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

See here for a more formal introduction to the topic of XAI and visual saliency explanations.

This framework is a part of the Explainable AI Toolkit (XAITK).

The xaitk-saliency package provides saliency algorithms for a wide range of image understanding tasks, including image classification, image similarity, object detection, and reinforcement learning. The current list of supported saliency algorithms can be found here.

This toolkit is intended to help data scientists and developers who want to add visual saliency explanations to their workflow or product. Functionality provided here is both directly accessible for targeted experimentation, and through Strategy and Adapter patterns to allow for modular integration into systems and applications.

Install the latest release via pip:

pip install xaitk-saliency

Some plugins may require additional dependencies in order to be utilized at runtime. Such details are described here.

See here for more installation documentation.

We provide a number of examples based on Jupyter notebooks in the ./examples/ directory to show usage of the xaitk-saliency package in a number of different contexts.

Contributions are welcome! See the CONTRIBUTING.md file for details.

Documentation snapshots for releases as well as the latest master are hosted on ReadTheDocs.

The sphinx-based documentation may also be built locally for the most up-to-date reference:

# Install dependencies
poetry install
# Navigate to the documentation root.
cd docs
# Build the docs.
poetry run make html
# Open in your favorite browser!
firefox _build/html/index.html

This associated project provides a local web-application that provides a demonstration of visual saliency generation in a user-interface. This provides an example of how visual saliency, as generated by this package, can be utilized in a user-interface to facilitate model and results exploration. This tool uses the trame framework.