MVTec Software GmbH

Deep learning classification with MVTec HALCON – 03 - Evaluate a trained classifier

Please note, that the classification approach shown here is legacy from HALCON 19.11 on. With HALCON 19.11, classification is done using the same operators and workflow as the deep learning methods object detection and semantic segmentation. Get more information here.

In the last two video tutorials (part 1 and part 2), you learned how to prepare your data and perform the training of a deep-learning-based classifier with MVTec HALCON.

In this video, we check out different methods that can help you evaluate this trained classifier, like confusion matrices and heatmaps. Furthermore, we use a procedure that can display, for example, all falsely classified images, which can help you evaluate limitations of your classifier. Lastly, different evaluation measures can be computed, like the F-score, the Top-K-error, precision, and recall.

You can download the HDevelop example used in this video below.