Image Segmentation
MLX
English
native-mlx
apple-silicon
medical-imaging
ct
liver
segmentation
nnunet
research
Instructions to use MedMLX/nnunet-liver-mlx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use MedMLX/nnunet-liver-mlx with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download MedMLX/nnunet-liver-mlx --local-dir nnunet-liver-mlx
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Download reference-comparison.json from MedMLX/nnunet-liver-mlx: direct link, hf CLI and curl.
- Browser
- Download file 5.73 kB
-
https://huggingface.co/MedMLX/nnunet-liver-mlx/resolve/main/reference-comparison.json
- Command line
-
hf download hf://MedMLX/nnunet-liver-mlx/reference-comparison.json
-
curl -L -o reference-comparison.json https://huggingface.co/MedMLX/nnunet-liver-mlx/resolve/main/reference-comparison.json
5.73 kB
| { | |
| "schema_version": "medmlx.reference.v1", | |
| "model": "nnunet-liver", | |
| "date": "2026-10-12", | |
| "case": { | |
| "dataset": "AMOS22 (validation)", | |
| "case_id": "amos_0008", | |
| "input_sha256": "e3ed0f4d7b1fcd1b785ffdb233750cc35eff9714fa724b07122dd625a9c85461", | |
| "inputs": [ | |
| { | |
| "role": "CT", | |
| "path": "<repositories>/radnn/work/nnunet-next-20261001/public-ct/amos_0008/image.nii.gz", | |
| "sha256": "e3ed0f4d7b1fcd1b785ffdb233750cc35eff9714fa724b07122dd625a9c85461", | |
| "dataset": "AMOS22 (validation)", | |
| "case_id": "amos_0008", | |
| "derivation": "staged AMOS CT; HU and geometry as released", | |
| "resolved_path": "<repositories>/research/radnn/work/nnunet-next-20261001/public-ct/amos_0008/image.nii.gz", | |
| "verified_sha256": "e3ed0f4d7b1fcd1b785ffdb233750cc35eff9714fa724b07122dd625a9c85461" | |
| } | |
| ] | |
| }, | |
| "original": { | |
| "code": "nnunetv2", | |
| "version_or_commit": "2.8.1", | |
| "upstream_commit": "468cf803df9b267150ae2b6c0c59b8ac84f16227", | |
| "precision": "FP32 network; upstream FP16 Gaussian sliding-window accumulation; CPU autocast OFF", | |
| "command": "<paper-repo>/work/reference/nnunet-brats/.venv/bin/python benchmarks/reference/nnunet-liver.py --threads 3", | |
| "folds": [ | |
| 0 | |
| ], | |
| "configuration": "3d_fullres", | |
| "trainer": "nnUNetTrainer", | |
| "use_mirroring": false, | |
| "mirroring_axes": [], | |
| "tile_step_size": 0.5, | |
| "device": "cpu", | |
| "checkpoint_path": "<repositories>/research/radnn/checkpoints/nnunet/nnunet_liver/fold_0/checkpoint_final.pth", | |
| "checkpoint_sha256": "9170ca9826668954256947f1aa6fb06faede02301edaf9db6446fd552ffefda9", | |
| "packages": { | |
| "nnunetv2": "2.8.1", | |
| "dynamic-network-architectures": "0.4.4", | |
| "torch": "2.13.0", | |
| "numpy": "2.5.2", | |
| "scipy": "1.18.1", | |
| "scikit-image": "0.26.0", | |
| "SimpleITK": "2.5.6", | |
| "nibabel": "5.4.2" | |
| }, | |
| "inference_wall_s": 5010.051914084004, | |
| "output_path": "work/reference/nnunet-liver/out/mask.nii.gz", | |
| "output_sha256": "459e23806fd3887c0835ba4c3766cf1383d24f3ecb991600660d3306dd6ee9ca" | |
| }, | |
| "mlx": { | |
| "run_record": "results/runs/nnunet-liver.json", | |
