BoneAGUNet inference assets

Model checkpoints and atlases used by the ManskeLab/BoneAGUNet Python package.

The supported inputs are already-cropped MCP2 or MCP3 joint image stacks. This repository contains four nnU-Net inference bundles (joint stripping, cortical edge, closed-edge/bone mask, and erosion segmentation) plus MC and PP atlases for MCP2 and MCP3.

The attention models require the Manske Lab multichannel-attention nnU-Net fork. Use the Python package's boneagunet-install command rather than downloading individual checkpoint files; nnU-Net requires the adjacent dataset.json and plans.json metadata.

These models are intended for research use and are not a medical device.