Three notebooks with code and explanations: how the synthetic dataset is generated, how the model is trained and how it infers from a new photograph.
How the synthetic bone dataset is generated (Blender/BlenderProc), how geometric features are extracted from each render and how the table that trains the model is assembled.
Open notebook → 2From dataset to model: hybrid architecture (geometry + CNN), why three seeds and balanced accuracy rather than raw accuracy, and what goes wrong along the way.
Open notebook → 3What happens when you upload a photo: segmentation, features, embedding, a two-stage cascade and the final prediction, step by step.
Open notebook →