wenglor sensoric - AI Lab - How can I evaluate AI model performance in the AI Lab?

In this tutorial, you'll learn how to review the performance of a trained AI model in AI Lab and understand the key metrics that help you evaluate its accuracy and reliability.

To demonstrate the workflow, we'll use one of the example datasets, as it already contains pre-trained models. Open the All Models page, select the model you want to analyze, and navigate to the Model Review section.

Here, you'll find comprehensive performance metrics, including the balanced accuracy, the number of incorrect predictions, inference time, and detailed precision and recall values for every class. These metrics provide valuable insights into how well your AI model performs.

Next, examine the Confusion Matrix, which compares the ground truth labels with the model's predictions. This visualization makes it easy to identify which classes are classified correctly and where misclassifications occur.

You can also review examples of both correct and incorrect predictions to better understand the model's behavior.

The Info tab provides additional details about the model and dataset, including the image resolution, dataset size, and the number of classes used during training.

To investigate errors more closely, click Review Wrong Predictions. This opens every misclassified image in full size, allowing you to inspect the results in greater detail.

For even deeper analysis, enable the Heatmap feature. The heatmap highlights the image regions that had the greatest influence on the model's prediction, making the AI's decision process more transparent.

If you discover mislabeled images, you can correct the labels and retrain the model to further improve its performance and accuracy.

Finally, AI Lab allows you to generate a printable model performance report containing all evaluation metrics and review information—ideal for documentation, validation, or sharing results with your team.