XAI Toolbox for User-Friendly GUI-Based Grad-CAM and LIME

We created a toolbox for easy Grad-CAM and LIME visualizations, helping researchers interpret model predictions without coding expertise.

We designed this MATLAB toolbox for visualizing Grad-CAM and LIME explanations, simplifying interpretability for researchers with limited coding skills. Our toolbox enables users to create visualizations through an intuitive interface, facilitating the interpretation of model predictions without deep programming expertise. It includes essential features for generating Grad-CAM and LIME visuals with customizable parameters. Tutorials and examples are provided to aid users in mastering the toolbox's capabilities.

Here are the features provided by each script:

Single Image Processing: Allows users to process one image at a time using Grad-CAM and LIME techniques (single_gradCAM_LIME_UI.m).

Batch Image Processing: Supports the processing of multiple images in a batch using Grad-CAM and LIME techniques (bulk_gradCAM_LIME_UI.m).

Masked LIME Explanations: Generates LIME explanations with masked regions (masked_LIME_UI.m).

IoU Calculation: Computes the Intersection over Union (IoU) metric between a binarized image and a ground truth image (iou_calculation_UI.m).

Screenshots:

Beyond its utility in interpreting model decisions, our toolbox extends its value to diverse fields like healthcare, agriculture, and remote sensing. These areas benefit greatly from enhanced understanding of model reasoning. Researchers in these domains can leverage our toolbox to gain insights into complex models and improve decision-making processes. With user-friendly tools and educational resources, our aim is to democratize the use of interpretability techniques, fostering broader adoption and innovation across scientific disciplines.