3D Object Detection on nuScenes v1.0 (val)
71.2mAP (Overall)BEVFusion (Liang et al.) + EA-LSS
Evaluation results
| Method | Links | |||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| BEVFusion (Liang et al.) + EA-LSSModality=Camera + LiDAR, Latency (ms)=194.92023.03 | 71.2 | — | — | — | — | — | — | — | — | 73.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CMTModality=Camera + LiDAR2023.03 | 70.3 | — | — | — | — | — | — | — | — | 72.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepInteractionModality=Camera + LiDAR, Latency (ms)=204.12023.03 | 69.9 | — | — | — | — | — | — | — | — | 72.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusion (Liang et al.)Modality=Camera + LiDAR, Latency (ms)=190.32023.03 | 69.6 | — | — | — | — | — | — | — | — | 72.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusion (Z. et al.) + EA-LSSModality=Camera + LiDAR, Latency (ms)=123.62023.03 | 69.4 | — | — | — | — | — | — | — | — | 71.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| 3D Dual-Fusion (T)Img Backbone=R50, LiDAR Backbone=VoxelNet2022.11 | 69.3 | — | — | — | — | — | — | — | — | 72.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusionBackbone=Voxel-based, Input=LiDAR-Camera fusion (LC)2023.05 | 68.52 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusionImg Backbone=Swin-Tiny, LiDAR Backbone=VoxelNet2022.11 | 68.5 | — | — | — | — | — | — | — | — | 71.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusion (Z. et al.)Modality=Camera + LiDAR, Latency (ms)=119.22023.03 | 68.5 | — | — | — | — | — | — | — | — | 71.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusionImg Backbone=Swin-Tiny, LiDAR Backbone=VoxelNet2022.11 | 67.9 | — | — | — | — | — | — | — | — | 71 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| No PruningBaseline Model=BEVfusion-mit, Sparsity=0%2024.09 | 67.8 | — | — | — | — | — | — | — | — | 70.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusionImg Backbone=R50, LiDAR Backbone=VoxelNet2022.11 | 67.5 | — | — | — | — | — | — | — | — | 71.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusionModality=Camera + LiDAR, Latency (ms)=156.62023.03 | 67.5 | — | — | — | — | — | — | — | — | 71.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| 3D Dual-Fusion (C)Img Backbone=R50, LiDAR Backbone=VoxelNet2022.11 | 67.3 | — | — | — | — | — | — | — | — | 71.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AlterMOMABaseline Model=BEVfusion-mit, Sparsity=80%2024.09 | 67.3 | — | — | — | — | — | — | — | — | 70.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MVPImg Backbone=DLA34, LiDAR Backbone=VoxelNet2022.11 | 67.1 | — | — | — | — | — | — | — | — | 70.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AutoAlignv2Img Backbone=CSPNet, LiDAR Backbone=VoxelNet2022.11 | 67.1 | — | — | — | — | — | — | — | — | 71.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusionBackbone=Voxel-based, Input=LiDAR-Camera fusion (LC)2023.05 | 67.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FBMNetBackbone=Voxel-based, Input=LiDAR-Camera fusion (LC)2023.05 | 66.93 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| No PruningBaseline Model=BEVfusion-pku, Sparsity=0%2024.09 | 66.9 | — | — | — | — | — | — | — | — | 70.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HEDNet2023.10 | 66.7 | — | — | — | — | — | — | — | — | 71.4 | — | — | — | — | — | 87.7 | 60.6 | 77.8 | 50.7 | 28.9 | 87.1 | 74.3 | 56.8 | 76.3 | 66.9 | — | — | — | — | — | — | — | — | — | — | |
| FBMNetBackbone=Voxel-based, Input=LC + Asynchronous Sensors (AS)2023.05 | 66.66 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AlterMOMABaseline Model=BEVfusion-pku, Sparsity=80%2024.09 | 66.5 | — | — | — | — | — | — | — | — | 70.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MVPBackbone=Voxel-based, Input=LiDAR-Camera fusion (LC)2023.05 | 65.97 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FBMNetBackbone=Voxel-based, Input=LC + Misaligned Sensor Placement (MSP)2023.05 | 65.88 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FocalsConvImg Backbone=R50, LiDAR Backbone=VoxelNet-FocalsConv2022.11 | 65.6 | — | — | — | — | — | — | — | — | 70.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FBMNetBackbone=Voxel-based, Input=LC + Degenerated Images (DI)2023.05 | 65.59 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusionBackbone=Voxel-based, Input=LC + Asynchronous Sensors (AS)2023.05 | 65.54 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusion-L2023.10 | 65.5 | — | — | — | — | — | — | — | — | 70.1 | — | — | — | — | — | 86.9 | 60.8 | 73.1 | 43.4 | 25.2 | 87.5 | 72.9 | 57.3 | 77.2 | 70.3 | — | — | — | — | — | — | — | — | — | — | |
