Andreas Geiger (MPI Tübingen) | Philip Lenz (KIT) | Christoph Stiller (KIT) | Raquel Urtasun (University of Toronto)

Bird's Eye View Evaluation 2017


The bird's eye view benchmark consists of 7481 training images and 7518 test images as well as the corresponding point clouds, comprising a total of 80.256 labeled objects. For evaluation, we compute precision-recall curves. To rank the methods we compute average precision. We require that all methods use the same parameter set for all test pairs. Our development kit provides details about the data format as well as MATLAB / C++ utility functions for reading and writing the label files.

We evaluate bird's eye view detection performance using the PASCAL criteria also used for 2D object detection. Far objects are thus filtered based on their bounding box height in the image plane. As only objects also appearing on the image plane are labeled, objects in don't car areas do not count as false positives. We note that the evaluation does not take care of ignoring detections that are not visible on the image plane — these detections might give rise to false positives. For cars we require a bounding box overlap of 70% in bird's eye view, while for pedestrians and cyclists we require an overlap of 50%. Difficulties are defined as follows:

  • Easy: Min. bounding box height: 40 Px, Max. occlusion level: Fully visible, Max. truncation: 15 %
  • Moderate: Min. bounding box height: 25 Px, Max. occlusion level: Partly occluded, Max. truncation: 30 %
  • Hard: Min. bounding box height: 25 Px, Max. occlusion level: Difficult to see, Max. truncation: 50 %

All methods are ranked based on the moderately difficult results.

Note 2: On 08.10.2019, we have followed the suggestions of the Mapillary team in their paper Disentangling Monocular 3D Object Detection and use 40 recall positions instead of the 11 recall positions proposed in the original Pascal VOC benchmark. This results in a more fair comparison of the results, please check their paper. The last leaderboards right before this change can be found here: Object Detection Evaluation, 3D Object Detection Evaluation, Bird's Eye View Evaluation.
Important Policy Update: As more and more non-published work and re-implementations of existing work is submitted to KITTI, we have established a new policy: from now on, only submissions with significant novelty that are leading to a peer-reviewed paper in a conference or journal are allowed. Minor modifications of existing algorithms or student research projects are not allowed. Such work must be evaluated on a split of the training set. To ensure that our policy is adopted, new users must detail their status, describe their work and specify the targeted venue during registration. Furthermore, we will regularly delete all entries that are 6 months old but are still anonymous or do not have a paper associated with them. For conferences, 6 month is enough to determine if a paper has been accepted and to add the bibliography information. For longer review cycles, you need to resubmit your results.
Additional information used by the methods
  • Stereo: Method uses left and right (stereo) images
  • Flow: Method uses optical flow (2 temporally adjacent images)
  • Multiview: Method uses more than 2 temporally adjacent images
  • Laser Points: Method uses point clouds from Velodyne laser scanner
  • Additional training data: Use of additional data sources for training (see details)

Car


Method Setting Code Moderate Easy Hard Runtime Environment
1 SA-SSD 91.03 % 95.03 % 85.96 % 0.04 s 1 core @ 2.5 Ghz (Python)
2 MMLab PV-RCNN
This method makes use of Velodyne laser scans.
 90.65 % 94.98 % 86.14 % 0.08 s 1 core @ 2.5 Ghz (Python + C/C++)
S. Shi, C. Guo, L. Jiang, Z. Wang, J. Shi, X. Wang and H. Li: PV-RCNN: Point-Voxel Feature Set Abstraction for 3D Object Detection. arXiv preprint arXiv:1912.13192 2019.
3 STD 89.19 % 94.74 % 86.42 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
Z. Yang, Y. Sun, S. Liu, X. Shen and J. Jia: STD: Sparse-to-Dense 3D Object Detector for Point Cloud. ICCV 2019.
4 Point-GNN
This method makes use of Velodyne laser scans.
 89.17 % 93.11 % 83.90 % 0.6 s GPU @ 2.5 Ghz (Python)
5 Noah CV Lab - SSL 89.16 % 90.18 % 81.73 % 0.1 s GPU @ 2.5 Ghz (Python)
6 ORP 89.07 % 93.03 % 81.79 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
7 3DSSD 89.02 % 92.66 % 85.86 % 0.04 s GPU @ 2.5 Ghz (Python + C/C++)
8 MLF_PointCas 88.99 % 92.60 % 81.74 % 0.1 s GPU @ 2.5 Ghz (Python)
9 HVNet 88.82 % 92.83 % 83.38 % 0.03 s GPU @ 2.0 Ghz (Python)
10 PointCSE 88.81 % 92.58 % 83.64 % 0.02 s 1 core @ 2.5 Ghz (C/C++)
11 ELE 88.80 % 94.52 % 85.69 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
12 RGB3D
This method makes use of Velodyne laser scans.
 88.69 % 92.84 % 81.76 % 0.39 s GPU @ 2.5 Ghz (Python)
13 FCPP 88.65 % 92.36 % 83.21 % 0.02 s 1 core @ 2.0 Ghz (Python + C/C++)
14 MuRF 88.56 % 91.57 % 83.46 % 0.05 s GPU @ 1.5 Ghz (Python + C/C++)
15 DENFIDet 88.56 % 92.42 % 83.76 % 0.02 s GPU @ 2.5 Ghz (C/C++)
16 EPNet 88.47 % 94.22 % 83.69 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
17 Patches
This method makes use of Velodyne laser scans.
 88.39 % 92.72 % 83.19 % 0.15 s GPU @ 2.0 Ghz
J. Lehner, A. Mitterecker, T. Adler, M. Hofmarcher, B. Nessler and S. Hochreiter: Patch Refinement: Localized 3D Object Detection. arXiv preprint arXiv:1910.04093 2019.
18 CAASS-3D 88.38 % 94.76 % 81.93 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
19 UberATG-MMF
This method makes use of Velodyne laser scans.
 88.21 % 93.67 % 81.99 % 0.08 s GPU @ 2.5 Ghz (Python)
M. Liang*, B. Yang*, Y. Chen, R. Hu and R. Urtasun: Multi-Task Multi-Sensor Fusion for 3D Object Detection. CVPR 2019.
20 Patches - EMP
This method makes use of Velodyne laser scans.
 88.17 % 94.49 % 84.75 % 0.5 s GPU @ 2.5 Ghz (Python)
J. Lehner, A. Mitterecker, T. Adler, M. Hofmarcher, B. Nessler and S. Hochreiter: Patch Refinement: Localized 3D Object Detection. arXiv preprint arXiv:1910.04093 2019.
21 PointPainting
This method makes use of Velodyne laser scans.
 88.11 % 92.45 % 83.36 % 0.4 s GPU @ 2.5 Ghz (Python + C/C++)
S. Vora, A. Lang, B. Helou and O. Beijbom: PointPainting: Sequential Fusion for 3D Object Detection. arXiv preprint arXiv:1911.10150 2019.
22 OHS-Dense 88.11 % 93.73 % 84.98 % 0.03 s 1 core @ 2.5 Ghz (Python/C++)
Q. Chen, L. Sun, Z. Wang, K. Jia and A. Yuille: Object as Hotspots: An Anchor-Free 3D Object Detection Approach via Firing of Hotspots. 2019.
23 CPRCCNN 88.10 % 94.11 % 83.43 % 0.1 s 1 core @ 2.5 Ghz (Python)
24 Associate-3Ddet
This method makes use of Velodyne laser scans.
 88.09 % 91.40 % 82.96 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
25 DEFT 88.06 % 92.06 % 83.22 % 1 s GPU @ 2.5 Ghz (Python)
26 deprecated 88.05 % 91.96 % 83.21 % 0.05 s GPU @ >3.5 Ghz (Python)
27 3D-CVF
This method makes use of Velodyne laser scans.
 88.04 % 91.97 % 83.22 % 0.05 s GPU @ >3.5 Ghz (Python)
28 UberATG-HDNET
This method makes use of Velodyne laser scans.
 87.98 % 93.13 % 81.23 % 0.05 s GPU @ 2.5 Ghz (Python)
B. Yang, M. Liang and R. Urtasun: HDNET: Exploiting HD Maps for 3D Object Detection. 2nd Conference on Robot Learning (CoRL) 2018.
29 OHS-Direct 87.95 % 93.59 % 83.21 % 0.03 s 1 core @ 2.5 Ghz (Python + C/C++)
Q. Chen, L. Sun, Z. Wang, K. Jia and A. Yuille: Object as Hotspots: An Anchor-Free 3D Object Detection Approach via Firing of Hotspots. 2019.
