Authors: Lorenzo Borghi, Luca Lumetti, Francesca Cremonini, Federico Rizzo, Costantino Grana, Luca Lombardo, Federico Bolelli

At a Glance

📄 Download Full Paper (PDF) 🖼️ Download Poster (PDF)

• First public dataset of 200 paired upper/lower dental arches with multi-dimensional occlusal annotations.
• Multi-task benchmark for predicting 5 occlusal traits from raw 3D point clouds.
• Open-source: Dataset, code, and pretrained models available for research.

Why This Matters

Occlusal classification is critical for orthodontic diagnosis, yet lacks standardized 3D datasets. Bits2Bites bridges this gap by providing:

• ✅ Clinically relevant annotations (sagittal, vertical, transverse, midline relationships).
• ✅ Benchmark models using point-based neural networks.
• ✅ Reproducible baselines with ablation studies on multi-task learning and landmark features.

TL;DR: We enable automated orthodontic analysis from raw 3D scans—no manual segmentation required.

Keywords

Intra-oral Scans · Medical Imaging · Dental Occlusion · 3D Point Cloud · Multi-Task Learning

Questions? Contact info@lorenzoborghi.it.