Research Article
DANC-Net: Dual-Attention and Negative Constraint Network for Point Cloud Classification
Table 3
Classification performance on ModelNet40. (The top three accuracies are highlighted by bold, underline, and italic.)
| | Methods | Input | Points (k) | mA (%) | OA (%) |
| | VoxNet (IROS 2015) | Points | 1 | 83.0 | 85.9 | | 3DShapeNets (CVPR 2016) | Points | 1 | 77.3 | 84.7 | | PointNet (CVPR 2017) | Points | 1 | 86.2 | 89.2 | | PointNet++ (CVPR 2017) | Points + normal | 5 | 87.9 | 91.9 | | Kd-Net (ICCV 2017) | Points | 32 | 88.5 | 91.8 | | PointCNN (NeurIPS 2018) | Points + normal | 1 | 88.1 | 92.2 | | DGCNN (TOG 2019) | Points | 1 | 90.2 | 92.3 | | A-CNN (CVPR 2019) | Points + normal | 1 | 89.9 | 92.2 | | SRN-PointNet++ (CVPR 2019) | Points | 1 | — | 91.5 | | PointHop (IEEE T MULTIMEDIA 2020) | Points | 1 | 84.4 | 89.1 | | DGANet (remote sensing 2021) | Points | 1 | 89.4 | 92.3 | | MRFGAT (INT J ANTENN PROPAG 2021) | Points | 1 | 90.1 | 92.5 | | Ours | Points + normal | 1 | 90.5 | 92.9 |
|
|
Input and points represent the input data type and the number of sampling points, respectively.
|