30 SPA 87.90 % 91.70 % 83.18 % 0.1 s 1 core @ 2.5 Ghz (Python)
31 Fast Point R-CNN
This method makes use of Velodyne laser scans.
 87.84 % 90.87 % 80.52 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
Y. Chen, S. Liu, X. Shen and J. Jia: Fast Point R-CNN. Proceedings of the IEEE international conference on computer vision (ICCV) 2019.
32 MMLab-PartA^2
This method makes use of Velodyne laser scans.
 code 87.79 % 91.70 % 84.61 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
S. Shi, Z. Wang, J. Shi, X. Wang and H. Li: From Points to Parts: 3D Object Detection from Point Cloud with Part-aware and Part-aggregation Network. arXiv preprint arXiv:1907.03670 2019.
33 HRI-FusionRCNN 87.77 % 93.18 % 80.20 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
34 MODet
This method makes use of Velodyne laser scans.
 87.56 % 90.80 % 82.69 % 0.05 s GTX1080Ti
Y. Zhang, Z. Xiang, C. Qiao and S. Chen: Accurate and Real-Time Object Detection Based on Bird's Eye View on 3D Point Clouds. 2019 International Conference on 3D Vision (3DV) 2019.
35 AB3DMOT
This method makes use of Velodyne laser scans.
This is an online method (no batch processing).
 code 87.53 % 91.99 % 81.03 % 0.0047s 1 core @ 2.5 Ghz (Python)
X. Weng and K. Kitani: A Baseline for 3D Multi-Object Tracking. arXiv:1907.03961 2019.
36 PointRGCN 87.49 % 91.63 % 80.73 % 0.26 s GPU @ V100 (Python)
37 MLF_SecCas 87.46 % 92.54 % 77.39 % 0.05 s 1 core @ 2.5 Ghz (Python)
38 SAIC-SA-3D
This method makes use of Velodyne laser scans.
 87.45 % 92.34 % 83.72 % 0.05 s GPU @ 2.5 Ghz (Python)
39 IE-PointRCNN 87.43 % 92.11 % 81.10 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
40 PC-CNN-V2
This method makes use of Velodyne laser scans.
 87.40 % 91.19 % 79.35 % 0.5 s GPU @ 2.5 Ghz (Matlab + C/C++)
X. Du, M. Ang, S. Karaman and D. Rus: A General Pipeline for 3D Detection of Vehicles. 2018 IEEE International Conference on Robotics and Automation (ICRA) 2018.
41 MMLab-PointRCNN
This method makes use of Velodyne laser scans.
 code 87.39 % 92.13 % 82.72 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
S. Shi, X. Wang and H. Li: Pointrcnn: 3d object proposal generation and detection from point cloud. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition 2019.
42 MPNet
This method makes use of Velodyne laser scans.
 87.31 % 91.27 % 83.25 % 0.02 s GPU @ 2.5 Ghz (Python + C/C++)
43 deprecated 87.28 % 90.44 % 75.09 % 0.05 s GPU @ 2.0 Ghz (Python)
44 PiP 87.25 % 90.87 % 83.38 % 0.05 s 1 core @ 2.5 Ghz (Python)
45 HRI-VoxelFPN 87.21 % 92.75 % 79.82 % 0.02 s GPU @ 2.5 Ghz (Python + C/C++)
H. Kuang, B. Wang, J. An, M. Zhang and Z. Zhang: Voxel-FPN:multi-scale voxel feature aggregation in 3D object detection from point clouds. sensors 2020.
46 CentrNet-v1
This method makes use of Velodyne laser scans.
 87.19 % 90.72 % 83.34 % 0.03 s GPU @ 2.5 Ghz (Python)
47 epBRM
This method makes use of Velodyne laser scans.
 code 87.13 % 90.70 % 81.92 % 0.1 s GPU @ >3.5 Ghz (Python + C/C++)
K. Shin: Improving a Quality of 3D Object Detection by Spatial Transformation Mechanism. arXiv preprint arXiv:1910.04853 2019.
48 Roadstar.ai 87.12 % 92.42 % 81.88 % 0.08 s GPU @ 2.0 Ghz (Python)
49 PCSC-Net 87.00 % 90.98 % 82.99 % 0.04 s 1 core @ 2.5 Ghz (C/C++)
50 Alibaba-AILabsX
This method makes use of Velodyne laser scans.
 87.00 % 92.50 % 79.61 % 0.05 s 1 core @ >3.5 Ghz (C/C++)
51 SARPNET 86.92 % 92.21 % 81.68 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
Y. Ye, H. Chen, C. Zhang, X. Hao and Z. Zhang: SARPNET: Shape Attention Regional Proposal Network for LiDAR-based 3D Object Detection. Neurocomputing 2019.
52 TBA 86.85 % 90.51 % 83.05 % 0.07 s 1 core @ 2.5 Ghz (Python)
53 ARPNET 86.81 % 90.06 % 79.41 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
Y. Ye, C. Zhang and X. Hao: ARPNET: attention region proposal network for 3D object detection. Science China Information Sciences 2019.
54 PointPiallars_SECA 86.79 % 90.15 % 82.87 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
55 C-GCN 86.78 % 91.11 % 80.09 % 0.147 s GPU @ V100 (Python)
56 Tencent_ADlab_Lidar
This method makes use of Velodyne laser scans.
 86.72 % 90.27 % 81.35 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
57 SRF 86.60 % 91.90 % 81.43 % 0.05 s GPU @ 2.5 Ghz (Python + C/C++)
58 PointPillars
This method makes use of Velodyne laser scans.
 code 86.56 % 90.07 % 82.81 % 16 ms 1080ti GPU and Intel i7 CPU
A. Lang, S. Vora, H. Caesar, L. Zhou, J. Yang and O. Beijbom: PointPillars: Fast Encoders for Object Detection from Point Clouds. CVPR 2019.
59 TANet code 86.54 % 91.58 % 81.19 % 0.035s GPU @ 2.5 Ghz (Python + C/C++)
Z. Liu, X. Zhao, T. Huang, R. Hu, Y. Zhou and X. Bai: TANet: Robust 3D Object Detection from Point Clouds with Triple Attention. AAAI 2020.
60 A-VoxelNet 86.53 % 89.94 % 79.08 % 0.029 s GPU @ 2.5 Ghz (Python)
61 PointRCNN-deprecated
This method makes use of Velodyne laser scans.
 86.52 % 92.51 % 81.39 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
62 SCNet
This method makes use of Velodyne laser scans.
 86.48 % 90.07 % 81.30 % 0.04 s GPU @ 3.0 Ghz (Python)
Z. Wang, H. Fu, L. Wang, L. Xiao and B. Dai: SCNet: Subdivision Coding Network for Object Detection Based on 3D Point Cloud. IEEE Access 2019.
63 MMV 86.46 % 90.04 % 79.04 % 0.4 s GPU @ 2.5 Ghz (C/C++)
64 RUC 86.46 % 90.06 % 82.20 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
65 DDB
This method makes use of Velodyne laser scans.
 86.45 % 89.91 % 82.21 % 0.05 s GPU @ 2.5 Ghz (Python)
66 PPFNet code 86.44 % 92.35 % 81.48 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
67 autonet 86.42 % 89.81 % 81.25 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
68 HR-SECOND code 86.40 % 91.68 % 81.40 % 0.11 s 1 core @ 2.5 Ghz (Python + C/C++)
69 SECOND-V1.5
This method makes use of Velodyne laser scans.
 code 86.37 % 91.81 % 81.04 % 0.04 s GPU @ 2.0 Ghz (Python + C/C++)
70 SegVoxelNet 86.37 % 91.62 % 83.04 % 0.04 s 1 core @ 2.5 Ghz (Python)
H. Yi, S. Shi, M. Ding, J. Sun, K. Xu, H. Zhou, Z. Wang, S. Li and G. Wang: SegVoxelNet: Exploring Semantic Context and Depth-aware Features for 3D Vehicle Detection from Point Cloud. ICRA 2020.
71 VOXEL_FPN_HR 86.36 % 90.28 % 81.20 % 0.12 s 8 cores @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
72 CP
This method makes use of Velodyne laser scans.
 86.30 % 92.14 % 82.97 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
73 Bit 86.27 % 89.74 % 81.19 % 0.11 s 1 core @ 2.5 Ghz (C/C++)
74 FOFNet
This method makes use of Velodyne laser scans.
 86.22 % 90.09 % 78.96 % 0.04 s GPU @ 2.5 Ghz (Python)
75 NU-optim 86.22 % 91.62 % 80.81 % 0.04 s GPU @ >3.5 Ghz (Python)
76 3D IoU Loss
This method makes use of Velodyne laser scans.
