Getting Started with xaitk-saliency

The xaitk-saliency package is an explainable AI (XAI) framework and toolkit for visual saliency algorithm interfaces and implementations, built for analytics and autonomy applications.

Saliency maps provide a visual form of explanation by (typically) overlaying a heatmap on the input, highlighting regions that the AI model deems “important” for its predictions.

Saliency methods are typically categorized as:

  • White-box: Requiring access to the internal state of the AI model.

  • Black-box: Operating without any knowledge of the model internals.

Because black-box methods are often better suited to testing and evaluation (T&E) scenarios—where internal model access may be restricted—xaitk-saliency prioritizes black-box saliency techniques.

Example: A First Look at xaitk-saliency

This associated project features a local web application that demonstrates visual saliency generation through a user interface (UI). It provides an example of how saliency maps produced byxaitk-saliencycan be integrated into a UI to support model prediction exploration and reasoning. The application is built using the trame framework.

Next Steps

To learn more about xaitk-saliency, read the Overview or dive right into a Tutorial.