Explainable Artificial Intelligence (XAI) for Methods Working on Point Cloud Data: A Survey

Abstract

In this work, we provide an overview of the XAI (Explainable Artificial Intelligence) works related to explaining the methods working on point cloud (PC) data. The recent decade has seen a surge in artificial intelligence (AI) and machine learning (ML) algorithms finding applications in various fields dealing with a wide variety of data types such as image and text data. Point cloud data is one of these datatypes that has seen an upward trend in the use of AI/ML algorithms. However, not all these AI algorithms are “white box” models that can be understood by humans easily. Many of them are hard to interpret or understand and thus, require methods to provide explanations for the decision-making process. These methods that attempt to provide explanations or insights into the working of AI models working on various datatypes are grouped under XAI. Even though the use of datatypes such as point clouds for AI models has seen an upward trajectory, we see a lack of survey works documenting the developments in the corresponding XAI field. This issue is addressed through our contribution. We classify the literature based on different criteria such as XAI mechanism used, AI models, their tasks, type of model learning and the type of point cloud data taken into consideration. This can help readers identify works that address specific tasks and obtain corresponding details easily. We also provide useful insights regarding the surveyed papers that highlight interesting aspects of the surveyed literature.