 86.22 % 91.36 % 81.20 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
D. Zhou, J. Fang, X. Song, C. Guan, J. Yin, Y. Dai and R. Yang: IoU Loss for 2D/3D Object Detection. International Conference on 3D Vision (3DV) 2019.
77 MP 86.16 % 90.24 % 78.86 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
78 FCY
This method makes use of Velodyne laser scans.
 86.11 % 89.74 % 80.99 % 0.02 s GPU @ 2.5 Ghz (Python)
79 R-GCN 86.05 % 91.91 % 81.05 % 0.16 s GPU @ 2.5 Ghz (Python)
80 UberATG-PIXOR++
This method makes use of Velodyne laser scans.
 86.01 % 93.28 % 80.11 % 0.035 s GPU @ 2.5 Ghz (Python)
B. Yang, M. Liang and R. Urtasun: HDNET: Exploiting HD Maps for 3D Object Detection. 2nd Conference on Robot Learning (CoRL) 2018.
81 PTS
This method makes use of Velodyne laser scans.
 code 85.95 % 91.42 % 80.81 % 0.01 s 1 core @ 2.5 Ghz (C/C++)
82 PPBA 85.85 % 91.30 % 80.92 % NA s GPU @ 2.5 Ghz (Python)
83 TBU 85.85 % 91.30 % 80.92 % NA s GPU @ 2.5 Ghz (Python)
84 F-ConvNet
This method makes use of Velodyne laser scans.
 code 85.84 % 91.51 % 76.11 % 0.47 s GPU @ 2.5 Ghz (Python + C/C++)
Z. Wang and K. Jia: Frustum ConvNet: Sliding Frustums to Aggregate Local Point-Wise Features for Amodal 3D Object Detection. IROS 2019.
85 RUC code 85.84 % 88.54 % 81.15 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
86 BVVF 85.83 % 91.20 % 80.76 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
87 PI-RCNN 85.81 % 91.44 % 81.00 % 0.1 s 1 core @ 2.5 Ghz (Python)
88 SAANet 85.69 % 91.72 % 78.77 % 0.10 s 1 core @ 2.5 Ghz (Python)
89 SFB-SECOND 85.63 % 91.38 % 78.60 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
90 UberATG-ContFuse
This method makes use of Velodyne laser scans.
 85.35 % 94.07 % 75.88 % 0.06 s GPU @ 2.5 Ghz (Python)
M. Liang, B. Yang, S. Wang and R. Urtasun: Deep Continuous Fusion for Multi-Sensor 3D Object Detection. ECCV 2018.
91 AVOD
This method makes use of Velodyne laser scans.
 code 84.95 % 89.75 % 78.32 % 0.08 s Titan X (pascal)
J. Ku, M. Mozifian, J. Lee, A. Harakeh and S. Waslander: Joint 3D Proposal Generation and Object Detection from View Aggregation. IROS 2018.
92 AVOD-FPN
This method makes use of Velodyne laser scans.
 code 84.82 % 90.99 % 79.62 % 0.1 s Titan X (Pascal)
J. Ku, M. Mozifian, J. Lee, A. Harakeh and S. Waslander: Joint 3D Proposal Generation and Object Detection from View Aggregation. IROS 2018.
93 Prune 84.81 % 90.48 % 77.40 % 0.11 s 1 core @ 2.5 Ghz (C/C++)
94 autoRUC 84.80 % 90.44 % 77.43 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
95 F-PointNet
This method makes use of Velodyne laser scans.
 code 84.67 % 91.17 % 74.77 % 0.17 s GPU @ 3.0 Ghz (Python)
C. Qi, W. Liu, C. Wu, H. Su and L. Guibas: Frustum PointNets for 3D Object Detection from RGB-D Data. arXiv preprint arXiv:1711.08488 2017.
96 PAD 84.46 % 88.66 % 80.61 % 0.15 s 1 core @ 2.5 Ghz (Python)
97 RUC code 84.40 % 89.11 % 79.33 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
98 MVSLN 84.26 % 90.30 % 78.94 % 0.1s s 1 core @ 2.5 Ghz (C/C++)
99 PP-3D 84.11 % 89.16 % 76.81 % 0.1 s 1 core @ 2.5 Ghz (Python)
100 3DBN
This method makes use of Velodyne laser scans.
 83.94 % 89.66 % 76.50 % 0.13s 1080Ti (Python+C/C++)
X. Li, J. Guivant, N. Kwok and Y. Xu: 3D Backbone Network for 3D Object Detection. CoRR 2019.
101 RADNet-Fusion
This method makes use of Velodyne laser scans.
 83.84 % 91.81 % 78.80 % 0.1 s 1 core @ 2.5 Ghz (Python)
102 RADNet-LIDAR
This method makes use of Velodyne laser scans.
 83.74 % 92.43 % 77.00 % 0.1 s 1 core @ 2.5 Ghz (Python)
103 3DNN 83.68 % 88.06 % 77.00 % 0.09 s GPU @ 2.5 Ghz (Python)
104 SCANet 83.19 % 88.68 % 77.84 % 0.17 s >8 cores @ 2.5 Ghz (Python)
105 FailNet-Fusion
This method makes use of Velodyne laser scans.
 82.78 % 93.20 % 75.84 % 0.1 s 1 core @ 2.5 Ghz (Python)
106 yl_net 82.70 % 87.27 % 80.23 % 0.03 s GPU @ 2.5 Ghz (Python)
107 MLOD
This method makes use of Velodyne laser scans.
 code 82.68 % 90.25 % 77.97 % 0.12 s GPU @ 1.5 Ghz (Python)
J. Deng and K. Czarnecki: MLOD: A multi-view 3D object detection based on robust feature fusion method. arXiv preprint arXiv:1909.04163 2019.
108 SECA 82.58 % 90.37 % 75.75 % 1 s GPU @ 2.5 Ghz (Python)
109 FailNet-LIDAR
This method makes use of Velodyne laser scans.
 82.41 % 92.85 % 75.39 % 0.1 s 1 core @ 2.5 Ghz (Python)
110 SDP-Net-s 81.93 % 86.57 % 75.83 % 0.01 s GPU @ 2.5 Ghz (Python)
111 NLK 81.93 % 89.93 % 76.80 % 0.02 s 1 core @ 2.5 Ghz (Python)
112 RTL3D 81.63 % 89.55 % 76.63 % 0.02 s GPU @ 2.5 Ghz (Python)
113 PointRes
This method makes use of Velodyne laser scans.
This method makes use of GPS/IMU information.
This is an online method (no batch processing).
 81.60 % 90.60 % 76.03 % 0.013 s 1 core @ 2.5 Ghz (Python + C/C++)
114 RuiRUC 80.20 % 86.90 % 67.77 % 0.12 s 1 core @ 2.5 Ghz (Python)
115 UberATG-PIXOR
This method makes use of Velodyne laser scans.
 80.01 % 83.97 % 74.31 % 0.035 s TITAN Xp (Python)
B. Yang, W. Luo and R. Urtasun: PIXOR: Real-time 3D Object Detection from Point Clouds. CVPR 2018.
116 MV3D (LIDAR)
This method makes use of Velodyne laser scans.
 78.98 % 86.49 % 72.23 % 0.24 s GPU @ 2.5 Ghz (Python + C/C++)
X. Chen, H. Ma, J. Wan, B. Li and T. Xia: Multi-View 3D Object Detection Network for Autonomous Driving. CVPR 2017.
117 MV3D
This method makes use of Velodyne laser scans.
 78.93 % 86.62 % 69.80 % 0.36 s GPU @ 2.5 Ghz (Python + C/C++)
X. Chen, H. Ma, J. Wan, B. Li and T. Xia: Multi-View 3D Object Detection Network for Autonomous Driving. CVPR 2017.
118 Multi-3D
This method makes use of Velodyne laser scans.
 78.45 % 85.99 % 67.14 % 0.15 s 1 core @ 2.5 Ghz (C/C++)
119 VoxelNet(Unofficial) 78.39 % 87.95 % 71.29 % 0.5 s GPU @ 2.0 Ghz (Python)
120 ANM 75.40 % 84.78 % 61.28 % 0.12 s 1 core @ 2.5 Ghz (Python)
121 LaserNet 74.52 % 79.19 % 68.45 % 12 ms GPU @ 2.5 Ghz (C/C++)
G. Meyer, A. Laddha, E. Kee, C. Vallespi-Gonzalez and C. Wellington: LaserNet: An Efficient Probabilistic 3D Object Detector for Autonomous Driving. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2019.
