What does a Panoptic Perception Model Learn?

Panoptic driving perception is a promising technology for autonomous driving, which aims to provide a comprehensive understanding of the surrounding environment by processing multiple tasks together. Given the constrained resource on autonomous vehicles, it is essential to develop panoptic perception models with high precision and low resource consumption to assist autonomous driving in real-time. To achieve this goal, it is urgent to understand what a panoptic perception model learns, and get insights to efficiently improve the model design. In this work, we focus on analyzing the explainability of panoptic perception models. Specifically, we use YOLOP as an example, analyze the model explainability and propose several insights for designing the panoptic perception model in the future. To facilitate further research, the source codes are released at https://github.com/lori930/panoptic_visualization.