Point Cloud Part Segmentation on ModelNet40
85.97mIoU (instance-average)nD-RoPE (vector attention)
Evaluation results
| Method | Links | |
|---|---|---|
| nD-RoPE (vector attention)Input Points=2048, Backbone=Point Transformer2026.06 | 85.97 | |
| nD-RoPE (vector attention)Input Points=1536, Backbone=Point Transformer2026.06 | 85.76 | |
| nD-RoPE (standard dot-product)Input Points=2048, Backbone=Point Transformer2026.06 | 85.07 | |
| nD-RoPE (vector attention)Input Points=3072, Backbone=Point Transformer2026.06 | 84.92 | |
| nD-RoPE (standard dot-product)Input Points=1536, Backbone=Point Transformer2026.06 | 84.66 | |
| nD-RoPE (standard dot-product)Input Points=3072, Backbone=Point Transformer2026.06 | 83.8 | |
| nD-RoPE (vector attention)Input Points=4096, Backbone=Point Transformer2026.06 | 82.75 | |
| nD-RoPE (vector attention)Input Points=1024, Backbone=Point Transformer2026.06 | 82.65 | |
| Learnable Rel. PEInput Points=2048, Backbone=Point Transformer2026.06 | 82.58 | |
| 3D RoPE-Axial + APEInput Points=2048, Backbone=Point Transformer2026.06 | 81.92 | |
| Learnable Rel. PEInput Points=1536, Backbone=Point Transformer2026.06 | 81.66 | |
| 3D RoPE-Axial + APEInput Points=1536, Backbone=Point Transformer2026.06 | 81.42 | |
| 3D RoPE-MixedInput Points=2048, Backbone=Point Transformer2026.06 | 81.4 | |
| 3D RoPE-Mixed + APEInput Points=2048, Backbone=Point Transformer2026.06 | 81.13 | |
| nD-RoPE (standard dot-product)Input Points=4096, Backbone=Point Transformer2026.06 | 81.06 | |
| 3D RoPE-AxialInput Points=2048, Backbone=Point Transformer2026.06 | 80.98 | |
| 3D RoPE-MixedInput Points=1536, Backbone=Point Transformer2026.06 | 80.91 | |
| nD-RoPE (standard dot-product)Input Points=1024, Backbone=Point Transformer2026.06 | 80.85 | |
| Learnable Rel. PEInput Points=3072, Backbone=Point Transformer2026.06 | 80.82 | |
| 3D RoPE-Axial + APEInput Points=3072, Backbone=Point Transformer2026.06 | 80.48 | |
| 3D RoPE-MixedInput Points=3072, Backbone=Point Transformer2026.06 | 80.15 | |
| 3D RoPE-AxialInput Points=1536, Backbone=Point Transformer2026.06 | 80.14 | |
| 3D RoPE-Mixed + APEInput Points=1536, Backbone=Point Transformer2026.06 | 79.98 | |
| 3D RoPE-AxialInput Points=3072, Backbone=Point Transformer2026.06 | 79.13 | |
| nD-RoPE (vector attention)Input Points=768, Backbone=Point Transformer2026.06 | 78.9 | |
| 3D RoPE-Mixed + APEInput Points=3072, Backbone=Point Transformer2026.06 | 78.86 | |
| Learnable Rel. PEInput Points=4096, Backbone=Point Transformer2026.06 | 78.15 | |
| 3D RoPE-Axial + APEInput Points=4096, Backbone=Point Transformer2026.06 | 78.09 | |
| 3D RoPE-Axial + APEInput Points=1024, Backbone=Point Transformer2026.06 | 77.94 | |
| 3D RoPE-MixedInput Points=4096, Backbone=Point Transformer2026.06 | 77.86 | |
| Learnable Rel. PEInput Points=1024, Backbone=Point Transformer2026.06 | 77.51 | |
| 3D RoPE-MixedInput Points=1024, Backbone=Point Transformer2026.06 | 76.48 | |
| 3D RoPE-AxialInput Points=4096, Backbone=Point Transformer2026.06 | 76.14 | |
| nD-RoPE (standard dot-product)Input Points=768, Backbone=Point Transformer2026.06 | 76.07 | |
| 3D RoPE-Mixed + APEInput Points=4096, Backbone=Point Transformer2026.06 | 75.25 | |
| 3D RoPE-Axial + APEInput Points=768, Backbone=Point Transformer2026.06 | 75.12 | |
| 3D RoPE-AxialInput Points=1024, Backbone=Point Transformer2026.06 | 73.74 | |
| Learnable Rel. PEInput Points=768, Backbone=Point Transformer2026.06 | 73.35 | |
| 3D RoPE-MixedInput Points=768, Backbone=Point Transformer2026.06 | 73.23 | |
| 3D RoPE-Mixed + APEInput Points=1024, Backbone=Point Transformer2026.06 | 72.52 | |
| 3D RoPE-AxialInput Points=768, Backbone=Point Transformer2026.06 | 70.16 | |
| 3D RoPE-Mixed + APEInput Points=768, Backbone=Point Transformer2026.06 | 68.95 | |
| nD-RoPE (vector attention)Input Points=256, Backbone=Point Transformer2026.06 | 55.37 | |
| 3D RoPE-Axial + APEInput Points=256, Backbone=Point Transformer2026.06 | 52.71 | |
| 3D RoPE-AxialInput Points=256, Backbone=Point Transformer2026.06 | 48.22 | |
| nD-RoPE (standard dot-product)Input Points=256, Backbone=Point Transformer2026.06 | 46.13 | |
| 3D RoPE-Mixed + APEInput Points=256, Backbone=Point Transformer2026.06 | 45.25 | |
| Learnable Rel. PEInput Points=256, Backbone=Point Transformer2026.06 | 43.78 | |
| 3D RoPE-MixedInput Points=256, Backbone=Point Transformer2026.06 | 40.41 |