122 PL++ (SDN+GDC)
This method uses stereo information.
This method makes use of Velodyne laser scans.
 code 73.80 % 84.61 % 65.59 % 0.6 s GPU @ 2.5 Ghz (C/C++)
Y. You, Y. Wang, W. Chao, D. Garg, G. Pleiss, B. Hariharan, M. Campbell and K. Weinberger: Pseudo-LiDAR++: Accurate Depth for 3D Object Detection in Autonomous Driving. International Conference on Learning Representations 2020.
123 anm 73.63 % 82.59 % 62.87 % 3 s 1 core @ 2.5 Ghz (C/C++)
124 A3DODWTDA
This method makes use of Velodyne laser scans.
 code 73.26 % 79.58 % 62.77 % 0.08 s GPU @ 3.0 Ghz (Python)
F. Gustafsson and E. Linder-Norén: Automotive 3D Object Detection Without Target Domain Annotations. 2018.
125 avodC 72.78 % 84.61 % 66.02 % 0.1 s GPU @ 2.5 Ghz (Python)
126 E-VoxelNet 69.69 % 81.10 % 60.88 % 0.1 s GPU @ 2.5 Ghz (Python)
127 Complexer-YOLO
This method makes use of Velodyne laser scans.
 68.96 % 77.24 % 64.95 % 0.06 s GPU @ 3.5 Ghz (C/C++)
M. Simon, K. Amende, A. Kraus, J. Honer, T. Samann, H. Kaulbersch, S. Milz and H. Michael Gross: Complexer-YOLO: Real-Time 3D Object Detection and Tracking on Semantic Point Clouds. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops 2019.
128 TopNet-Retina
This method makes use of Velodyne laser scans.
 68.16 % 80.16 % 63.43 % 52ms GeForce 1080Ti (tensorflow-gpu, v1.12)
S. Wirges, T. Fischer, C. Stiller and J. Frias: Object Detection and Classification in Occupancy Grid Maps Using Deep Convolutional Networks. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
129 SF
This method uses stereo information.
This method makes use of Velodyne laser scans.
 65.74 % 74.20 % 58.35 % 0.5 s 1 core @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
130 DSGN
This method uses stereo information.
 code 65.05 % 82.90 % 56.60 % 0.67 s NVIDIA Tesla V100
Y. Chen, S. Liu, X. Shen and J. Jia: DSGN: Deep Stereo Geometry Network for 3D Object Detection. arXiv preprint arXiv:2001.03398 2020.
131 TopNet-DecayRate
This method makes use of Velodyne laser scans.
 64.60 % 79.74 % 58.04 % 92 ms NVIDIA GeForce 1080 Ti (tensorflow-gpu)
S. Wirges, T. Fischer, C. Stiller and J. Frias: Object Detection and Classification in Occupancy Grid Maps Using Deep Convolutional Networks. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
132 3D FCN
This method makes use of Velodyne laser scans.
 61.67 % 70.62 % 55.61 % >5 s 1 core @ 2.5 Ghz (C/C++)
B. Li: 3D Fully Convolutional Network for Vehicle Detection in Point Cloud. IROS 2017.
133 BirdNet
This method makes use of Velodyne laser scans.
 59.83 % 84.17 % 57.35 % 0.11 s Titan Xp GPU
J. Beltrán, C. Guindel, F. Moreno, D. Cruzado, F. García and A. Escalera: BirdNet: A 3D Object Detection Framework from LiDAR Information. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
134 TopNet-UncEst
This method makes use of Velodyne laser scans.
 59.67 % 72.05 % 51.67 % 0.09 s NVIDIA GeForce 1080 Ti (tensorflow-gpu)
S. Wirges, M. Braun, M. Lauer and C. Stiller: Capturing Object Detection Uncertainty in Multi-Layer Grid Maps. 2019.
135 Pseudo-LiDAR E2E
This method uses stereo information.
 58.84 % 79.58 % 52.06 % 0.4 s GPU @ 2.5 Ghz (Python)
136 Pseudo-LiDAR++
This method uses stereo information.
 code 58.01 % 78.31 % 51.25 % 0.4 s GPU @ 2.5 Ghz (Python)
Y. You, Y. Wang, W. Chao, D. Garg, G. Pleiss, B. Hariharan, M. Campbell and K. Weinberger: Pseudo-LiDAR++: Accurate Depth for 3D Object Detection in Autonomous Driving. International Conference on Learning Representations 2020.
137 ZoomNet
This method uses stereo information.
 code 54.91 % 72.94 % 44.14 % 0.3 s 1 core @ 2.5 Ghz (C/C++)
L. Z. Xu: ZoomNet: Part-Aware Adaptive Zooming Neural Network for 3D Object Detection. Proceedings of the AAAI Conference on Artificial Intelligence 2020.
138 VoxelJones code 53.96 % 66.21 % 47.66 % .18 s 1 core @ 2.5 Ghz (Python + C/C++)
M. Motro and J. Ghosh: Vehicular Multi-object Tracking with Persistent Detector Failures. arXiv preprint arXiv:1907.11306 2019.
139 TopNet-HighRes
This method makes use of Velodyne laser scans.
 53.05 % 67.84 % 46.99 % 101ms NVIDIA GeForce 1080 Ti (tensorflow-gpu)
S. Wirges, T. Fischer, C. Stiller and J. Frias: Object Detection and Classification in Occupancy Grid Maps Using Deep Convolutional Networks. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
140 Disp R-CNN (velo)
This method uses stereo information.
 52.34 % 74.07 % 43.77 % 0.42 s GPU @ 2.5 Ghz (Python + C/C++)
141 Disp R-CNN
This method uses stereo information.
 52.34 % 73.82 % 43.64 % 0.42 s GPU @ 2.5 Ghz (Python + C/C++)
142 OC Stereo
This method uses stereo information.
 51.47 % 68.89 % 42.97 % 0.35 s 1 core @ 2.5 Ghz (Python + C/C++)
A. Pon, J. Ku, C. Li and S. Waslander: Object-Centric Stereo Matching for 3D Object Detection. arXiv preprint arXiv:1909.07566 2019.
143 stereo_sa
This method uses stereo information.
 49.61 % 71.47 % 42.71 % 0.3 s GPU @ 2.5 Ghz (Python)
144 RT3DStereo
This method uses stereo information.
 46.82 % 58.81 % 38.38 % 0.08 s GPU @ 2.5 Ghz (C/C++)
H. Königshof, N. Salscheider and C. Stiller: Realtime 3D Object Detection for Automated Driving Using Stereo Vision and Semantic Information. Proc. IEEE Intl. Conf. Intelligent Transportation Systems 2019.
145 Pseudo-Lidar
This method uses stereo information.
 code 45.00 % 67.30 % 38.40 % 0.4 s GPU @ 2.5 Ghz (Python + C/C++)
Y. Wang, W. Chao, D. Garg, B. Hariharan, M. Campbell and K. Weinberger: Pseudo-LiDAR From Visual Depth Estimation: Bridging the Gap in 3D Object Detection for Autonomous Driving. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2019.
146 RT3D
This method makes use of Velodyne laser scans.
 44.00 % 56.44 % 42.34 % 0.09 s GPU @ 1.8Ghz
Y. Zeng, Y. Hu, S. Liu, J. Ye, Y. Han, X. Li and N. Sun: RT3D: Real-Time 3-D Vehicle Detection in LiDAR Point Cloud for Autonomous Driving. IEEE Robotics and Automation Letters 2018.
147 m-prcnn
This method uses stereo information.
 42.81 % 67.82 % 33.63 % 0.43 s 1 core @ 2.5 Ghz (Python)
148 IDA-3D
This method uses stereo information.
 42.47 % 61.87 % 34.59 % 0.08 s 1 core @ 2.5 Ghz (Python + C/C++)
149 Stereo R-CNN
This method uses stereo information.
 code 41.31 % 61.92 % 33.42 % 0.3 s GPU @ 2.5 Ghz (Python)
P. Li, X. Chen and S. Shen: Stereo R-CNN based 3D Object Detection for Autonomous Driving. CVPR 2019.
150 Licar
This method makes use of Velodyne laser scans.
 38.47 % 46.67 % 35.78 % 0.09 s GPU @ 2.0 Ghz (Python)
151 ASOD 33.63 % 54.61 % 26.76 % 0.28 s GPU @ 2.5 Ghz (Python)
152 StereoFENet
This method uses stereo information.
