How-To Guides
This section provides practical, task-oriented examples demonstrating how to apply XAITK-Saliency
across a range of domains—from image classification and object detection to explainability in
similarity scoring and reinforcement learning. These guides are implemented as Jupyter notebooks
located in the docs/examples/ folder of the repository.
Each notebook walks through how to accomplish a specific task using XAITK-Saliency’s tools and APIs. For further detail on these APIs, refer to the Reference topics Implementations and Interfaces.
Image Classification
Classifying COVID-19 in Chest X-rays
Interpret model predictions on chest X-ray images using saliency maps.
Generating Saliency for MNIST with scikit-learn
Apply saliency techniques to scikit-learn classifiers on the MNIST dataset.
Comparing Saliency Across Models
Visualize and compare explanations from different classifiers.
Object Detection
Generating Detection Saliency via Serialization
Produce saliency maps for serialized detections in COCO format.
Applying Occlusion Saliency in VIAME
Perform occlusion-based saliency analysis for classifying marine species with the VIAME toolkit.
Advanced Saliency Techniques
Swapping Saliency Techniques in a Classification Pipeline
Modularize and switch between saliency methods in an application workflow.
Estimating Saliency with Multi-Color RISE
Generate saliency maps with uncertainty quantification using MC-RISE.
Applying Radial Perturbations to Images
Analyze model sensitivity by applying radial perturbations to input images.
Generating Superpixel-Based Saliency Maps
Use superpixels as spatial units for interpretable saliency mapping.
Other Applications
Applying Saliency to Atari Game Agents
Visualize saliency in deep reinforcement learning agents trained on Atari games.
Explaining Similarity Scores with Saliency
Use saliency maps to interpret similarity scoring systems.