EvoAttack: An Evolutionary Search-Based Adversarial Attack for Object Detection Models

State-of-the-art deep neural networks in image classification, recognition, and detection tasks are increasingly being used in a range of real-world applications. Applications include those that are safety critical, where the failure of the system may cause serious harm, injuries, or even deaths. Adversarial examples are expected inputs that are maliciously modified such that the machine learning models fail to classify them correctly. While a number of evolutionary search-based approaches have been developed to generate adversarial examples against image classification problems, evolutionary search-based attacks against object detection algorithms remain unexplored. This paper explores how evolutionary search-based techniques can be used as a black-box, model- and data- agnostic approach to attack state-of-the-art object detection algorithms (e.g., RetinaNet and Faster R-CNN). A proof-of-concept implementation is provided to demonstrate how evolutionary search can generate adversarial examples that existing models fail to correctly process. We applied our approach to benchmark datasets, Microsoft COCO and Waymo Open Dataset, applying minor perturbations to generate adversarial examples that prevented correct model detections and classifications on areas of interest.