SegNeXt: Rethinking Convolutional Attention Design for Semantic Segmentation - 42Papers