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