Table 1 Results of the state-of-the-art methods. None of the comparison methods (the first seven rows) employ post-processing technique to optimize the segmentation results.
From: Transformer based 3D tooth segmentation via point cloud region partition
Method | Parameter | Inference | OAcc(%) | mIoU(%) | ||||
|---|---|---|---|---|---|---|---|---|
Size(M) | Time(ms) | Maxillary | Mandible | All | Maxillary | Mandible | All | |
PointNet32 | 3.531 | 50.960 | 60.091 | 68.005 | 64.423 | 27.512 | 49.737 | 40.055 |
PointNet++20 | 1.756 | 54.968 | 91.659 | 91.807 | 91.740 | 76.803 | 72.930 | 74.617 |
PointConv22 | 12.779 | 119.254 | 93.389 | 94.496 | 93.995 | 80.717 | 81.149 | 80.961 |
DGCNN50 | 1.462 | 141.980 | 94.905 | 96.108 | 95.564 | 85.346 | 89.939 | 87.938 |
PCT16 | 2.482 | 123.768 | 96.538 | 96.139 | 96.320 | 88.801 | 89.090 | 88.964 |
PVT14 | 2.690 | 127.880 | 94.718 | 94.523 | 94.612 | 84.292 | 85.883 | 85.190 |
PointMLP52 | 15.296 | 114.900 | 95.999 | 95.877 | 95.932 | 87.561 | 88.864 | 88.297 |
PointRegion w/o post-process | 2.490 | 90.288 | 96.866 | 96.729 | 96.791 | 90.739 | 91.614 | 91.233 |
PointRegion with post-process | – | – | 97.336 | 96.921 | 97.109 | 92.775 | 92.308 | 92.511 |