 32.96 % 49.29 % 25.90 % 0.15 s 1 core @ 3.5 Ghz (Python)
W. Bao, B. Xu and Z. Chen: MonoFENet: Monocular 3D Object Detection with Feature Enhancement Networks. IEEE Transactions on Image Processing 2019.
153 DPSM 19.33 % 28.63 % 15.31 % 0.1 s GPU @ 2.5 Ghz (Python)
154 RefinedMPL 17.60 % 28.08 % 13.95 % 0.15 s GPU @ 2.5 Ghz (Python + C/C++)
J. Vianney, S. Aich and B. Liu: RefinedMPL: Refined Monocular PseudoLiDAR for 3D Object Detection in Autonomous Driving. arXiv preprint arXiv:1911.09712 2019.
155 AM3D 17.32 % 25.03 % 14.91 % 0.4 s GPU @ 2.5 Ghz (Python + C/C++)
X. Ma, Z. Wang, H. Li, P. Zhang, W. Ouyang and X. Fan: Accurate Monocular Object Detection via Color- Embedded 3D Reconstruction for Autonomous Driving. Proceedings of the IEEE international Conference on Computer Vision (ICCV) 2019.
156 D4LCN code 16.02 % 22.51 % 12.55 % 0.2 s GPU @ 2.5 Ghz (Python + C/C++)
M. Ding, Y. Huo, H. Yi, Z. Wang, J. Shi, Z. Lu and P. Luo: Learning Depth-Guided Convolutions for Monocular 3D Object Detection. arXiv preprint arXiv:1912.04799 2019.
157 YoloMono3D 15.01 % 24.39 % 10.67 % 0.05 s GPU @ 2.5 Ghz (Python)
158 MonoPair 14.83 % 19.28 % 12.89 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
159 Decoupled-3D 14.82 % 23.16 % 11.25 % 0.08 s GPU @ 2.5 Ghz (C/C++)
160 Decoupled-3D v2 14.66 % 24.62 % 11.46 % 0.08 s GPU @ 2.5 Ghz (C/C++)
161 MonoSS 14.52 % 20.91 % 12.63 % 0.03 s GPU @ 2.5 Ghz (Python + C/C++)
162 SMOKE 14.49 % 20.83 % 12.75 % 0.03 s GPU @ 2.5 Ghz (Python)
163 Mono3CN 14.17 % 19.82 % 12.30 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
164 RTM3D code 14.05 % 18.81 % 11.68 % 0.05 s GPU @ 1.0 Ghz (Python)
P. Li, H. Zhao, P. Liu and F. Cao: RTM3D: Real-time Monocular 3D Detection from Object Keypoints for Autonomous Driving. 2020.
165 Mono3D_PLiDAR code 13.92 % 21.27 % 11.25 % 0.1 s NVIDIA GeForce 1080 (pytorch)
X. Weng and K. Kitani: Monocular 3D Object Detection with Pseudo-LiDAR Point Cloud. arXiv:1903.09847 2019.
166 SS3D_HW 13.70 % 20.28 % 9.86 % 0.4 s GPU @ 2.5 Ghz (Python)
167 M3D-RPN code 13.67 % 21.02 % 10.23 % 0.16 s GPU @ 1.5 Ghz (Python)
G. Brazil and X. Liu: M3D-RPN: Monocular 3D Region Proposal Network for Object Detection . ICCV 2019 .
168 MonoDIS 13.19 % 17.23 % 11.12 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
169 CSoR
This method makes use of Velodyne laser scans.
 13.07 % 18.67 % 10.34 % 3.5 s 4 cores @ >3.5 Ghz (Python + C/C++)
L. Plotkin: PyDriver: Entwicklung eines Frameworks für räumliche Detektion und Klassifikation von Objekten in Fahrzeugumgebung. 2015.
170 RAR-Net 13.01 % 20.63 % 10.19 % 0.5 s 1 core @ 2.5 Ghz (C/C++)
171 MonoPSR code 12.58 % 18.33 % 9.91 % 0.2 s GPU @ 3.5 Ghz (Python)
J. Ku*, A. Pon* and S. Waslander: Monocular 3D Object Detection Leveraging Accurate Proposals and Shape Reconstruction. CVPR 2019.
172 PG-MonoNet 12.45 % 19.79 % 9.68 % 0.19 s GPU @ 2.5 Ghz (Python)
173 SS3D 11.52 % 16.33 % 9.93 % 48 ms Tesla V100 (Python)
E. Jörgensen, C. Zach and F. Kahl: Monocular 3D Object Detection and Box Fitting Trained End-to-End Using Intersection-over-Union Loss. CoRR 2019.
174 MonoGRNet code 11.17 % 18.19 % 8.73 % 0.04s NVIDIA P40
Z. Qin, J. Wang and Y. Lu: MonoGRNet: A Geometric Reasoning Network for 3D Object Localization. The Thirty-Third AAAI Conference on Artificial Intelligence (AAAI-19) 2019.
175 MonoFENet 11.03 % 17.03 % 9.05 % 0.15 s 1 core @ 3.5 Ghz (Python)
W. Bao, B. Xu and Z. Chen: MonoFENet: Monocular 3D Object Detection with Feature Enhancement Networks. IEEE Transactions on Image Processing 2019.
176 RADNet-Mono 10.57 % 15.22 % 8.74 % 0.1 s 1 core @ 2.5 Ghz (Python)
177 OACV 10.13 % 16.24 % 8.28 % 0.23 s GPU @ 2.5 Ghz (Python)
178 mylsi-faster-rcnn 9.96 % 15.55 % 8.11 % 0.3 s 1 core @ 2.5 Ghz (Python)
179 FailNet-Mono 9.11 % 14.41 % 7.11 % 0.1 s 1 core @ 2.5 Ghz (Python)
180 A3DODWTDA (image) code 8.66 % 10.37 % 7.06 % 0.8 s GPU @ 3.0 Ghz (Python)
F. Gustafsson and E. Linder-Norén: Automotive 3D Object Detection Without Target Domain Annotations. 2018.
181 mymask-rcnn 8.29 % 15.56 % 6.53 % 0.3 s 1 core @ 2.5 Ghz (Python)
182 Shift R-CNN (mono) code 6.82 % 11.84 % 5.27 % 0.25 s GPU @ 1.5 Ghz (Python)
A. Naiden, V. Paunescu, G. Kim, B. Jeon and M. Leordeanu: Shift R-CNN: Deep Monocular 3D Object Detection With Closed-form Geometric Constraints. ICIP 2019.
183 GS3D 6.08 % 8.41 % 4.94 % 2 s 1 core @ 2.5 Ghz (C/C++)
B. Li, W. Ouyang, L. Sheng, X. Zeng and X. Wang: GS3D: An Efficient 3D Object Detection Framework for Autonomous Driving. IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2019.
184 MVRA + I-FRCNN+ 5.84 % 9.05 % 4.50 % 0.18 s GPU @ 2.5 Ghz (Python)
H. Choi, H. Kang and Y. Hyun: Multi-View Reprojection Architecture for Orientation Estimation. The IEEE International Conference on Computer Vision (ICCV) Workshops 2019.
185 ROI-10D 4.91 % 9.78 % 3.74 % 0.2 s GPU @ 3.5 Ghz (Python)
F. Manhardt, W. Kehl and A. Gaidon: ROI-10D: Monocular Lifting of 2D Detection to 6D Pose and Metric Shape. Computer Vision and Pattern Recognition (CVPR) 2019.
186 3D-GCK 4.57 % 5.79 % 3.64 % 24 ms Tesla V100
187 FQNet 3.23 % 5.40 % 2.46 % 0.5 s 1 core @ 2.5 Ghz (Python)
L. Liu, J. Lu, C. Xu, Q. Tian and J. Zhou: Deep Fitting Degree Scoring Network for Monocular 3D Object Detection. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition 2019.
188 3D-SSMFCNN code 2.63 % 3.20 % 2.40 % 0.1 s GPU @ 1.5 Ghz (C/C++)
L. Novak: Vehicle Detection and Pose Estimation for Autonomous Driving. 2017.
189 monoref3d 1.70 % 3.22 % 1.21 % 0.1 s 1 core @ 2.5 Ghz (Python)
190 ref3D 1.70 % 3.22 % 1.21 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
191 3DVSSD 1.31 % 1.74 % 1.08 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
192 VeloFCN
This method makes use of Velodyne laser scans.
