InSiSo ModelBench™ · Cloud and Desktop
Vision models for your machines, without an ML team
Label a few images. ModelBench labels the rest, trains a detector on your own classes, and exports it to the controller on the line. In your browser, or on one PC with no network at all.
Already have an account? Sign in above.
Who it is for
Built for the people who keep machines running.
Machine builders
Add a camera and a detection model to the machine you already sell. Train it on your parts, your fixtures, your lighting, and ship it on the controller your customers already run.
Maintenance engineers
The line has a problem a camera could catch: a missing clip, a wrong orientation, a worn tool. Photograph it, label fifty images, and have a model watching by the end of the week.
ctrlX and automation partners
ModelBench exports straight to ctrlX AUTOMATION controllers, with CPU, NVIDIA and Hailo targets. Resellers provision customers through organizations and redemption codes.
How it works
From fifty photos to a model on the controller.
Annotate
Upload images, name your own classes and draw boxes, polygons or rotated boxes. About fifty labelled images are enough to start.
Auto-annotate
Train once and let the model label the rest of your dataset. Review, fix what it got wrong, train again. Every round is more accurate.
Train
The INSISO detector trains on your classes, on CPU or GPU, with augmentation presets, early stopping and a validation results panel that shows what it learned.
Export to the line
One question, not nine settings: pick where it will run. ONNX Runtime, NVIDIA CUDA/TensorRT or a Hailo accelerator. Every export is checked against the trained model before you get it.
Runs where the machine is
Export once. Deploy on the hardware the line already has.
A model trained in ModelBench runs in InSiSo Visual Detector™ on ctrlX AUTOMATION and on ordinary industrial PCs. The export carries your class names and the exact pre-processing the model expects, so a single file holds everything the target needs.
Any CPU. The default target for ctrlX AUTOMATION apps.
IPCs with an NVIDIA card. A bundle with a worked example, ready for trtexec.
Verified end to end on a Hailo-8 inside a ctrlX CORE X7: 19x faster than the same model on the X7's CPU.
Two ways to run it
Same tool, same models. Pick where the data lives.
ModelBench Cloud
Open it in a browser. Your team shares one organization with roles, invitations and an audit log. Training runs on our CPU and GPU servers, so a laptop is enough.
- Nothing to install
- Organizations, roles and invitations
- CPU or GPU training on our servers
- Nightly backups
ModelBench Desktop
An installer for Windows and macOS. Everything stays on one PC: images, models and the database. Uses the PC's NVIDIA GPU when it has one.
- Works with no network at all
- Data never leaves the machine
- Windows and macOS installers
- Compile Hailo bundles on the PC
Editions
Standard to get started. Pro for accelerators and measurement.
| Feature | Standard | Pro |
|---|---|---|
| Annotate, auto-annotate, train | ✓ | ✓ |
| ONNX Runtime export for ctrlX | ✓ | ✓ |
| NVIDIA CUDA / TensorRT export | — | ✓ |
| Hailo export and compile to .hef | — | ✓ |
| Measurement in 2D and 3D, with CSV / Excel export | — | ✓ |
| Bulk review and generated augmented images | — | ✓ |
A trial includes everything in Pro.
Questions we get asked
Where does my data go?
On ModelBench Cloud every image, annotation and model belongs to your organization and is never visible to another. The service runs on AWS with nightly backups. If the data cannot leave your site, ModelBench Desktop runs on one PC with no network at all.
Do I need a GPU?
No. Training runs on our servers, on CPU or GPU. On the desktop, an NVIDIA card is used when there is one and the CPU otherwise.
What happens after the trial?
A trial gives your organization everything in Pro. When it ends, nothing is deleted and nobody is locked out: you continue on Standard, or upgrade to Pro.
How is it priced?
An annual licence per organization for the edition, and prepaid credits for the compute you use. We will walk you through it on the call.
Request a trial
Tell us what the camera should catch and which controller it will run on. We set up your organization on ModelBench Cloud, or send you the desktop installer, and you get everything in Pro for the trial.
- 01Send the form. We get back to you to set up the trial.
- 02We create your organization and email your invitation.
- 03Upload your first images and train the same day.