Geodesic heatmap-based segmentation-free framework for robust tooth landmark detection

Weijie Liu, Shaojie Zhuang, Yeying Fan, Guangshun Wei, Yuanfeng Zhou

Abstract: Highlights•A two-stage framework detects tooth landmarks without prior segmentation.•Geodesic heatmaps reduce spatial ambiguity and boost localization accuracy.•Clinically inspired augmentation improves robustness for small and crowded teeth.•Confidence-aware loss ensures stable training and precise landmark prediction.•Outperforms prior methods with +4.5% mAP and +6.8% mAR on benchmark datasets.

External IDs:dblp:journals/cg/LiuZFWZ25