 0.14 % 0.02 % 0.21 % 1 s GPU @ 2.5 Ghz (Python + C/C++)
B. Li, T. Zhang and T. Xia: Vehicle Detection from 3D Lidar Using Fully Convolutional Network. RSS 2016 .
193 ANM 0.00 % 0.00 % 0.00 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
194 ref3D 0.00 % 0.00 % 0.00 % 0.1 s 1 core @ 2.5 Ghz (Python)
195 multi-task CNN 0.00 % 0.00 % 0.00 % 25.1 ms GPU @ 2.0 Ghz (Python)
M. Oeljeklaus, F. Hoffmann and T. Bertram: A Fast Multi-Task CNN for Spatial Understanding of Traffic Scenes. IEEE Intelligent Transportation Systems Conference 2018.
196 mBoW
This method makes use of Velodyne laser scans.
 0.00 % 0.00 % 0.00 % 10 s 1 core @ 2.5 Ghz (C/C++)
J. Behley, V. Steinhage and A. Cremers: Laser-based Segment Classification Using a Mixture of Bag-of-Words. Proc. of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2013.
Table as LaTeX | Only published Methods

Pedestrian


Method Setting Code Moderate Easy Hard Runtime Environment
1 DENFIDet 51.96 % 61.15 % 49.03 % 0.02 s GPU @ 2.5 Ghz (C/C++)
2 A-VoxelNet 51.79 % 61.34 % 47.93 % 0.029 s GPU @ 2.5 Ghz (Python)
3 TANet code 51.38 % 60.85 % 47.54 % 0.035s GPU @ 2.5 Ghz (Python + C/C++)
Z. Liu, X. Zhao, T. Huang, R. Hu, Y. Zhou and X. Bai: TANet: Robust 3D Object Detection from Point Clouds with Triple Attention. AAAI 2020.
4 Noah CV Lab - SSL 50.66 % 57.27 % 46.55 % 0.1 s GPU @ 2.5 Ghz (Python)
5 MMLab PV-RCNN
This method makes use of Velodyne laser scans.
 50.57 % 59.86 % 46.74 % 0.08 s 1 core @ 2.5 Ghz (Python + C/C++)
S. Shi, C. Guo, L. Jiang, Z. Wang, J. Shi, X. Wang and H. Li: PV-RCNN: Point-Voxel Feature Set Abstraction for 3D Object Detection. arXiv preprint arXiv:1912.13192 2019.
6 VMVS
This method makes use of Velodyne laser scans.
 50.34 % 60.34 % 46.45 % 0.25 s GPU @ 2.5 Ghz (Python)
J. Ku, A. Pon, S. Walsh and S. Waslander: Improving 3D object detection for pedestrians with virtual multi-view synthesis orientation estimation. IROS 2019.
7 AVOD-FPN
This method makes use of Velodyne laser scans.
 code 50.32 % 58.49 % 46.98 % 0.1 s Titan X (Pascal)
J. Ku, M. Mozifian, J. Lee, A. Harakeh and S. Waslander: Joint 3D Proposal Generation and Object Detection from View Aggregation. IROS 2018.
8 3DSSD 49.94 % 60.54 % 45.73 % 0.04 s GPU @ 2.5 Ghz (Python + C/C++)
9 PointPainting
This method makes use of Velodyne laser scans.
 49.93 % 58.70 % 46.29 % 0.4 s GPU @ 2.5 Ghz (Python + C/C++)
S. Vora, A. Lang, B. Helou and O. Beijbom: PointPainting: Sequential Fusion for 3D Object Detection. arXiv preprint arXiv:1911.10150 2019.
10 MMLab-PartA^2
This method makes use of Velodyne laser scans.
 code 49.81 % 59.04 % 45.92 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
S. Shi, Z. Wang, J. Shi, X. Wang and H. Li: From Points to Parts: 3D Object Detection from Point Cloud with Part-aware and Part-aggregation Network. arXiv preprint arXiv:1907.03670 2019.
11 F-PointNet
This method makes use of Velodyne laser scans.
 code 49.57 % 57.13 % 45.48 % 0.17 s GPU @ 3.0 Ghz (Python)
C. Qi, W. Liu, C. Wu, H. Su and L. Guibas: Frustum PointNets for 3D Object Detection from RGB-D Data. arXiv preprint arXiv:1711.08488 2017.
12 OHS-Direct 49.48 % 55.90 % 45.79 % 0.03 s 1 core @ 2.5 Ghz (Python + C/C++)
Q. Chen, L. Sun, Z. Wang, K. Jia and A. Yuille: Object as Hotspots: An Anchor-Free 3D Object Detection Approach via Firing of Hotspots. 2019.
13 PPBA 49.34 % 57.23 % 46.86 % NA s GPU @ 2.5 Ghz (Python)
14 F-ConvNet
This method makes use of Velodyne laser scans.
 code 48.96 % 57.04 % 44.33 % 0.47 s GPU @ 2.5 Ghz (Python + C/C++)
Z. Wang and K. Jia: Frustum ConvNet: Sliding Frustums to Aggregate Local Point-Wise Features for Amodal 3D Object Detection. IROS 2019.
15 HVNet 48.86 % 54.84 % 46.33 % 0.03 s GPU @ 2.0 Ghz (Python)
16 CentrNet-v1
This method makes use of Velodyne laser scans.
 48.78 % 57.58 % 45.94 % 0.03 s GPU @ 2.5 Ghz (Python)
17 STD 48.72 % 60.02 % 44.55 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
Z. Yang, Y. Sun, S. Liu, X. Shen and J. Jia: STD: Sparse-to-Dense 3D Object Detector for Point Cloud. ICCV 2019.
18 PointPillars
This method makes use of Velodyne laser scans.
 code 48.64 % 57.60 % 45.78 % 16 ms 1080ti GPU and Intel i7 CPU
A. Lang, S. Vora, H. Caesar, L. Zhou, J. Yang and O. Beijbom: PointPillars: Fast Encoders for Object Detection from Point Clouds. CVPR 2019.
19 DDB
This method makes use of Velodyne laser scans.
 48.35 % 57.68 % 45.44 % 0.05 s GPU @ 2.5 Ghz (Python)
20 PiP 48.14 % 56.16 % 45.27 % 0.05 s 1 core @ 2.5 Ghz (Python)
21 PPFNet code 47.92 % 55.04 % 44.95 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
22 LDAM 47.35 % 52.08 % 45.23 % 24 ms GTX 1080 ti GPU
23 Tencent_ADlab_Lidar
This method makes use of Velodyne laser scans.
 47.24 % 56.06 % 44.61 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
24 Point-GNN
This method makes use of Velodyne laser scans.
 47.07 % 55.36 % 44.61 % 0.6 s GPU @ 2.5 Ghz (Python)
25 TBU 46.76 % 55.15 % 44.60 % NA s GPU @ 2.5 Ghz (Python)
26 PP-3D 46.74 % 56.74 % 44.01 % 0.1 s 1 core @ 2.5 Ghz (Python)
27 SCNet
This method makes use of Velodyne laser scans.
 46.73 % 56.87 % 42.74 % 0.04 s GPU @ 3.0 Ghz (Python)
Z. Wang, H. Fu, L. Wang, L. Xiao and B. Dai: SCNet: Subdivision Coding Network for Object Detection Based on 3D Point Cloud. IEEE Access 2019.
28 MMLab-PointRCNN
This method makes use of Velodyne laser scans.
 code 46.13 % 54.77 % 42.84 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
S. Shi, X. Wang and H. Li: Pointrcnn: 3d object proposal generation and detection from point cloud. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition 2019.
29 Multi-3D
This method makes use of Velodyne laser scans.
 46.09 % 54.37 % 41.42 % 0.15 s 1 core @ 2.5 Ghz (C/C++)
30 ARPNET 45.92 % 55.48 % 42.54 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
Y. Ye, C. Zhang and X. Hao: ARPNET: attention region proposal network for 3D object detection. Science China Information Sciences 2019.
31 epBRM
This method makes use of Velodyne laser scans.
 code 45.49 % 52.48 % 42.75 % 0.10 s 1 core @ 2.5 Ghz (C/C++)
K. Shin: Improving a Quality of 3D Object Detection by Spatial Transformation Mechanism. arXiv preprint arXiv:1910.04853 2019.
32 MLOD
This method makes use of Velodyne laser scans.
 code 45.40 % 55.09 % 41.42 % 0.12 s GPU @ 1.5 Ghz (Python)
J. Deng and K. Czarnecki: MLOD: A multi-view 3D object detection based on robust feature fusion method. arXiv preprint arXiv:1909.04163 2019.
