Image Segmentation
MLX
English
native-mlx
apple-silicon
medical-imaging
mri
brain-tumor
segmentation
research
Instructions to use MedMLX/segresnet-brats-mlx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use MedMLX/segresnet-brats-mlx with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download MedMLX/segresnet-brats-mlx --local-dir segresnet-brats-mlx
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Document verified public runtime and checkpoint scope
Browse files
README.md
CHANGED
|
@@ -33,15 +33,16 @@ MLX implementation by [MedMLX](https://github.com/MedMLX).
|
|
| 33 |
|
| 34 |
## Requirements
|
| 35 |
|
| 36 |
-
Apple Silicon Mac with Metal GPU access, Python 3.
|
| 37 |
The weights are included here; checkpoint conversion is not needed.
|
| 38 |
|
| 39 |
## Quick start
|
| 40 |
|
| 41 |
```bash
|
| 42 |
-
uv venv --python 3.
|
| 43 |
source .venv/bin/activate
|
| 44 |
-
uv pip install "segresnet-mlx @
|
|
|
|
| 45 |
|
| 46 |
hf download MedMLX/segresnet-brats-mlx --local-dir models
|
| 47 |
```
|
|
@@ -75,7 +76,11 @@ labels use the bundle’s BraTS convention: 0 background, 1 tumor core,
|
|
| 75 |
|
| 76 |
## Technical details
|
| 77 |
|
| 78 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 79 |
|
| 80 |
Native whole-volume output compared with the original MONAI bundle; the October 12 renal operator repair also has a full-case BraTS regression.
|
| 81 |
|
|
|
|
| 33 |
|
| 34 |
## Requirements
|
| 35 |
|
| 36 |
+
Apple Silicon Mac with Metal GPU access, Python 3.13 or 3.14 and [uv](https://docs.astral.sh/uv/).
|
| 37 |
The weights are included here; checkpoint conversion is not needed.
|
| 38 |
|
| 39 |
## Quick start
|
| 40 |
|
| 41 |
```bash
|
| 42 |
+
uv venv --python 3.13
|
| 43 |
source .venv/bin/activate
|
| 44 |
+
uv pip install "segresnet-mlx @ https://github.com/MedMLX/segresnet-mlx/releases/download/v0.2.1/segresnet_mlx-0.2.1-py3-none-any.whl"
|
| 45 |
+
uv tool install huggingface-hub
|
| 46 |
|
| 47 |
hf download MedMLX/segresnet-brats-mlx --local-dir models
|
| 48 |
```
|
|
|
|
| 76 |
|
| 77 |
## Technical details
|
| 78 |
|
| 79 |
+
The companion runtime is [v0.2.1](https://github.com/MedMLX/segresnet-mlx/releases/tag/v0.2.1), source `b5cc265cd0426fe8f64b40a8ed9be4c42a599724`, with `medmlx-core==0.2.3`. All 83 tensors here exactly match a fresh CPU host conversion of the pinned BraTS source checkpoint through that installed release's public `convert_state_dict` API. The weights were retained unchanged.
|
| 80 |
+
|
| 81 |
+
The current fresh-wheel suite passed 145 tests with zero skips: 133 ordinary cases and 12 released **renal** checkpoint cases. Those renal results do not qualify this BraTS checkpoint. The full-volume BraTS comparison below is historical and retains its original software revision and one-case scope; no new full-volume BraTS or clinical performance claim is made.
|
| 82 |
+
|
| 83 |
+
Historical native result: **success**. Original-code comparison: **pass**, `minimum_per_label_dice` = `1.0`.
|
| 84 |
|
| 85 |
Native whole-volume output compared with the original MONAI bundle; the October 12 renal operator repair also has a full-case BraTS regression.
|
| 86 |
|