End-to-End 3D Tooth Landmark Detection
This work is the third solution of MICCAI 2024 Challenge 3DTeethLand.
It is also an extended application of KeypointDETR for tooth landmark detection.
- First, individual teeth are cropped based on the segmentation ground truth provided by Teeth3DS. scripts/seg_to_single/segment_patch
- Assign ground truth landmarks to each tooth patch. scripts/seg_to_single/tooth_landmarks
- Generate the geodesic distance maps for all landmarks on each tooth patch. scripts/geodesic_distance
python train.py
python test.py
- First, the 3D tooth segmentation method is executed to crop tooth patches from the segmentation results.
- Run TL-DETR to detect landmarks on each tooth and map back to the original jaw models.
@inproceedings{jin2024keypointdetr,
title={KeypointDETR: an end-to-end 3d keypoint detector},
author={Jin, Hairong and Shen, Yuefan and Lou, Jianwen and Zhou, Kun and Zheng, Youyi},
booktitle={ECCV},
year={2024}
}