33 OHS-Dense 44.59 % 50.87 % 42.14 % 0.03 s 1 core @ 2.5 Ghz (Python/C++)
Q. Chen, L. Sun, Z. Wang, K. Jia and A. Yuille: Object as Hotspots: An Anchor-Free 3D Object Detection Approach via Firing of Hotspots. 2019.
34 Roadstar.ai 44.35 % 49.63 % 41.39 % 0.08 s GPU @ 2.0 Ghz (Python)
35 FCY
This method makes use of Velodyne laser scans.
 43.88 % 51.21 % 41.41 % 0.02 s GPU @ 2.5 Ghz (Python)
36 SCANet 42.81 % 53.84 % 38.94 % 0.17 s >8 cores @ 2.5 Ghz (Python)
37 FOFNet
This method makes use of Velodyne laser scans.
 42.31 % 51.39 % 38.81 % 0.04 s GPU @ 2.5 Ghz (Python)
38 VOXEL_FPN_HR 41.62 % 50.18 % 38.30 % 0.12 s 8 cores @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
39 deprecated 41.32 % 53.09 % 37.16 % 0.05 s GPU @ 2.0 Ghz (Python)
40 HR-SECOND code 40.06 % 50.05 % 36.47 % 0.11 s 1 core @ 2.5 Ghz (Python + C/C++)
41 AB3DMOT
This method makes use of Velodyne laser scans.
This is an online method (no batch processing).
 code 38.79 % 47.51 % 35.85 % 0.0047s 1 core @ 2.5 Ghz (python)
X. Weng and K. Kitani: A Baseline for 3D Multi-Object Tracking. arXiv:1907.03961 2019.
42 MP 38.77 % 47.59 % 35.50 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
43 anm 38.01 % 49.07 % 34.00 % 3 s 1 core @ 2.5 Ghz (C/C++)
44 CSW3D
This method makes use of Velodyne laser scans.
 37.96 % 49.27 % 33.83 % 0.03 s 4 cores @ 2.5 Ghz (C/C++)
J. Hu, T. Wu, H. Fu, Z. Wang and K. Ding: Cascaded Sliding Window Based Real-Time 3D Region Proposal for Pedestrian Detection. ROBIO 2019.
45 SparsePool code 34.15 % 43.33 % 31.78 % 0.13 s 8 cores @ 2.5 Ghz (Python)
Z. Wang, W. Zhan and M. Tomizuka: Fusing bird view lidar point cloud and front view camera image for deep object detection. arXiv preprint arXiv:1711.06703 2017.
46 SAANet 33.94 % 42.34 % 31.75 % 0.10 s 1 core @ 2.5 Ghz (Python)
47 AVOD
This method makes use of Velodyne laser scans.
 code 33.57 % 42.58 % 30.14 % 0.08 s Titan X (pascal)
J. Ku, M. Mozifian, J. Lee, A. Harakeh and S. Waslander: Joint 3D Proposal Generation and Object Detection from View Aggregation. IROS 2018.
48 SparsePool code 33.22 % 41.55 % 29.66 % 0.13 s 8 cores @ 2.5 Ghz (Python)
Z. Wang, W. Zhan and M. Tomizuka: Fusing bird view lidar point cloud and front view camera image for deep object detection. arXiv preprint arXiv:1711.06703 2017.
49 32.32 % 40.87 % 29.52 %
50 SF
This method uses stereo information.
This method makes use of Velodyne laser scans.
 29.77 % 37.16 % 26.61 % 0.5 s 1 core @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
51 BirdNet
This method makes use of Velodyne laser scans.
 23.06 % 28.20 % 21.65 % 0.11 s Titan Xp GPU
J. Beltrán, C. Guindel, F. Moreno, D. Cruzado, F. García and A. Escalera: BirdNet: A 3D Object Detection Framework from LiDAR Information. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
52 OC Stereo
This method uses stereo information.
 20.80 % 29.79 % 18.62 % 0.35 s 1 core @ 2.5 Ghz (Python + C/C++)
A. Pon, J. Ku, C. Li and S. Waslander: Object-Centric Stereo Matching for 3D Object Detection. arXiv preprint arXiv:1909.07566 2019.
53 DSGN
This method uses stereo information.
 code 20.75 % 26.61 % 18.86 % 0.67 s NVIDIA Tesla V100
Y. Chen, S. Liu, X. Shen and J. Jia: DSGN: Deep Stereo Geometry Network for 3D Object Detection. arXiv preprint arXiv:2001.03398 2020.
54 Complexer-YOLO
This method makes use of Velodyne laser scans.
 18.26 % 21.42 % 17.06 % 0.06 s GPU @ 3.5 Ghz (C/C++)
M. Simon, K. Amende, A. Kraus, J. Honer, T. Samann, H. Kaulbersch, S. Milz and H. Michael Gross: Complexer-YOLO: Real-Time 3D Object Detection and Tracking on Semantic Point Clouds. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops 2019.
55 TopNet-Retina
This method makes use of Velodyne laser scans.
 14.57 % 18.04 % 12.48 % 52ms GeForce 1080Ti (tensorflow-gpu, v1.12)
S. Wirges, T. Fischer, C. Stiller and J. Frias: Object Detection and Classification in Occupancy Grid Maps Using Deep Convolutional Networks. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
56 TopNet-HighRes
This method makes use of Velodyne laser scans.
 13.50 % 19.43 % 11.93 % 101ms NVIDIA GeForce 1080 Ti (tensorflow-gpu)
S. Wirges, T. Fischer, C. Stiller and J. Frias: Object Detection and Classification in Occupancy Grid Maps Using Deep Convolutional Networks. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
57 RefinedMPL 7.92 % 13.09 % 7.25 % 0.15 s GPU @ 2.5 Ghz (Python + C/C++)
J. Vianney, S. Aich and B. Liu: RefinedMPL: Refined Monocular PseudoLiDAR for 3D Object Detection in Autonomous Driving. arXiv preprint arXiv:1911.09712 2019.
58 MonoPair 7.04 % 10.99 % 6.29 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
59 TopNet-DecayRate
This method makes use of Velodyne laser scans.
 6.59 % 8.78 % 6.25 % 92 ms NVIDIA GeForce 1080 Ti (tensorflow-gpu)
S. Wirges, T. Fischer, C. Stiller and J. Frias: Object Detection and Classification in Occupancy Grid Maps Using Deep Convolutional Networks. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
60 Shift R-CNN (mono) code 5.66 % 8.58 % 4.49 % 0.25 s GPU @ 1.5 Ghz (Python)
A. Naiden, V. Paunescu, G. Kim, B. Jeon and M. Leordeanu: Shift R-CNN: Deep Monocular 3D Object Detection With Closed-form Geometric Constraints. ICIP 2019.
61 SS3D_HW 5.47 % 8.81 % 4.79 % 0.4 s GPU @ 2.5 Ghz (Python)
62 TopNet-UncEst
This method makes use of Velodyne laser scans.
 4.60 % 6.88 % 3.79 % 0.09 s NVIDIA GeForce 1080 Ti (tensorflow-gpu)
S. Wirges, M. Braun, M. Lauer and C. Stiller: Capturing Object Detection Uncertainty in Multi-Layer Grid Maps. 2019.
63 MonoPSR code 4.56 % 7.24 % 4.11 % 0.2 s GPU @ 3.5 Ghz (Python)
J. Ku*, A. Pon* and S. Waslander: Monocular 3D Object Detection Leveraging Accurate Proposals and Shape Reconstruction. CVPR 2019.
64 RTM3D code 4.22 % 6.39 % 3.50 % 0.05 s GPU @ 1.0 Ghz (Python)
P. Li, H. Zhao, P. Liu and F. Cao: RTM3D: Real-time Monocular 3D Detection from Object Keypoints for Autonomous Driving. 2020.
65 M3D-RPN code 4.05 % 5.65 % 3.29 % 0.16 s GPU @ 1.5 Ghz (Python)
G. Brazil and X. Liu: M3D-RPN: Monocular 3D Region Proposal Network for Object Detection . ICCV 2019 .
66 Mono3CN 4.02 % 6.03 % 3.40 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
67 D4LCN code 3.86 % 5.06 % 3.59 % 0.2 s GPU @ 2.5 Ghz (Python + C/C++)
M. Ding, Y. Huo, H. Yi, Z. Wang, J. Shi, Z. Lu and P. Luo: Learning Depth-Guided Convolutions for Monocular 3D Object Detection. arXiv preprint arXiv:1912.04799 2019.
68 RT3DStereo
This method uses stereo information.
