Low-Resolution Self-Attention for Semantic Segmentation

Semantic segmentation tasks naturally require high-resolution information for pixel-wise segmentation and global context information for class prediction.

BibTex: