraries make you start from an architecture name. Starting from the job and the target gets you to a real shortlist faster.
Domain
Start from the world the model was trained for — industrial, transport, retail, environmental — rather than from an architecture name.
Task
Detection, classification, segmentation, counting or reading. The task decides the model far more often than the brand does.
Architecture
Compare families directly when you already know the shape of what you need, and see what each trades away.
Format
ONNX, PyTorch, TFLite. Filter to what your toolchain actually consumes so nothing on your shortlist is a dead end.
Supported hardware
Filter to what will run on your target — an edge node, an accelerator, or a server — before performance becomes a surprise.
A Few Places to Start
The categories most people arrive looking for.
Safety & PPE
Hard hats, hi-vis, and exclusion-zone occupancy — the models behind most site compliance monitoring.
Forklift & Industrial
Plant, forklifts and industrial vehicles in yards, warehouses and laydown areas.
Wildlife
Species presence and movement for environmental monitoring, offsets and remote-site obligations.
Barcode & QR
Reading codes off assets, packaging and plant so an image becomes an identifier.
Fire & Smoke
Early visual detection of fire and smoke on remote sites, well before anybody is close enough to call it in.
Traffic
Vehicle counting and classification for gates, haul roads, car parks and public streets.
Curated, Not Exhaustive
Evaluation is the slow part
Finding a model is easy. Working out which of forty candidates runs on your hardware, in your format, under a licence you can actually use, is the part that eats a month.
Formats decided early
A model that only ships as a training checkpoint is a different project to one with an ONNX export. Knowing that on day one changes what you shortlist.
Licence visible up front
Licence terms are listed alongside the architecture, so a commercial blocker surfaces before the pilot rather than after it.
Built from deployment experience
The library is curated from what we actually run on nodes for customers — biased toward models that survive contact with a real site.
A path onto real hardware
Anything you shortlist can be deployed to a Vision Node 301, which hot-swaps detection models from the browser. Evaluation and production use the same shape.
Bring your own
If nothing in the library fits, we train and deploy a custom model against your imagery — the zoo is the starting point, not the ceiling.
A Model Is an Input. A Site Is the Output.
Customers are not shopping for models. They are trying to answer questions about a site — is the exclusion zone clear, does the gate count match the induction list, has that stockpile moved since Friday.
Those are model questions underneath. But the thing people actually want is a live operational picture of the site, where the rules that matter are written once and enforced continuously.
Put a Model to Work on Your Site
See live occupancy, queues and crowd flow — and how they’d work on your site.
