Criss-Cross Attention for Semantic Segmentation in pure Pytorch with a faster and more precise implementation.