3D Object Detection on nuScenes (val) (Core Metrics)
76.7NDSOurs
Evaluation results
| Method | Links | |||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| OursModality=L+C2026.05 | 76.7 | 73.4 | — | — | 38.6 | — | — | — | — | — | — | 90.2 | — | — | — | — | — | 88.2 | 63.1 | 75.1 | 67.4 | 78.2 | 83.2 | 60.2 | 90.3 | |
| SparseLIFsModality=L+C2026.05 | 76.5 | 73.4 | — | — | 82 | — | — | — | — | — | — | 75 | — | — | — | — | — | 90.7 | 72.3 | 46.3 | 33.1 | 90.8 | 83.9 | 76.9 | 83 | |
| MV2DFusionModality=L+C2026.05 | 75.4 | 73.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Co-Fusion4DImage Encoder=Swin-T2026.05 | 75.3 | 73.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MV2DFusion + RayMambaData=LC2026.04 | 75 | 73.4 | — | — | 37.8 | — | — | — | — | — | — | 83.6 | — | — | — | — | — | 90 | 67.5 | 79.2 | 47.4 | 75.6 | 84.8 | 77.1 | 90.5 | |
| MV2DFusionData=LC2026.04 | 74.8 | 73 | — | — | 37.8 | — | — | — | — | — | — | 83.9 | — | — | — | — | — | 89 | 66.1 | 80 | 45.7 | 75.3 | 85.2 | 76.6 | 91 | |
| UniTRModality=L2026.05 | 74.5 | 70.9 | — | — | 39.2 | — | — | — | — | — | — | 89.7 | — | — | — | — | — | 87.9 | 60.2 | 72.2 | 65.1 | 76.8 | 75.8 | 52.2 | 89.4 | |
| IS-FusionModality=L+C2026.05 | 74 | 72.8 | — | — | 38.4 | — | — | — | — | — | — | 89.2 | — | — | — | — | — | 88.3 | 62.7 | 74.9 | 67.3 | 78.1 | 82.4 | 59.5 | 89.3 | |
| IS-FusionImage Encoder=Swin-T2026.05 | 74 | 72.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DepthFusionVenue=TMM’26, Modality=LC2026.06 | 74 | 71.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IS-FusionMasked-Modal Training=true2026.05 | 73.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Ours2025.07 | 73.6 | 71.5 | 89.8 | 68.5 | 35.1 | 77.2 | 45.5 | 78 | 80.5 | 68.3 | 90.1 | 82 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Co-Fix3DImage Encoder=ResNet-502026.05 | 73.6 | 70.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusion4DData=LC2026.04 | 73.5 | 72 | — | — | 32.9 | — | — | — | — | — | — | 80.9 | — | — | — | — | — | 90.6 | 70.3 | 81.5 | 47.1 | 71.6 | 81.5 | 73 | 90.2 | |
| BEVFusion4DImage Encoder=Swin-T2026.05 | 73.5 | 72 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GAFusionImage Encoder=Swin-T2026.05 | 73.5 | 72.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GraphBEV++ (Query)Modality=LC2026.06 | 73.4 | 71.4 | — | — | 32.3 | — | — | — | — | — | — | 82.5 | — | — | — | — | — | 91.6 | 66.2 | 77.4 | 45.5 | 77.8 | 81.9 | 68.4 | 90.4 | |
| SparseFusionMasked-Modal Training=true2026.05 | 73.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SparseFusionMasked-Modal Training=false2026.05 | 73.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFormer-MVenue=TPAMI’25, Modality=LC2026.06 | 73.2 | 71.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GraphBEV++ (LSS)Modality=LC2026.06 | 73.2 | 70.7 | — | — | 31.8 | — | — | — | — | — | — | 83.4 | — | — | — | — | — | 90.5 | 65.3 | 76.4 | 44.4 | 76.6 | 80.8 | 68.1 | 89.7 | |
| SparseFusionModality=L+C2026.05 | 73.1 | 71 | — | — | 38.3 | — | — | — | — | — | — | 87.5 | — | — | — | — | — | 87.6 | 59.8 | 71.6 | 64.5 | 78.8 | 78.1 | 59.4 | 90.5 | |
