Table 2 Semantic segmentation results on the Semantic3D dataset
From: Pseudo-labelling-aided semantic segmentation on sparsely annotated 3D point clouds
Method | Overall accuracy (%) | Average F-score (%) | Per-class F-scores(%) | |||||
|---|---|---|---|---|---|---|---|---|
| Â | Â | Â | Terrain | Vegetation | Building | Hardscape | Artefacts | Cars |
Pointwise [1] | 49.5 | 29.3 | 87.2 | 13.6 | 59.0 | 12.9 | 2.2 | 1.0 |
CRF-reg [22] | 65.0 | 42.8 | 96.2 | 32.0 | 73.5 | 28.4 | 24.8 | 1.8 |
Seg-aided [17] | 74.4 | 43.1 | 95.7 | 28.1 | 83.0 | 24.7 | 22.9 | 4.0 |
Supervised baseline | 86.2 | 51.9 | 97.1 | 36.5 | 91.7 | 66.3 | 6.7 | 13.0 |
Ours no kdist | 88.1 | 56.9 | 97.7 | 51.8 | 92.7 | 54.9 | 4.8 | 39.2 |
Ours with kdist | 95.6 | 66.7 | 94.2 | 61.2 | 97.7 | 84.6 | 9.0 | 53.3 |