Table 1 Comparison of related methods.
From: Feature-guided multilayer encoding–decoding network for segmentation for 3D intraoral scan data
Method | Technical description | Limitations |
|---|---|---|
PointNet13 | First end-to-end point cloud segmentation network | Fails to consider local features and context |
MeshSegNet12 | Uses mesh cells to represent local structure | Low training/inference efficiency |
PointMLP14 | Residual MLP network for point clouds | Sensitive to point cloud density variations |
MBESegNet15 | Multi-scale bidirectional enhancement | Limited to 160 samples; works best on normal teeth |
PointeNet16 | Lightweight architecture | Poor coarse/fine feature fusion |
Our method | Hierarchical feature-guided normalization | Weak generalization for long-tail data |
