Explainable Dataset for Ground Based Drones in Urban Environments

Recent interest in Explainable Artificial Intelligence has led to an explosion in post-hoc explanation methods for Deep Learning aimed at image understanding. However, there has been limited emphasis on the validation of these explainers. Without a means to effectively evaluate these explainers they undermine the core concept of explainability, risk misleading development and exacerbate the dangers of deploying Deep Networks in the wild. Here, we introduce the first dataset aimed at the development and validation of explainable scene and object recognition for autonomous ground based vehicles. This work aims at accelerating the deployment of Autonomous Driving systems.