Model-agnostic explanations of black box classifiers for image recognition

This repository contains the source code of my master thesis (Model-agnostic explanations of black box classifiers for image recognition) and the annotated data set used in the experiments.

This work is an extension of LIME (M. T. Ribeiro, S. Singh, and C. Guestrin).

Relevant folders:

• chosen_1000_images contains the 1000 images used in the experiments
• immagini_ritagliate contains the reference areas for each image in chosen_1000_images (i.e. the annotated dataset)
• master thesis includes all the LaTeX files and pdf version of the thesis

Relevant files:

• predizioni_bb.txt contains the predictions of the black box on each of the 1000 images in the test set
• chosen_classes_for_validation.txt specifies the chosen labels from the ILSVRC dataset
• requirements.txt specifies the required Python libraries

Experiments

• Source code to run the experiments can be found under lime/lime/

Jupyter Notebooks

• Under lime/doc/notebooks/ you can find various example notebooks used to generate plots and figures