wenglor sensoric - AI Lab - How can I train AI models in the AI Lab?

In this tutorial, you'll learn how to train an AI model in AI Lab and configure the most important settings to achieve the right balance between performance, accuracy, and deployment requirements.

Once your dataset is fully labeled, you're ready to begin training. Click Train Model to open the Model Configuration page.

Start by assigning a name to your model and selecting the target device for deployment, such as a B60 or an MVC.

Next, choose the model architecture that best fits your application. Select Latency for faster inference or Accuracy when achieving the highest possible classification precision is your priority.

You can also define the image resolution used during training. AI Lab only displays resolutions that are compatible with your selected hardware. In general, lower resolutions reduce both training time and inference time, while higher resolutions can improve classification performance when fine image details are important. A good rule of thumb is to choose the lowest resolution that still provides reliable results.

Use the Resized Image Preview to see how your images will appear at the selected resolution before starting the training process.

AI Lab also provides estimates for the training duration, inference time, and the required AI credits, helping you plan your project efficiently.

If your application is not sensitive to object rotation, you can enable Rotation to automatically generate additional training variations and improve model robustness. For rotation-dependent applications, this option should remain disabled.

On the right side of the configuration page, you'll find a dataset summary, including the number of classes, annotation coverage, and overall dataset usage. Reviewing these statistics helps ensure your dataset is ready for training.

Once you've verified all settings, click Train Model to begin the training process.

That's it! Your first AI Lab model is now being trained and will soon be ready for evaluation and deployment.