| EA-LSSModality=L+C2026.05 | 73.1 | 71.2 | — | — | 71.7 | — | — | — | — | — | — | 79.5 | — | — | — | — | — | 85.2 | 62.1 | 64.1 | 38.9 | 86.3 | 80.9 | 61.6 | 86.2 | |
| SparseFusionImage Encoder=ResNet-502026.05 | 73.1 | 71 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFormer-MVenue=TPAMI’25, Modality=LC, re-implementation=true2026.06 | 73 | 70.9 | — | — | 32.1 | — | — | — | — | — | — | 82.2 | — | — | — | — | — | 90.4 | 65.7 | 77.1 | 45 | 77.5 | 81.2 | 68 | 89.7 | |
| GraphBEVPresent at=ECCV 242025.07 | 72.9 | 70.1 | 89.8 | 64.2 | 31.2 | 75.8 | 43.5 | 75.6 | 79.3 | 66.3 | 88.6 | 80.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CMTMasked-Modal Training=true2026.05 | 72.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HD-FusionVenue=TII’26, Modality=LC2026.06 | 72.9 | 70.5 | — | — | 35.4 | — | — | — | — | — | — | 83.3 | — | — | — | — | — | 88.8 | 60.5 | 80.7 | 48.1 | 68.1 | 79.8 | 69.7 | 90.3 | |
| GraphBEVVenue=ECCV’24, Modality=LC2026.06 | 72.9 | 70.1 | — | — | 31.1 | — | — | — | — | — | — | 82.2 | — | — | — | — | — | 89.9 | 64.7 | 76 | 43.8 | 76 | 80.1 | 67.5 | 89.2 | |
| UniMambaPresent at=CVPR 252025.07 | 72.6 | 68.5 | 88.7 | 64.7 | 28.7 | 79.7 | 47.9 | 72.3 | 74.6 | 59.1 | 89.7 | 79.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IS-FusionMasked-Modal Training=false2026.05 | 72.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| M3NetPresent at=AAAI 252025.07 | 72.4 | 69 | 89 | 64.5 | 30.3 | 77.9 | 47.5 | 73.2 | 76.5 | 61.4 | 89.2 | 80.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ObjectFusionVenue=ICCV’23, Modality=LC2026.06 | 72.3 | 69.8 | — | — | 31.8 | — | — | — | — | — | — | 81.1 | — | — | — | — | — | 89.7 | 65.6 | 77.7 | 42.8 | 75.2 | 79.4 | 65 | 89.3 | |
| LION-MambaPresent at=NeurIPS 242025.07 | 72.1 | 68 | 87.9 | 64.9 | 28.5 | 77.6 | 44.4 | 71.6 | 75.6 | 59.4 | 89.6 | 80.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusionData=LC2026.04 | 72.1 | 69.6 | — | — | 30.9 | — | — | — | — | — | — | 79.3 | — | — | — | — | — | 89.1 | 66.7 | 77.7 | 42.6 | 73.5 | 79 | 67.5 | 89.4 | |
| BEVDiffuserPresent at=CVPR 252025.07 | 71.9 | 69.2 | 88.5 | 63.5 | 31 | 75.3 | 46.2 | 73.2 | 77.5 | 62.8 | 87.9 | 80.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CMTMasked-Modal Training=false2026.05 | 71.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FSHNetPresent at=CVPR 252025.07 | 71.7 | 68.1 | 88.7 | 61.4 | 26.3 | 79.3 | 47.8 | 72.3 | 76.7 | 60.5 | 89.3 | 78.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusion-MITPresent at=ICRA 232025.07 | 71.4 | 68.5 | 88.2 | 61.7 | 30.2 | 75.1 | 41.5 | 72.5 | 76.3 | 64.2 | 87.5 | 81 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusionMasked-Modal Training=false2026.05 | 71.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusionImage Encoder=Swin-T2026.05 | 71.4 | 68.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusionData=LC2026.04 | 71.3 | 67.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusion-LCImage Encoder=ResNet-502026.05 | 71.3 | 67.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DGFSDVenue=MM’25, Modality=L2026.06 | 71.3 | 66.8 | — | — | 26.4 | — | — | — | — | — | — | 75.7 | — | — | — | — | — | 87.9 | 63.4 | 77.9 | 47.2 | 69.8 | 75.8 | 56.1 | 88.1 | |
