Table 1.
Notations and nomenclatures.
| FPS | Farthest Point Sampling |
| VBS | Voxelization-based Sampling |
| VFPS | Voxelized Farthest Point Sampling |
| DNFEB | Double-Neighborhood Feature Extraction Block |
| DGFFM | Double-Granularity Feature Fusion Module |
| AM | Attention Module |
| CA | Channel attention |
| SA | Spatial attention |
| DHL | Double-hinge Loss |
| GT | Ground truth |
| MLP | Multilayer perceptron |
| ReLU | Rectified linear unit activation |
| PE | Position encoding |
| EC | EdgeConv operation |
| AP | Attentive pooling |
|
| |
| F c , Ff | Feature maps after decoding |
| F DGF | Aggregated feature map after DGFFM |
| L, Lsem, Lins, LDHL, Ls, Ld, Lreg | The loss functions |
| C | The number of semantic classes |
| N | The number of points in a point cloud |
| p i | A point in XYZ space |
| f i , ri, hi, fi′, fei | A point vector in feature space |
| K | The parameter of KNN |
| α, β, γ, δs, δd | Parameters for loss functions |
| ∪ | Feature concatenation |
| max[∙] | The maximum value across the inputs |
| IoU[∙, ∙] | IoU of the two entities |
| MLP[∙] | MLP operation with shared parameters |