| "run_record_sha256": "fe645df577e03b19e43d87ef49a50d81df61c8315e8b85e0fe87e0f0dd4a5231", | |
| "output_path": "<paper-repo>/work/runs/nnunet-liver/out/mask.nii.gz", | |
| "output_sha256": "dc3dd8177e62f6250ba1ba10576dbb245d86aeb15db3a1a97e50ec5b67dbbb12", | |
| "use_mirroring": false, | |
| "mirroring_axes": [] | |
| }, | |
| "comparison": { | |
| "metric": "minimum_per_label_dice", | |
| "value": 0.9999366324587556, | |
| "per_label": { | |
| "0": { | |
| "dice": 0.9999996624633793, | |
| "original_voxels": 59252850, | |
| "mlx_voxels": 59252810 | |
| }, | |
| "1": { | |
| "dice": 0.9999366324587556, | |
| "original_voxels": 315599, | |
| "mlx_voxels": 315639 | |
| }, | |
| "2": { | |
| "dice": 1.0, | |
| "original_voxels": 3775, | |
| "mlx_voxels": 3775 | |
| } | |
| }, | |
| "threshold": 0.99, | |
| "result": "pass", | |
| "differing_voxel_count": 40, | |
| "labels_compared": "Every label present in either output, including background", | |
| "geometry": { | |
| "original": { | |
| "shape": [ | |
| 768, | |
| 768, | |
| 101 | |
| ], | |
| "affine": [ | |
| [ | |
| -0.6158854365348816, | |
| -0.0, | |
| 0.0, | |
| 223.5 | |
| ], | |
| [ | |
| -0.0, | |
| 0.6158854365348816, | |
| -0.0, | |
| -404.8841247558594 | |
| ], | |
| [ | |
| 0.0, | |
| 0.0, | |
| 5.0, | |
| 5.49993896484375 | |
| ], | |
| [ | |
| 0.0, | |
| 0.0, | |
| 0.0, | |
| 1.0 | |
| ] | |
| ], | |
| "spacing": [ | |
| 0.6158854365348816, | |
| 0.6158854365348816, | |
| 5.0 | |
| ], | |
| "orientation": [ | |
| "L", | |
| "A", | |
| "S" | |
| ], | |
| "qform_code": 1, | |
| "sform_code": 1 | |
| }, | |
| "mlx": { | |
| "shape": [ | |
| 768, | |
| 768, | |
| 101 | |
| ], | |
| "affine": [ | |
| [ | |
| -0.6158854365348816, | |
| -0.0, | |
| 0.0, | |
| 223.5 | |
| ], | |
| [ | |
| -0.0, | |
| 0.6158854365348816, | |
| -0.0, | |
| -404.8841247558594 | |
| ], | |
| [ | |
| 0.0, | |
| 0.0, | |
| 5.0, | |
| 5.49993896484375 | |
| ], | |
| [ | |
| 0.0, | |
| 0.0, | |
| 0.0, | |
| 1.0 | |
| ] | |
| ], | |
| "spacing": [ | |
| 0.6158854365348816, | |
| 0.6158854365348816, | |
| 5.0 | |
| ], | |
| "orientation": [ | |
| "L", | |
| "A", | |
| "S" | |
| ], | |
| "qform_code": 1, | |
| "sform_code": 1 | |
| }, | |
| "input": { | |
| "shape": [ | |
| 768, | |
| 768, | |
| 101 | |
| ], | |
| "affine": [ | |
| [ | |
| -0.6158854365348816, | |
| -0.0, | |
| 0.0, | |
| 223.5 | |
| ], | |
| [ | |
| 0.0, | |
| 0.6158854365348816, | |
| 0.0, | |
| -404.8841247558594 | |
| ], | |
| [ | |
| 0.0, | |
| 0.0, | |
| 5.0, | |
| 5.49993896484375 | |
| ], | |
| [ | |
| 0.0, | |
| 0.0, | |
| 0.0, | |
| 1.0 | |
| ] | |
| ], | |
| "spacing": [ | |
| 0.6158854365348816, | |
| 0.6158854365348816, | |
| 5.0 | |
| ], | |
| "orientation": [ | |
| "L", | |
| "A", | |
| "S" | |
| ], | |
| "qform_code": 1, | |
| "sform_code": 1 | |
| } | |
| } | |
| }, | |
| "notes": "One recorded case; reproducibility comparison, not accuracy or clinical qualification. Relocated RadNN paths are recorded and verified against the original run hashes.", | |
| "failure": null | |
| } | |