| AlterMOMABaseline Model=BEVfusion-mit, Sparsity=85%2024.09 | 65.5 | — | — | — | — | — | — | — | — | 69.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusionBackbone=Voxel-based, Input=LC + Misaligned Sensor Placement (MSP)2023.05 | 65.34 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Supervised 100%Labels=Human, Supervision level=100%2024.05 | 65.2 | — | — | — | — | — | — | — | — | 59.7 | — | — | — | — | — | 75.4 | — | — | — | — | 74.2 | — | 45.9 | — | — | 53.2 | — | — | — | — | — | — | — | — | — | |
| TransFusion-LImg Backbone=-, LiDAR Backbone=VoxelNet2022.11 | 65.1 | — | — | — | — | — | — | — | — | 70.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusionBackbone=Voxel-based, Input=LiDAR-only (L)2023.05 | 64.98 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FBMNetBackbone=Voxel-based, Input=LiDAR-only (L)2023.05 | 64.98 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusionBackbone=Voxel-based, Input=LC + Degenerated Images (DI)2023.05 | 64.64 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FUTR3DImg Backbone=R101, LiDAR Backbone=VoxelNet2022.11 | 64.5 | — | — | — | — | — | — | — | — | 68.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusionBackbone=Voxel-based, Input=LC + Degenerated Images (DI)2023.05 | 64.42 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ProsPrBaseline Model=BEVfusion-mit, Sparsity=80%2024.09 | 64.3 | — | — | — | — | — | — | — | — | 69.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AlterMOMABaseline Model=BEVfusion-pku, Sparsity=85%2024.09 | 64.2 | — | — | — | — | — | — | — | — | 68.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusionBackbone=Voxel-based, Input=LC + Asynchronous Sensors (AS)2023.05 | 64.19 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ProsPrBaseline Model=BEVfusion-pku, Sparsity=80%2024.09 | 63.6 | — | — | — | — | — | — | — | — | 68.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AlterMOMABaseline Model=BEVfusion-mit, Sparsity=90%2024.09 | 63.5 | — | — | — | — | — | — | — | — | 66.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEVFusionBackbone=Voxel-based, Input=LC + Misaligned Sensor Placement (MSP)2023.05 | 63.41 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SynFlowBaseline Model=BEVfusion-mit, Sparsity=80%2024.09 | 63.2 | — | — | — | — | — | — | — | — | 67.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MVPBackbone=Pillar-based, Input=LiDAR-Camera fusion (LC)2023.05 | 62.72 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SynFlowBaseline Model=BEVfusion-pku, Sparsity=80%2024.09 | 62.4 | — | — | — | — | — | — | — | — | 67.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AlterMOMABaseline Model=BEVfusion-pku, Sparsity=90%2024.09 | 62.3 | — | — | — | — | — | — | — | — | 66 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SNIPBaseline Model=BEVfusion-mit, Sparsity=80%2024.09 | 62.2 | — | — | — | — | — | — | — | — | 67.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ProsPrBaseline Model=BEVfusion-mit, Sparsity=85%2024.09 | 61.9 | — | — | — | — | — | — | — | — | 66.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SNIPBaseline Model=BEVfusion-pku, Sparsity=80%2024.09 | 61.8 | — | — | — | — | — | — | — | — | 67.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Uni3DETRtest-time augmentation=false2023.10 | 61.7 | — | — | — | — | — | — | — | — | 68.5 | 0.288 | 0.249 | 0.303 | 0.216 | 0.181 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FocalsConvImg Backbone=-, LiDAR Backbone=VoxelNet-FocalsConv2022.11 | 61.2 | — | — | — | — | — | — | — | — | 68.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| UVTRtest-time augmentation=false2023.10 | 60.9 | — | — | — | — | — | — | — | — | 67.7 | 0.334 | 0.257 | 0.3 | 0.204 | 0.182 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VoxelNeXttest-time augmentation=false, implementation=OpenPCDet2023.10 | 60.5 | — | — | — | — | — | — | — | — | 66.7 | 0.301 | 0.252 | 0.406 | 0.217 | 0.186 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VoxelNeXt2023.10 | 60.5 | — | — | — | — | — | — | — | — | 66.7 | — | — | — | — | — | 83.9 | 55.5 | 70.5 | 38.1 | 21.1 | 84.6 | 62.8 | 50 | 69.4 | 69.4 | — | — | — | — | — | — | — | — | — | — | |
