Fetching the paper…

PSViT: Better Vision Transformer via Token Pooling and Attention Sharing · Around