Meanwhile more and more users have made their first experiences with AI. On the one hand, they discover that often much more images are needed to train the networks than promised. But there are also problems in case of misclassifications, i.e. understanding why an AI has classified the image as an error. The new Halcon version offers an interesting approach to this. A heatmap shows which parts of an image had a strong influence on the classification. This is at least the first step away from a black box system and towards a white box AI and above all to a much better understanding how the new algorithms really tick.
TeDo Verlag GmbH
