SeedSeg model weights

ONNX weights for SeedSeg, which detects implanted gold fiducial markers in T1-weighted prostate MRI. The SeedSeg browser application and the Neurodesk OpenRecon container load their models from this repository.

Code and usage: https://github.com/astewartau/seedseg

Files

Path	Contents
v2/seedseg-v2.json	Preprocessing and inference settings for the v2 networks (patch size, spacing, normalisation) and file checksums
v2/browser/seedseg-v2_fold{0-4}.wg16.onnx	v2: five nnU-Net networks (one per cross-validation fold), fp16-stored weights, graph adapted for WebGPU; used by the browser application
v2/fp32/seedseg-v2_fold{0-4}.onnx	v2: the same five networks in full precision; used by the OpenRecon container
v1/seedseg-model-seed{42,123,456,789}.onnx	v1: the previous 4-network 3D U-Net

The v2 networks take one channel (N4-corrected T1-weighted image, z-score normalised) and output three classes: background, marker, prostate.