lmvseg-poseidon3d

🦷 LMVSegRNN and Poseidon3D: Addressing Challenging Teeth Segmentation Cases in 3D Dental Surface Orthodontic Scans

   

[Tibor Kubik](https://scholar.google.com/citations?user=Zb6MSKcAAAAJ) and [Michal Spanel](https://scholar.google.com/citations?hl=sk&user=75XIbgQAAAAJ) 

🔥 News and Todo

• 💻 June 2025: Preparation of this GitHub repository and data release.

Abstract

The segmentation of teeth in 3D dental scans is difficult due to variations in teeth shapes, misalignments, occlusions, or the present dental appliances. Existing methods consistently adhere to geometric representations, omitting the perceptual aspects of the inputs. In addition, current works often lack evaluation on anatomically complex cases due to the unavailability of such datasets. We present a projection-based approach towards accurate teeth segmentation that operates in a detect-and-segment manner locally on each tooth in a multi-view fashion. Information is spatially correlated via recurrent units. We show that a projection-based framework can precisely segment teeth in cases with anatomical anomalies with negligible information loss. It outperforms point-based, edge-based, and Graph Cut-based geometric approaches, achieving an average weighted IoU score of 0.97122 ± 0.038 and a Hausdorff distance at 95 percentile of 0.49012 ± 0.571 mm. We also release Poseidon’s Teeth 3D (Poseidon3D), a novel dataset of real orthodontic cases with various dental anomalies like teeth crowding and missing teeth.

Accessing Data

The dataset is available for download at Zenodo. It contains 3D dental scans of orthodontic cases, including various anomalies such as teeth crowding and missing teeth. The dataset is intended for research purposes and can be used to train and evaluate segmentation models.

Parsing the Dataset

tbd: add description

---

Requirements

The code was tested on

• Ubuntu 24.04
• Python 3.12
• PyTorch 2.7.0
• 1 NVIDIA GPU with CUDA version 11.8 (the method is not memory heavy, at least when using 256 embeddings, so any gpu with at least 8GB will work).

Setup an environment

conda create -n lmvsegrnn python==3.12
conda activate lmvsegrnn

Install PyTorch

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

Other Dependencies

pip install -r requirements.txt

Note: to avoid dependency conflicts, please make sure to use the exact package versions specified in requirements.txt.

Data Preparation and Training

todo: describe how to generate synchronized embeddings and run training loop

🔗 BibTeX

If you find this work useful for your research and applications, please cite using this BibTeX:

@article{kubik24poseidon3d,
author = {Kubík, Tibor and Španěl, Michal},
title = {LMVSegRNN and Poseidon3D: Addressing Challenging Teeth Segmentation Cases in 3D Dental Surface Orthodontic Scans},
journal = {Bioengineering},
volume = {11},
year = {2024},
number = {10},
url = {https://www.mdpi.com/2306-5354/11/10/1014},
pubmedid = {39451390},
issn = {2306-5354},
doi = {10.3390/bioengineering11101014}
}