| TransFusionBackbone=Pillar-based, Input=LiDAR-Camera fusion (LC)2023.05 | 59.96 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ProsPrBaseline Model=BEVfusion-pku, Sparsity=85%2024.09 | 59.9 | — | — | — | — | — | — | — | — | 66.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FBMNetBackbone=Pillar-based, Input=LiDAR-Camera fusion (LC)2023.05 | 59.88 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PillarNettest-time augmentation=false2023.10 | 59.8 | — | — | — | — | — | — | — | — | 67.4 | 0.277 | 0.252 | 0.289 | 0.247 | 0.191 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CenterPointImg Backbone=-, LiDAR Backbone=VoxelNet2022.11 | 59.6 | — | — | — | — | — | — | — | — | 66.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MVPBackbone=Voxel-based, Input=LiDAR-only (L)2023.05 | 59.53 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IMPBaseline Model=BEVfusion-mit, Sparsity=80%2024.09 | 59.3 | — | — | — | — | — | — | — | — | 66.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CenterPoint2023.10 | 59.2 | — | — | — | — | — | — | — | — | 66.5 | — | — | — | — | — | 84.9 | 57.4 | 70.7 | 38.1 | 16.9 | 85.1 | 59 | 42 | 69.8 | 68.3 | — | — | — | — | — | — | — | — | — | — | |
| FBMNetBackbone=Pillar-based, Input=LC + Misaligned Sensor Placement (MSP)2023.05 | 58.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ProsPrBaseline Model=BEVfusion-mit, Sparsity=90%2024.09 | 58.6 | — | — | — | — | — | — | — | — | 62.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FBMNetBackbone=Pillar-based, Input=LC + Asynchronous Sensors (AS)2023.05 | 58.48 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MVPBackbone=Voxel-based, Input=LC + Degenerated Images (DI)2023.05 | 57.67 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IMPBaseline Model=BEVfusion-pku, Sparsity=80%2024.09 | 57.3 | — | — | — | — | — | — | — | — | 65.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FBMNetBackbone=Pillar-based, Input=LC + Degenerated Images (DI)2023.05 | 56.96 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SynFlowBaseline Model=BEVfusion-mit, Sparsity=85%2024.09 | 56.9 | — | — | — | — | — | — | — | — | 64.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ProsPrBaseline Model=BEVfusion-pku, Sparsity=90%2024.09 | 56.7 | — | — | — | — | — | — | — | — | 62.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CenterPointtest-time augmentation=false2023.10 | 56.6 | — | — | — | — | — | — | — | — | 64.9 | 0.291 | 0.252 | 0.324 | 0.284 | 0.189 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SNIPBaseline Model=BEVfusion-mit, Sparsity=85%2024.09 | 56.4 | — | — | — | — | — | — | — | — | 63.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Full Precision BaselineDetector=CenterPoint, Backbone=VoxelNet, Bits=FP32, Modality=LiDAR2026.02 | 56.3 | — | — | — | — | — | — | — | — | 64.8 | 0.288 | 0.254 | 0.326 | 0.282 | 0.187 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SynFlowBaseline Model=BEVfusion-pku, Sparsity=85%2024.09 | 55.4 | — | — | — | — | — | — | — | — | 63.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusionBackbone=Pillar-based, Input=LiDAR-only (L)2023.05 | 55.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FBMNetBackbone=Pillar-based, Input=LiDAR-only (L)2023.05 | 55.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Cross-Level Sensor FusionBackbone=V2-99, Image resolution=800 x 3202025.12 | 54.99 | — | — | — | — | — | — | — | — | 56.85 | 0.6334 | 0.2558 | 0.5006 | 0.486 | 0.1885 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SNIPBaseline Model=BEVfusion-pku, Sparsity=85%2024.09 | 54.7 | — | — | — | — | — | — | — | — | 62.