ToothInstanceNet

Welcome to the repository of our solution to the 3DTeethLand challenge.

Install

First, setup a Conda environment.

conda create -n 3dteethland python=3.10
conda activate 3dteethland

Then, install the Pip requirements.

pip install -r requirements.txt

Lastly, compile and install the Cuda kernels.

pip install -v -e .

Inference

Please specify the root directory where your scans are stored using the root keyword in teethland/config/config.yaml. The file names of the scans are expected to contain whether they are of the lower or upper jaw as STEM_(lower|upper).(stl|ply|obj). Please rename the scans using this format to get the correct FDI labels.

Then, download the checkpoints from here and specify their paths using the checkpoint_path keywords in teethland/config/config.yaml.

Finally, run the model with

python infer.py landmarks --devices DEVICES

where DEVICES can be set to use multiple GPUs for inference. The tooth instance segmentations will be saved next to the scan as STEM_(lower|upper).json and the detected landmarks will be saved next to the scan as STEM_(lower|upper)__kpt.json.