2D TransUNet for prompt-free glioma sub-region segmentation, trained on BraTS2023-GLI (t1c).
Self-contained: predictor.py snapshot-downloads the bundled transunet_src/ (torch-only) and weights.pth at load. Input: 4-ch (t1n,t1c,t2w,t2f), uses t1c. Output {0=bg,1=NCR,2=edema,3=ET} on the native grid.
