Evaluation of Vision Transformers for Traffic Sign Classification

Table 3

Hyperparameters for ViT and its variants.

Hyperparameter	Value
Number of classes	Refer to Table 1
Image patch size	32
Output dimension of the encoder	1024
Number of Transformer blocks	6
Number of heads in multihead attention layer	16
Dimension of the MLP layer	2048
Dropout rate	0.1
Embedding dropout rate	0.1