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| InlierQDetector=CenterPoint, Backbone=VoxelNet, Bits=W8A8, Modality=LiDAR2026.02 | 54.7 | — | — | — | — | — | — | — | — | 63.6 | 0.297 | 0.257 | 0.346 | 0.293 | 0.182 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BRECQDetector=CenterPoint, Backbone=VoxelNet, Bits=W8A8, Modality=LiDAR2026.02 | 54.4 | — | — | — | — | — | — | — | — | 63.4 | 0.298 | 0.257 | 0.348 | 0.294 | 0.182 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LiDAR-PTQDetector=CenterPoint, Backbone=VoxelNet, Bits=W8A8, Modality=LiDAR2026.02 | 54.4 | — | — | — | — | — | — | — | — | 63.4 | 0.299 | 0.258 | 0.353 | 0.29 | 0.183 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RayDN + Dynamic Token SelectionInput Resolution=320×800, Token selection ratio (r)=0.5, Model Size (M)=333.5, GFLOPs=6,396, Latency τ (ms)=495, Latency E (ms)=5182026.04 | 54 | — | — | — | — | — | — | — | — | 62.3 | 0.519 | 0.252 | 0.275 | 0.231 | 0.193 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 2.27 | |
| RayDN + Dynamic Token SelectionInput Resolution=320×800, Token selection ratio (r)=0.3, Model Size (M)=333.5, GFLOPs=5,484, Latency τ (ms)=480, Latency E (ms)=5032026.04 | 53.8 | — | — | — | — | — | — | — | — | 62.1 | 0.523 | 0.252 | 0.274 | 0.233 | 0.192 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 2.45 | |
| Sparse4Dv2Input Resolution=320×800, Model Size (M)=316.5, GFLOPs=9,004, Latency τ (ms)=246, Latency E (ms)=2792026.04 | 53.8 | — | — | — | — | — | — | — | — | 61.3 | 0.529 | 0.259 | 0.292 | 0.287 | 0.193 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 2.54 | |
| RayDN + Dynamic Token SelectionInput Resolution=320×800, Token selection ratio (r)=0.1, Model Size (M)=333.5, GFLOPs=4,909, Latency τ (ms)=442, Latency E (ms)=4642026.04 | 53.7 | — | — | — | — | — | — | — | — | 62.1 | 0.522 | 0.252 | 0.275 | 0.234 | 0.193 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 2.27 | |
| RayDNInput Resolution=320×800, Model Size (M)=331.9, GFLOPs=9,787, Latency τ (ms)=615, Latency E (ms)=6392026.04 | 53.6 | — | — | — | — | — | — | — | — | 62.3 | 0.54 | 0.257 | 0.25 | 0.201 | 0.204 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 3.09 | |
| Sparse4Dv2 + Dynamic Token SelectionInput Resolution=320×800, Token selection ratio (r)=0.5, Model Size (M)=318.1, GFLOPs=5,751, Latency τ (ms)=220, Latency E (ms)=2522026.04 | 53.5 | — | — | — | — | — | — | — | — | 61 | 0.533 | 0.259 | 0.294 | 0.292 | 0.19 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 2.63 | |
| Sparse4Dv2 + Dynamic Token SelectionInput Resolution=320×800, Token selection ratio (r)=0.3, Model Size (M)=318.1, GFLOPs=4,674, Latency τ (ms)=207, Latency E (ms)=2392026.04 | 53 | — | — | — | — | — | — | — | — | 60.8 | 0.541 | 0.259 | 0.288 | 0.286 | 0.193 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 2.27 | |
| Sparse4Dv2 + Dynamic Token SelectionInput Resolution=320×800, Token selection ratio (r)=0.1, Model Size (M)=318.1, GFLOPs=3,879, Latency τ (ms)=198, Latency E (ms)=2302026.04 | 52.5 | — | — | — | — | — | — | — | — | 60.5 | 0.546 | 0.259 | 0.294 | 0.289 | 0.194 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 2.54 | |
| MVPBackbone=Pillar-based, Input=LiDAR-only (L)2023.05 | 52.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusionBackbone=Pillar-based, Input=LC + Misaligned Sensor Placement (MSP)2023.05 | 52.39 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MVPBackbone=Pillar-based, Input=LC + Degenerated Images (DI)2023.05 | 52.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| InlierQDetector=CenterPoint, Backbone=VoxelNet, Bits=W4A8, Modality=LiDAR2026.02 | 52.2 | — | — | — | — | — | — | — | — | 61.7 | 0.304 | 0.259 | 0.376 | 0.304 | 0.192 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CBGStest-time augmentation=false2023.10 | 51.9 | — | — | — | — | — | — | — | — | 61.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BRECQDetector=CenterPoint, Backbone=VoxelNet, Bits=W4A8, Modality=LiDAR2026.02 | 51.7 | — | — | — | — | — | — | — | — | 61.4 | 0.309 | 0.261 | 0.378 | 0.309 | 0.193 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusionBackbone=Pillar-based, Input=LC + Degenerated Images (DI)2023.05 | 51.38 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransFusionBackbone=Pillar-based, Input=LC + Asynchronous Sensors (AS)2023.05 | 51.36 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IMPBaseline Model=BEVfusion-mit, Sparsity=85%2024.09 | 51.2 | — | — | — | — | — | — | — | — | 59.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |