Abstract: Despite the success of pixel-based black-box models in most semantic segmentation algorithms for autonomous driving technology, identifying the causes of recognition errors remains ...