BrainTumor-MajorProject REL-002 — model weights
Frozen inference weights for the BrainTumor-MajorProject REL-002 engineering release. Engineering prototype — not a clinical diagnostic device. Never use for diagnosis, triage, or therapy planning.
Files
| File | Model | SHA256 | Size |
|---|---|---|---|
CLS-001-best.pt |
classifier (ConvNeXt-Tiny) | 451e4fc4b12446764262c159253e7550b5a08db49bb2956059f71613e268095e |
111,358,647 |
SEG-001-best.pt |
segmenter (UNet, default) | ce29df5e4225ed8e7aa0fb29599e53c6abd7a9e27d5bbb2d7f35a3332a2d1b80 |
124,231,879 |
ROB-001-best.pt |
segmenter (robustness variant, documented alternative) | 36091d63cea9af94c568ee3eaf9895658c7eae9a7179cf1fee6c397f1ad82620 |
372,593,293 |
Hashes match docs/release_manifest.md in the source release.
ROB-001 trades clean performance for noise performance; SEG-001 stays default.
Use
git clone https://github.com/NeutronZero/BrainTumor-MajorProject.git
cd BrainTumor-MajorProject
pip install -r requirements.lock
pip install huggingface-hub
python scripts/fetch_models.py # downloads + SHA256-verifies into checkpoints/
python scripts/launch.py check