Start by defining your use case: objects you want to detect and the real-world scene where detections happen.
We use synthetic data as a replacement for
expensive and noisy real data. Learn more about synthetic data here.
1. Pick a category
Start by choosing a category of objects. Check our public catalogue and leverage all existing categories.
2. Create your catalogue
Next, define your private catalogue by selecting specific objects from a category. Spot your objects visually or use the search functionality.
3. Define the scene
Generate a synthetic dataset tailored to your real-life scenario. Choose from a range of pre-defined scenes or customise your own.
Once the dataset is ready, you will find it under your Datasets collection. It is then time to add and customise a model.
We use Neural Networks that are customised for your specific Computer Vision use case and trained with your custom dataset. Learn more about our technology here.
1. Input your business constraints
Choose how fast or big your model should be. We automatically select the best model that fits your business constraints.
2. Select a deployment
Tell us upfront how you would like your model to be made available to you. Work with the deployment of your choice.
3. Customise the
Select post-processing options to convert the detections to your desired business intelligence.
Going to Production
Validate, deploy and monitor your model
Finally, it is time to validate the model
before deploying it in the real world. Afterwards, keeping an eye on the model is easy.
We validate the model once it is trained to ensure high standards of quality and robustness. Learn more about our approach here.
1. Validate the performance
Upload real images and check how the model performs. Unhappy? Resume the model development for further improvements.
2. Ready to deploy
Manage cloud or local deployments with a click of a button. You can export your model to a variety of industry standard formats.
3. Post-deployment maintenance
Get real-time performance statistics once deployed. Step in whenever the model needs an update.