CrossViT: Cross-Attention Multi-Scale Vision Transformer for Image Classification (Paper Review)