Abstract: Recently, lightweight Vision Transformers (ViTs) demon-strate superior performance and lower latency, compared with lightweight Convolutional Neural Networks (CNNs), on resource-constrained ...