Adaptive patching of high-resolution images based on image details can drastically reduce the number of patches fed to vision transformer models, while maintaining or improving segmentation quality.