| DSVTRepr.=Pillar, Key Operators=Transformer2025.06 | 71.1 | 66.4 | — | — | 25.3 | — | — | — | — | — | — | 77.9 | — | — | — | — | — | 87.4 | 62.6 | 75.9 | 42.1 | 71 | 74.8 | 58.7 | 88.2 | |
| HEDnetModality=L2026.05 | 71.09 | 66.97 | — | — | 27.64 | — | — | — | — | — | — | 76.19 | — | — | — | — | — | 87.55 | 62.29 | 77.51 | 46.93 | 69.25 | 75.05 | 59.67 | 87.57 | |
| SAFDNetPresent at=CVPR 242025.07 | 71 | 66.3 | 87.6 | 60.8 | 26.6 | 78 | 43.5 | 69.7 | 75.5 | 58 | 87.8 | 75 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusion-PKUPresent at=NeurIPS 222025.07 | 71 | 67.9 | 87.3 | 59.8 | 28.9 | 73.5 | 41.2 | 73.8 | 74.6 | 59.8 | 85.4 | 68.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SAFDNetRepr.=Voxel, Key Operators=SpConv2025.06 | 71 | 66.3 | — | — | 26.6 | — | — | — | — | — | — | 75 | — | — | — | — | — | 87.6 | 60.8 | 78 | 43.5 | 69.7 | 75.5 | 58 | 87.8 | |
| FALORepr.=Pillar, Key Operators=Conv, MLP, HP2025.06 | 70.8 | 65.7 | — | — | 25.5 | — | — | — | — | — | — | 77.5 | — | — | — | — | — | 87.1 | 62.3 | 75.3 | 43 | 68.7 | 71.9 | 57.9 | 87.6 | |
| TransFusionMasked-Modal Training=false2026.05 | 70.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepInteractionMasked-Modal Training=true2026.05 | 70.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusionMasked-Modal Training=true2026.05 | 70.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FSDv2Repr.=Voxel, Key Operators=SpConv2025.06 | 70.4 | 64.7 | — | — | 28.5 | — | — | — | — | — | — | 72.9 | — | — | — | — | — | 84.4 | 57.3 | 75.9 | 44.1 | 73.6 | 69.5 | 57.4 | 86.9 | |
| FSDv2Venue=TPAMI’24, Modality=L2026.06 | 70.4 | 64.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusionModality Availability=Both Modalities2026.05 | 70.33 | 66.91 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusion-LPresent at=CVPR 222025.07 | 70.2 | 65.5 | 86.2 | 54.8 | 26.5 | 70.1 | 42.3 | 72.1 | 69.8 | 53.9 | 86.4 | 70.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Enhanced HOPErouting=adaptive LID routing, GFLOPs=21.72026.05 | 70.1 | 65.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusion-LRepr.=Voxel, Key Operators=SpConv2025.06 | 70.1 | 65.5 | — | — | 25.2 | — | — | — | — | — | — | 77.2 | — | — | — | — | — | 86.9 | 60.8 | 73.1 | 43.4 | 70.3 | 72.9 | 57.3 | 87.5 | |
| TransFusion-LVenue=CVPR’22, Modality=L2026.06 | 70.1 | 65.1 | — | — | 25.4 | — | — | — | — | — | — | 74.1 | — | — | — | — | — | 86.5 | 59.6 | 74.4 | 42.2 | 74.1 | 72.1 | 56 | 86.6 | |
| Enhanced HOPEmode=static, ℓ=6, GFLOPs=35.12026.05 | 69.8 | 65.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SB-BEVFusion (Avg)Modality Availability=Both Modalities2026.05 | 69.7 | 67.37 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SB-BEVFusion (Avg)Modality Availability=LiDAR Only2026.05 | 69.59 | 64.48 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LinKRepr.=Voxel, Key Operators=SpConv2025.06 | 69.5 | 63.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusionModality=L+C2026.05 | 69.43 | 67.75 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PillarNeSt-BaseRepr.=Pillar, Key Operators=Conv, MLP2025.06 | 69.2 | 63.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusionGFLOPs=52.72026.05 | 69.1 | 64.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LargeKernel3DRepr.=Voxel, Key Operators=SpConv2025.06 | 69.1 | 63.9 | — | — | 24.3 | — | — | — | — | — | — | 72.3 | — | — | — | — | — | 85.1 | 60.1 | 72.6 | 41.4 | 67.7 | 70.8 | 59.2 | 85.6 | |
