Semantic Segmentation on nuScenes
79.5mIoU (Average)SphereFormer
Evaluation Results
| Method | Links | ||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SphereFormerTime (ms)=59, GPU (GB)=0.81, Param (M)=32.3, Hardware=Nvidia RTX 3090 GPU, Batch size=12023.03 | 79.5 | — | — | — | — | — | 78.7 | 46.7 | 95.2 | 93.7 | 54 | 88.9 | 81.1 | 68 | 74.2 | 86.2 | 97.2 | 74.3 | 76.3 | 75.8 | 91.4 | 89.7 | |
| OA-CNNs2024.03 | 78.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SphereFormer2024.03 | 78.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CurveCloudNetType=Curve, Time (ms)=87, GPU (GB)=1.14, Param (M)=28.8, Hardware=Nvidia RTX 3090 GPU, Batch size=12023.03 | 78 | — | — | — | — | — | 77.3 | 45.7 | 92.4 | 91.9 | 59.4 | 84.5 | 78.5 | 64.1 | 69.6 | 85 | 96.9 | 72.7 | 75.6 | 75.2 | 90.5 | 89 | |
| SPVNAS2024.03 | 77.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MinkowskiNetType=Voxel, Time (ms)=44, GPU (GB)=0.29, Param (M)=36.6, Hardware=Nvidia RTX 3090 GPU, Batch size=12023.03 | 76.2 | — | — | — | — | — | 75.4 | 43.9 | 91.9 | 93 | 49 | 84.3 | 78.3 | 64.6 | 65.9 | 85.7 | 96.1 | 71.5 | 67.5 | 74.8 | 86.5 | 84.9 | |
| Cylinder3DTime (ms)=80, GPU (GB)=1.57, Param (M)=55.9, Hardware=Nvidia RTX 3090 GPU, Batch size=12023.03 | 76.1 | — | — | — | — | — | 76.4 | 40.3 | 91.2 | 93.8 | 51.3 | 78 | 78.9 | 64.9 | 62.1 | 84.4 | 96.8 | 71.6 | 76.4 | 75.4 | 90.5 | 87.4 | |
| Cylinder3D2024.03 | 76.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SparseUNet2024.03 | 73.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PolarNetHardware=Nvidia RTX 3090 GPU, Batch size=12023.03 | 71 | — | — | — | — | — | 74.7 | 28.2 | 85.3 | 90.9 | 35.1 | 77.5 | 71.3 | 58.8 | 57.4 | 76.1 | 96.5 | 71.1 | 74.7 | 74 | 87.3 | 85.7 | |
| PointMLPType=Point, Time (ms)=164, GPU (GB)=4.94, Param (M)=16.8, Hardware=Nvidia RTX 3090 GPU, Batch size=12023.03 | 67.9 | — | — | — | — | — | 72.3 | 27.8 | 88.2 | 86.3 | 37.2 | 51 | 60.7 | 50.6 | 56.4 | 71.1 | 95.7 | 70.6 | 70.9 | 72 | 88.8 | 87.2 | |
| DINOv3 + CDArch.=ViT-H+/16, Evaluation Protocol=linear probing2026.02 | 65.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PointNextType=Point, Time (ms)=155, GPU (GB)=0.62, Param (M)=41.5, Hardware=Nvidia RTX 3090 GPU, Batch size=12023.03 | 65 | — | — | — | — | — | 68.7 | 1.2 | 86.9 | 87.5 | 41.8 | 57.4 | 54.3 | 34.9 | 55.3 | 75.1 | 95.7 | 68.9 | 70.2 | 71.5 | 86.5 | 84 | |
| DINOv3Arch.=ViT-H+/16, Evaluation Protocol=linear probing2026.02 | 64.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RandLA-NetType=Point, Time (ms)=21, GPU (GB)=0.18, Param (M)=1.2, Hardware=Nvidia RTX 3090 GPU, Batch size=12023.03 | 62.9 | — | — | — | — | — | 72.5 | 12.6 | 36.6 | 81.8 | 38.7 | 72.3 | 68.5 | 37.3 | 44.7 | 59.7 | 95.3 | 87 | 69.7 | 71.1 | 73.2 | 85.9 | |
| DepthAnythingV2 + CDArch.=ViT-B/14, Evaluation Protocol=linear probing2026.02 | 55.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DepthAnythingV2Arch.=ViT-B/14, Evaluation Protocol=linear probing2026.02 | 53.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PointNet++Type=Point, Time (ms)=274, GPU (GB)=0.6, Param (M)=1.5, Hardware=Nvidia RTX 3090 GPU, Batch size=12023.03 | 51.1 | — | — | — | — | — | 60.1 | 6.5 | 58.4 | 66.3 | 16.4 | 20 | 50.8 | 12.6 | 31.5 | 42 | 94 | 60.8 | 63.8 | 69.2 | 82.4 | 82.3 | |
| IBOT + CDArch.=ViT-L/16, Evaluation Protocol=linear probing2026.02 | 49.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLIP + CDArch.=ViT-B/16, Evaluation Protocol=linear probing2026.02 | 49.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DINO + CDArch.=ViT-B/8, Evaluation Protocol=linear probing2026.02 | 48.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IBOTArch.=ViT-L/16, Evaluation Protocol=linear probing2026.02 | 47.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DINO + CDArch.=ResNet502026.02 | 46.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DINO + CDArch.=ResNet50, Evaluation Protocol=linear probing2026.02 | 46.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLIPArch.=ViT-B/16, Evaluation Protocol=linear probing2026.02 | 45.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DistillBEVArch.=ResNet502026.02 | 44.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DINOArch.=ViT-B/8, Evaluation Protocol=linear probing2026.02 | 44.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MoCov2 + CDArch.=ResNet50, Evaluation Protocol=linear probing2026.02 | 41.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MoCov2Arch.=ResNet50, Evaluation Protocol=linear probing2026.02 | 38.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DINOArch.=ResNet502026.02 | 38.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DINOArch.=ResNet50, Evaluation Protocol=linear probing2026.02 | 38.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MoVieDriveDiverse conditions?=✓, Multi-modal?=✓2025.08 | 37.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVDistillArch.=ResNet502026.02 | 36.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CogVideoX + SyntheOccDiverse conditions?=✓, Multi-modal?=✗2025.08 | 32.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MagicDriveDiverse conditions?=✓, Multi-modal?=✗2025.08 | 23.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Le MuMo JEPA2026.03 | 22.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MagicDrive-V2Diverse conditions?=✓, Multi-modal?=✗2025.08 | 22.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ImageBind2026.03 | 19.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MultiMAE-SS2026.03 | 19.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MultiMAE-MT2026.03 | 19.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ALSOBackbone=MinkUNet [74]2022.12 | — | 26.2 | 37.4 | 59 | 69.8 | 71.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ALSOBackbone=SPVCNN [78]2022.12 | — | 24.8 | 37.4 | 58.4 | 69.5 | 71.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DepthContrastBackbone=MinkUNet [74]2022.12 | — | 21.7 | 34.6 | 57.4 | 69.2 | 71.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| No pre-trainingBackbone=MinkUNet [74]2022.12 | — | 21.6 | 35 | 57.3 | 69 | 71.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| No pre-trainingBackbone=SPVCNN [78]2022.12 | — | 22.2 | 34.4 | 57.1 | 69 | 70.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PointContrastBackbone=MinkUNet [74]2022.12 | — | 27.1 | 37 | 58.9 | 69.4 | 71.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |