To improve the security and robustness of autonomous driving models, this paper presents SMET, a scenariobased metamorphic testing tool for autonomous driving models. The metamorphic relationship is divided into three dimensions (time, space, and event) and demonstrates its effectiveness through case studies in two types of autonomous driving models with different outputs.Experimental results show that this tool can well detect potential defects of the autonomous driving model, and complex scenes are more effective than simple scenes.