Evaluation of Vision Transformers for Traffic Sign Classification

Traffic sign recognition is one of the most important tasks in autonomous driving. Camera-based computer vision techniques have been proposed for this task, and various convolutional neural network structures are used validated with multiple open datasets. Recently, novel Transformer-based models achieved state-of-the-art performance, outperforming networks several tasks. In study, our goal to investigate whether success Vision Transformers can be replicated within traffic area. Based on existing resources, we first extract contribute three classification these datasets, experiment seven five Transformers. We find that not as competitive task. Specifically, there performance gaps up 12.81%, 2.01%, 4.37% German, Indian, Chinese respectively. Furthermore, propose some suggestions improve