| DeepInteractionMasked-Modal Training=false2026.05 | 69.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PillarNeXt-BRepr.=Pillar, Key Operators=SpConv2025.06 | 68.8 | 62.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusionMasked-Modal Training=true2026.05 | 68.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| UniBEVModality Availability=Both Modalities2026.05 | 68.5 | 64.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Uni3DETRRepr.=Voxel, Key Operators=SpConv2025.06 | 68.5 | 61.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SparseDriveGFLOPs=38.42026.05 | 68.3 | 63.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FastPillars-mRepr.=Pillar, Key Operators=Conv, Attention2025.06 | 68.2 | 61.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VISTARepr.=Voxel, Key Operators=SpConv2025.06 | 68.1 | 60.8 | — | — | 36.8 | — | — | — | — | — | — | 84.1 | — | — | — | — | — | 84.8 | 57.2 | 20.5 | 67.6 | 69.7 | 67.7 | 50.7 | 69 | |
| VoxelNextModality=L2026.05 | 67.66 | 60.48 | — | — | 22.58 | — | — | — | — | — | — | 71.01 | — | — | — | — | — | 84.04 | 54.59 | 66.46 | 38.53 | 65.37 | 65.26 | 50.78 | 86.1 | |
| PillarNet-34Repr.=Pillar, Key Operators=SpConv2025.06 | 67.6 | 60.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VoxelNeXtRepr.=Voxel, Key Operators=SpConv2025.06 | 66.7 | 60.5 | — | — | 21.1 | — | — | — | — | — | — | 69.4 | — | — | — | — | — | 83.9 | 55.5 | 70.5 | 38.1 | 69.4 | 62.8 | 50 | 84.6 | |
| CenterPoint + RayMambaData=L2026.04 | 65.6 | 57.4 | — | — | 16.7 | — | — | — | — | — | — | 66.6 | — | — | — | — | — | 85.5 | 54.4 | 68.2 | 38.2 | 63.5 | 58.7 | 38.5 | 84.1 | |
| UniBEVModality Availability=LiDAR Only2026.05 | 65.3 | 58.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CenterPointGFLOPs=30.22026.05 | 64.8 | 56.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CenterPointRepr.=Voxel, Key Operators=SpConv2025.06 | 64.8 | 56.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CenterPointData=L2026.04 | 64.7 | 56.4 | — | — | 16.5 | — | — | — | — | — | — | 64.3 | — | — | — | — | — | 84.6 | 53.9 | 67 | 37.5 | 65 | 55.3 | 36.7 | 83.3 | |
| CenterPointAnnotation Rate=100%2026.02 | 64.61 | 56.11 | — | — | 16.4 | — | — | — | — | — | — | 64.91 | — | — | — | — | — | 84.63 | 52.87 | 66.85 | 36.85 | 65.79 | 53.97 | 35.39 | 83.46 | |
| RaCFormerInput=C+R, Backbone=R101, Image Size=1408 x 512, CBGS=false, Future Frames=false2025.05 | 63 | 57.3 | — | — | — | — | — | — | — | — | — | — | 47.6 | 26.1 | 42.8 | 21.3 | 18 | — | — | — | — | — | — | — | — | |
| StreamPETRBackbone=EVA-02-L, Image Resolution=800×1600, Model Size (M)=316.6, GFLOPs=45,044, Latency tau_E (ms)=2,571, Latency tau (ms)=2,724, Avg. rank=3.362026.04 | 63 | 55.5 | — | — | — | — | — | — | — | — | — | — | 53.9 | 25.5 | 25.8 | 23.1 | 19.9 | — | — | — | — | — | — | — | — | |
