🦷 Bits2Bites

This repository is a customized fork of Pointcept tailored for multi-task classification on 3D dental point cloud data. It includes several major extensions and changes that may affect the default behavior of Pointcept. We recommend using this fork only to explore our implementation.

📌 What's in this Fork

We implemented:

• ✅ Custom model: MultiTaskClassifier with multiple classification heads → Located in: pointcept/models/multi_task_classifier

• ✅ Custom configs → Located in: configs/

• ✅ Custom data loader for the Bits2Bites dataset → Located in: pointcept/datasets/ → Dataset info: Bits2Bites Dataset

• ✅ Custom inference scripts → Located in: tools/

• ✅ Custom evaluation hook for multi-task learning → Located in: pointcept/engines/hooks/evaluator

• ⚠️ Additional modifications to core components → This may impact compatibility with upstream Pointcept. Please use this fork only for reproducing our results.

📄 Citation & Paper

This repository supports the methods described in our MICCAI 2025 paper: 📝 "Bits2Bites: Intra-oral Scans Occlusal Classification" Authors: L. Borghi, L. Lumetti, et al. 📍 To be presented at ODIN - MICCAI 2025 – South Korea 📄 Paper (link coming soon)

📦 Getting Started

Please refer to Pointcept's installation guide and ensure you clone this repository instead of the original.