3D Object Detection on KITTI official leaderboard (test)
90.9AP3D (Easy)Voxel R-CNN
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Voxel R-CNNFPS (Hz)=25.2, Input Modality=LiDAR-only, Method Category=Voxel-based, AP Calculation Protocol=R402020.12 | 90.9 | 81.62 | 77.06 | |
| PV-RCNNFPS (Hz)=8.9, Input Modality=LiDAR-only, Method Category=Point-based, AP Calculation Protocol=R402020.12 | 90.25 | 81.43 | 76.82 | |
| SA-SSDFPS (Hz)=25, Input Modality=LiDAR-only, Method Category=Voxel-based, AP Calculation Protocol=R402020.12 | 88.75 | 79.79 | 74.16 | |
| PatchesInput Modality=LiDAR-only, Method Category=Point-based, AP Calculation Protocol=R402020.12 | 88.67 | 77.2 | 71.82 | |
| UberATG-MMFInput Modality=RGB+LiDAR, AP Calculation Protocol=R402020.12 | 88.4 | 77.43 | 70.22 | |
| 3DSSDFPS (Hz)=26.3, Input Modality=LiDAR-only, Method Category=Point-based, AP Calculation Protocol=R402020.12 | 88.36 | 79.57 | 74.55 | |
| STDFPS (Hz)=12.5, Input Modality=LiDAR-only, Method Category=Point-based, AP Calculation Protocol=R402020.12 | 87.95 | 79.71 | 75.09 | |
| Part-A2Input Modality=LiDAR-only, Method Category=Voxel-based, AP Calculation Protocol=R402020.12 | 87.81 | 78.49 | 73.51 | |
| HVNetFPS (Hz)=31, Input Modality=LiDAR-only, Method Category=Voxel-based, AP Calculation Protocol=R402020.12 | 87.21 | 77.58 | 71.79 | |
| PointRCNNFPS (Hz)=10, Input Modality=LiDAR-only, Method Category=Point-based, AP Calculation Protocol=R402020.12 | 86.96 | 75.64 | 70.7 | |
| PointSIFT+SENetInput Modality=RGB+LiDAR, AP Calculation Protocol=R402020.12 | 85.99 | 72.72 | 64.58 | |
| TANetFPS (Hz)=28.7, Input Modality=LiDAR-only, Method Category=Voxel-based, AP Calculation Protocol=R402020.12 | 85.94 | 75.76 | 68.32 | |
| SECONDFPS (Hz)=30.4, Input Modality=LiDAR-only, Method Category=Voxel-based, AP Calculation Protocol=R402020.12 | 83.34 | 72.55 | 65.82 | |
| AVOD-FPNFPS (Hz)=10, Input Modality=RGB+LiDAR, AP Calculation Protocol=R402020.12 | 83.07 | 71.76 | 65.73 | |
| PointPillarsFPS (Hz)=42.4, Input Modality=LiDAR-only, Method Category=Voxel-based, AP Calculation Protocol=R402020.12 | 82.58 | 74.31 | 68.99 | |
| F-PointNetFPS (Hz)=5.9, Input Modality=RGB+LiDAR, AP Calculation Protocol=R402020.12 | 82.19 | 69.79 | 60.59 | |
| VoxelNetInput Modality=LiDAR-only, Method Category=Voxel-based, AP Calculation Protocol=R402020.12 | 77.47 | 65.11 | 57.73 | |
| MV3DInput Modality=RGB+LiDAR, AP Calculation Protocol=R402020.12 | 74.97 | 63.63 | 54 | |
| Kinematic3DInput Modality=image + video, Inference Time (ms)=1202021.04 | 19.07 | 12.72 | 9.17 | |
| MonoRCNNInput Modality=image, Inference Time (ms)=70, Backbone=ResNet-502021.04 | 18.36 | 12.65 | 10.03 | |
| DA-3DdetInput Modality=image + depth, Inference Time (ms)=400, Backbone=ResNet-1012021.04 | 16.77 | 11.5 | 8.93 | |
| D4LCNInput Modality=image + depth, Inference Time (ms)=200, Backbone=ResNet-1012021.04 | 16.65 | 11.72 | 9.51 | |
| AM3DInput Modality=image + depth, Inference Time (ms)=400, Backbone=ResNet-1012021.04 | 16.5 | 10.74 | 9.52 | |
| PatchNetInput Modality=image + depth, Inference Time (ms)=400, Backbone=ResNet-1012021.04 | 15.68 | 11.12 | 10.17 | |
| MoVi-3DInput Modality=image2021.04 | 15.19 | 10.9 | 9.26 | |
| M3D-RPNInput Modality=image, Inference Time (ms)=160, Backbone=DenseNet-1212021.04 | 14.76 | 9.71 | 7.42 | |
| RTM3DInput Modality=image, Inference Time (ms)=50, Backbone=DLA-34, Training Images=additional images from right cameras2021.04 | 14.41 | 10.34 | 8.77 | |
| MonoPairInput Modality=image, Inference Time (ms)=60, Backbone=DLA-342021.04 | 13.04 | 9.99 | 8.65 | |
| ROI-10DInput Modality=image + depth, Inference Time (ms)=2002021.04 | 4.32 | 2.02 | 1.46 | |
| FQNetInput Modality=image, Inference Time (ms)=5002021.04 | 2.77 | 1.51 | 1.01 |