Siltframe Get a fix →

Pricing

Start with $19. Go as deep as you need.

Test your model yourself, have us find out why it fails, or have us fix it. Every fix is priced before you commit.

Individual

$19one-time

Students, researchers, solo builders

Test your model yourself.

  • Stress-test pack: 2,460 labelled off-road images — dust, night, fog, rain and mud on the lens, 3 strengths each, plus real rain and snow
  • Scoring script: your failure map in about 10 minutes
  • Reference results from 3 AI models to compare against
  • Commercial use allowed (CC BY-SA 4.0)
  • E-mail support
Get the pack — $19 →

Team

from $300per audit

Startups and small robotics teams

We find where and why your model fails.

  • Everything in Individual
  • Your model tested on your own frames (up to 300)
  • All 5 conditions × 3 strengths, cross-checked with independent simulators
  • Root cause: which objects it loses, in which conditions, and what its training data is missing
  • Written report with a fix plan and a fixed price for the fix
  • 30-minute review call · NDA on request · about 10 business days
Book an audit →

Recommended

Pro

from $600per model

Teams shipping a product

We find it — and fix it.

  • Everything in Team
  • Gap data: commercially licensed real data that covers what your model never saw, mapped to your classes
  • Simulated hard cases on your own frames — only where real data doesn’t exist
  • Your model retrained (one model, up to two rounds)
  • Before/after on your real held-out frames, against a control run
  • You keep the new model file, the training data and the recipe
Get a fix →

Enterprise

Custom

Fleets, several models or cameras

Robustness as part of every release.

  • Everything in Pro
  • Several models, cameras or operating sites
  • Regression testing on every model release
  • New conditions built for your environment — scoped up front, we tell you what we can and can’t simulate
  • On-premise option: we run the tools on your servers, your data never leaves
  • Priority support and invoicing
Talk to us →

What exactly do we fix?

The behaviour you see in the field. You get back a new version of your own model — same architecture, a drop-in model file — that handles the condition better, the data that taught it, and a before/after report on your own real frames.

Three typical cases · what you see in the field

Under trees, the robot calls the canopy “sky”.

Why
Its training data was shot in open fields; roads under trees were missing.
What we do
Add real forest-road frames (openly licensed) mapped to its classes, retrain.
Proof
In our case study: 80% of the canopy called sky → 0% on the same frame; real bad-weather scores up 26–35 points on 4 models.

At night it stops seeing the person on the trail.

Why
Almost no dark frames in training, so the person blends into the noise.
What we do
Real night data where it exists; otherwise headlight-and-sensor-noise night simulated on your own frames, labels kept exact.
Proof
Person detection at night measured before and after, on your real night frames if you have them.

Dust or mud on the lens and it loses the trail edge.

Why
The camera never looked through dirt during training.
What we do
Dust and lens mud simulated on your frames at three strengths — conditions that are hard to capture on demand.
Proof
Trail-edge accuracy in each condition, before and after, cross-checked on independent simulators.

you get back

  1. The fixed model — new weights for the same network, ready to deploy.
  2. The training data we added, labelled in your classes.
  3. The report — before/after on your real held-out frames, against a control run.
  4. The recipe — so your team can repeat it on the next model.

what we won’t promise

  • Simulated weather alone rarely fixes real-world failures. In our own tests the right real data did far more — so we use simulation only where it measurably helps, and tell you when it doesn’t.
  • If the audit shows the fix needs data we can’t source or simulate, we say so before you pay for the fix.

Pricing questions

What decides the price of an audit or a fix?

How many frames and classes, how many conditions, and whether we can use openly licensed real data or need to simulate. The audit ends with a fixed price for the fix — no open-ended billing.

Do we have to share our model and data?

For Team and Pro we need either the model or the ability to run it — you can also run our tools yourself and send back scores. Enterprise can run fully on your own servers. An NDA is available for every plan.

Which models do you support?

Semantic segmentation models in PyTorch or ONNX — SegFormer, Mask2Former, DeepLab, OneFormer and similar. Tell us about anything else in the brief.

Can we start with the $19 pack and upgrade later?

Yes. Send us your pack results and we’ll scope the audit from them.