This repository features RARE-UNet a resolution-aware 3D U-Net for adaptive medical segmentation. It uses multi-scale entry blocks and resolution-based routing to dynamically adjust the inference path to input resolution. Combined with consistency-based training, RARE-UNet delivers accurate, efficient segmentation across resolutions.
brain-mri dynamic-routing multi-scale medical-image-segmentation tumor-segmentation 3d-unet adaptive-inference resolution-aware consistency-training hippocampus-segmentation rare-unet
-
Updated
Feb 5, 2026 - Python