We validate CvT by conducting extensive experiments, showing that this approach achieves 
state-of-the-art performance over other Vision Transformers and ResNets on ImageNet-1k, with fewer parameters and lower FLOPs.