Fix stale CPU epoch timing (~15 min claimed, ~43 min measured)
Corrects Training-Guide.md hardware table and Evaluation-Results.md
config table to match the actual 50-epoch run timing.
Add real 8-class training and test results
Fills in the per-class metrics table with actual results from the 50-epoch
run: Val/Test Mean Tumor Dice 0.7350/0.7387. Points the model download link
to the upcoming v2.0.0 release.
Update all wiki pages to reflect 8-class segmentation model
- Model-Architecture.md: num_classes 4→8, full 8-class listing with WHO categories,
updated parameter table (CPU=2.2M), updated class weights
- Training-Guide.md: --num-classes default 8, training output format for all 7 classes,
test-set evaluation output, updated hardware table with actual timing
- Evaluation-Results.md: dataset size 8,673/1,858/1,860, config table updated
(2.2M params, depth=2, ~15min/epoch), per-class metrics table ready for results
- Dataset-Setup.md: actual folder names (Gliomas/, Meningothelial Tumors/, etc.),
class mapping table with all 7 WHO categories + pseudo-label explanation
Fill evaluation results with real metrics from completed training run (Val Dice 0.7894)
Add Dataset Setup, Model Architecture, Training Guide, Evaluation Results pages; fix Home link