Table 3.
Classification performance on point cloud with different models.
| Methods | ModelNet 10 | ModelNet 40 | |||
|---|---|---|---|---|---|
| Class Accuracy | Instance Accuracy | Class Accuracy | Instance Accuracy | Training | |
| PointNet [6] | - | - | 86.2 | 89.2 | 3–6 h |
| PointNet++ [43] | - | - | - | 91.9 | 20 h |
| Deepsets [100] | - | - | - | 90.0 | - |
| SO-Net [57] | 95.5 | 95.7 | 90.8 | 93.4 | 3 h |
| Dynamic Graph CNN [49] | - | - | - | 92.2 | - |
| PointCNN [48] | - | - | 91.7 | - | - |
| Kd-Net [78] | 93.5 | 94.0 | 88.5 | 91.8 | 120 h |
| 3DContextNet [44] | - | - | - | 91.1 | - |
| MRTNet [46] | - | - | - | 91.7 | - |
| SPLATNet [56] | - | - | 83.7 | 86.4 | - |
| FoldingNet [95] | - | 94.4 | - | 88.4 | - |
| NeuralSampler [55] | - | 95.3 | - | 88.7 | - |