Semantic Image Fuzzing of AI Perception Systems

This repository contains code and high level algorithm descriptions for the paper Semantic Image Fuzzing of AI Perception Systems (ICSE '22).

Setting up the codebase

1. Clone this repository
2. Initialize and update the git submodules in the root of the repository. This sets up the Cityscapes scripts dependency. This can be done by:

git submodule init
git submodule update

3. Set up the Python dependencies in requirements.txt

pip3 install -r requirements.txt

4. Set up the environment variables in init.bash.
5. Download the Cityscapes dataset. See this document for more information.
6. Each time you start a new shell to run the tester, run the following to set up the environment variables.

. init.bash

6. Set up the Systems Under Test (SUTs) that you intend to test with the system. For information on the five SUTs used in the study, please see the study data README. This is not needed if you are only trying to recreate the mutated images from the study.

Recreating the Tests from the Study

For information on how to recreate the tests from the study, please see this README.