| StreamPETR + ToC3DBackbone=EVA-02-L, Image Resolution=800×1600, Model Size (M)=324.4, GFLOPs=37,060, Latency tau_E (ms)=2,020, Latency tau (ms)=2,171, Avg. rank=2.902026.04 | 63 | 55.2 | — | — | — | — | — | — | — | — | — | — | 53.4 | 25.7 | 24.1 | 23.1 | 20 | — | — | — | — | — | — | — | — | |
| StreamPETR + Ours (Dynamic Token Selection)Backbone=EVA-02-L, Image Resolution=800×1600, Token selection rate (r)=0.5, Model Size (M)=318.2, GFLOPs=26,798, Latency tau_E (ms)=1,954, Latency tau (ms)=2,098, Avg. rank=2.272026.04 | 63 | 55.5 | — | — | — | — | — | — | — | — | — | — | 53.9 | 25.5 | 25.6 | 23 | 19.6 | — | — | — | — | — | — | — | — | |
| StreamPETR + Ours (Dynamic Token Selection)Backbone=EVA-02-L, Image Resolution=800×1600, Token selection rate (r)=0.3, Model Size (M)=318.2, GFLOPs=21,131, Latency tau_E (ms)=1,737, Latency tau (ms)=1,880, Avg. rank=2.272026.04 | 63 | 55.5 | — | — | — | — | — | — | — | — | — | — | 54.1 | 25.4 | 25.8 | 23.1 | 19.7 | — | — | — | — | — | — | — | — | |
| Refine3DVenue=AAAI’26, Modality=C2026.06 | 62.4 | 53.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RICCARDOInput=C+R, Backbone=R101, Image Size=1408 x 512, CBGS=false, Future Frames=true2025.05 | 62.2 | 54.4 | — | — | — | — | — | — | — | — | — | — | 48.1 | 26.6 | 32.5 | 23.7 | 18.9 | — | — | — | — | — | — | — | — | |
| RQR3DInput=C+R, Backbone=R101, Image Size=1408 x 512, CBGS=false, Future Frames=false2025.05 | 62.2 | 54.7 | — | — | — | — | — | — | — | — | — | — | 41.7 | 28.1 | 38.1 | 22.8 | 18.8 | — | — | — | — | — | — | — | — | |
| CRT-FusionInput=C+R, Backbone=R101, Image Size=1408 x 512, CBGS=false, Future Frames=false2025.05 | 62.1 | 55.4 | — | — | — | — | — | — | — | — | — | — | 42.5 | 26.4 | 43.3 | 23.7 | 19.3 | — | — | — | — | — | — | — | — | |
| StreamPETR + tgGBCBackbone=EVA-02-L, Image Resolution=800×1600, Model Size (M)=316.6, GFLOPs=44,964, Latency tau_E (ms)=2,491, Latency tau (ms)=2,550, Avg. rank=3.542026.04 | 62.1 | 54.2 | — | — | — | — | — | — | — | — | — | — | 56 | 25.6 | 25.5 | 22.3 | 20.4 | — | — | — | — | — | — | — | — | |
| StreamPETR + Ours (Dynamic Token Selection)Backbone=EVA-02-L, Image Resolution=800×1600, Token selection rate (r)=0.1, Model Size (M)=318.2, GFLOPs=18,512, Latency tau_E (ms)=1,652, Latency tau (ms)=1,789, Avg. rank=2.092026.04 | 62.1 | 54.1 | — | — | — | — | — | — | — | — | — | — | 55.6 | 25.4 | 25.4 | 23.4 | 19.4 | — | — | — | — | — | — | — | — | |
| CMCRArchitecture=VoxelNet + CenterPoint, Labeled data percentage=20%2024.12 | 62 | 52.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HyDRaInput=C+R, Backbone=R101, Image Size=1408 x 512, CBGS=false, Future Frames=false2025.05 | 61.7 | 53.6 | — | — | — | — | — | — | — | — | — | — | 41.6 | 26.4 | 40.7 | 23.1 | 18.6 | — | — | — | — | — | — | — | — | |
| CSCArchitecture=VoxelNet + CenterPoint, Labeled data percentage=20%2024.12 | 61.3 | 51.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RaCFormerInput=C+R, Backbone=R50, Image Size=704 x 256, CBGS=false, Future Frames=false2025.05 | 61.3 | 54.1 | — | — | — | — | — | — | — | — | — | — | 47.8 | 26.1 | 42.8 | 21.3 | 18 | — | — | — | — | — | — | — | — |