 3.65 % 4.72 % 3.00 % 0.08 s GPU @ 2.5 Ghz (C/C++)
H. Königshof, N. Salscheider and C. Stiller: Realtime 3D Object Detection for Automated Driving Using Stereo Vision and Semantic Information. Proc. IEEE Intl. Conf. Intelligent Transportation Systems 2019.
69 PG-MonoNet 3.16 % 4.28 % 2.57 % 0.19 s GPU @ 2.5 Ghz (Python)
70 mylsi-faster-rcnn 2.11 % 3.08 % 1.89 % 0.3 s 1 core @ 2.5 Ghz (Python)
71 SS3D 2.09 % 2.48 % 1.61 % 48 ms Tesla V100 (Python)
E. Jörgensen, C. Zach and F. Kahl: Monocular 3D Object Detection and Box Fitting Trained End-to-End Using Intersection-over-Union Loss. CoRR 2019.
72 mymask-rcnn 1.60 % 2.60 % 1.48 % 0.3 s 1 core @ 2.5 Ghz (Python)
73 mBoW
This method makes use of Velodyne laser scans.
 0.00 % 0.00 % 0.00 % 10 s 1 core @ 2.5 Ghz (C/C++)
J. Behley, V. Steinhage and A. Cremers: Laser-based Segment Classification Using a Mixture of Bag-of-Words. Proc. of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2013.
Table as LaTeX | Only published Methods

Cyclist


Method Setting Code Moderate Easy Hard Runtime Environment
1 Noah CV Lab - SSL 74.45 % 85.96 % 64.23 % 0.1 s GPU @ 2.5 Ghz (Python)
2 PointPainting
This method makes use of Velodyne laser scans.
 71.54 % 83.91 % 62.97 % 0.4 s GPU @ 2.5 Ghz (Python + C/C++)
S. Vora, A. Lang, B. Helou and O. Beijbom: PointPainting: Sequential Fusion for 3D Object Detection. arXiv preprint arXiv:1911.10150 2019.
3 HVNet 71.17 % 83.97 % 63.65 % 0.03 s GPU @ 2.0 Ghz (Python)
4 MMLab PV-RCNN
This method makes use of Velodyne laser scans.
 68.89 % 82.49 % 62.41 % 0.08 s 1 core @ 2.5 Ghz (Python + C/C++)
S. Shi, C. Guo, L. Jiang, Z. Wang, J. Shi, X. Wang and H. Li: PV-RCNN: Point-Voxel Feature Set Abstraction for 3D Object Detection. arXiv preprint arXiv:1912.13192 2019.
5 F-ConvNet
This method makes use of Velodyne laser scans.
 code 68.88 % 84.16 % 60.05 % 0.47 s GPU @ 2.5 Ghz (Python + C/C++)
Z. Wang and K. Jia: Frustum ConvNet: Sliding Frustums to Aggregate Local Point-Wise Features for Amodal 3D Object Detection. IROS 2019.
6 MMLab-PartA^2
This method makes use of Velodyne laser scans.
 code 68.73 % 83.43 % 61.85 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
S. Shi, Z. Wang, J. Shi, X. Wang and H. Li: From Points to Parts: 3D Object Detection from Point Cloud with Part-aware and Part-aggregation Network. arXiv preprint arXiv:1907.03670 2019.
7 3DSSD 67.62 % 85.04 % 61.14 % 0.04 s GPU @ 2.5 Ghz (Python + C/C++)
8 PPBA 67.28 % 82.69 % 60.53 % NA s GPU @ 2.5 Ghz (Python)
9 TBU 67.28 % 82.69 % 60.53 % NA s GPU @ 2.5 Ghz (Python)
10 Point-GNN
This method makes use of Velodyne laser scans.
 67.28 % 81.17 % 59.67 % 0.6 s GPU @ 2.5 Ghz (Python)
11 MMLab-PointRCNN
This method makes use of Velodyne laser scans.
 code 67.24 % 82.56 % 60.28 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
S. Shi, X. Wang and H. Li: Pointrcnn: 3d object proposal generation and detection from point cloud. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition 2019.
12 STD 67.23 % 81.36 % 59.35 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
Z. Yang, Y. Sun, S. Liu, X. Shen and J. Jia: STD: Sparse-to-Dense 3D Object Detector for Point Cloud. ICCV 2019.
13 OHS-Direct 67.20 % 79.66 % 61.04 % 0.03 s 1 core @ 2.5 Ghz (Python + C/C++)
Q. Chen, L. Sun, Z. Wang, K. Jia and A. Yuille: Object as Hotspots: An Anchor-Free 3D Object Detection Approach via Firing of Hotspots. 2019.
14 OHS-Dense 66.86 % 82.13 % 60.86 % 0.03 s 1 core @ 2.5 Ghz (Python/C++)
Q. Chen, L. Sun, Z. Wang, K. Jia and A. Yuille: Object as Hotspots: An Anchor-Free 3D Object Detection Approach via Firing of Hotspots. 2019.
15 ARPNET 66.39 % 82.32 % 58.80 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
Y. Ye, C. Zhang and X. Hao: ARPNET: attention region proposal network for 3D object detection. Science China Information Sciences 2019.
16 AB3DMOT
This method makes use of Velodyne laser scans.
This is an online method (no batch processing).
 code 65.85 % 80.00 % 58.69 % 0.0047s 1 core @ 2.5 Ghz (Python)
X. Weng and K. Kitani: A Baseline for 3D Multi-Object Tracking. arXiv:1907.03961 2019.
17 DENFIDet 65.49 % 82.13 % 57.76 % 0.02 s GPU @ 2.5 Ghz (C/C++)
18 PiP 65.12 % 79.51 % 58.25 % 0.05 s 1 core @ 2.5 Ghz (Python)
19 FOFNet
This method makes use of Velodyne laser scans.
 65.06 % 80.44 % 57.55 % 0.04 s GPU @ 2.5 Ghz (Python)
20 VOXEL_FPN_HR 65.02 % 81.07 % 58.44 % 0.12 s 8 cores @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
21 HR-SECOND code 64.21 % 78.79 % 57.82 % 0.11 s 1 core @ 2.5 Ghz (Python + C/C++)
22 Multi-3D
This method makes use of Velodyne laser scans.
 64.09 % 80.81 % 54.67 % 0.15 s 1 core @ 2.5 Ghz (C/C++)
23 TANet code 63.77 % 79.16 % 56.21 % 0.035s GPU @ 2.5 Ghz (Python + C/C++)
Z. Liu, X. Zhao, T. Huang, R. Hu, Y. Zhou and X. Bai: TANet: Robust 3D Object Detection from Point Clouds with Triple Attention. AAAI 2020.
24 LDAM 63.17 % 77.22 % 57.34 % 24 ms GTX 1080 ti GPU
25 PointPillars
This method makes use of Velodyne laser scans.
 code 62.73 % 79.90 % 55.58 % 16 ms 1080ti GPU and Intel i7 CPU
A. Lang, S. Vora, H. Caesar, L. Zhou, J. Yang and O. Beijbom: PointPillars: Fast Encoders for Object Detection from Point Clouds. CVPR 2019.
26 Tencent_ADlab_Lidar
This method makes use of Velodyne laser scans.
 62.34 % 78.91 % 55.37 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
27 F-PointNet
This method makes use of Velodyne laser scans.
 code 61.37 % 77.26 % 53.78 % 0.17 s GPU @ 3.0 Ghz (Python)
C. Qi, W. Liu, C. Wu, H. Su and L. Guibas: Frustum PointNets for 3D Object Detection from RGB-D Data. arXiv preprint arXiv:1711.08488 2017.
28 A-VoxelNet 60.71 % 76.90 % 53.62 % 0.029 s GPU @ 2.5 Ghz (Python)
29 MP 60.16 % 77.57 % 54.01 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
30 epBRM
This method makes use of Velodyne laser scans.
 code 59.79 % 75.13 % 53.36 % 0.10 s 1 core @ 2.5 Ghz (C/C++)
K. Shin: Improving a Quality of 3D Object Detection by Spatial Transformation Mechanism. arXiv preprint arXiv:1910.04853 2019.
31 Roadstar.ai 59.50 % 68.73 % 53.60 % 0.08 s GPU @ 2.0 Ghz (Python)
32 FCY
This method makes use of Velodyne laser scans.
 59.35 % 76.73 % 52.63 % 0.02 s GPU @ 2.5 Ghz (Python)
33 CentrNet-v1
This method makes use of Velodyne laser scans.
 58.05 % 75.80 % 51.17 % 0.03 s GPU @ 2.5 Ghz (Python)
34 SAANet 57.98 % 74.71 % 51.95 % 0.10 s 1 core @ 2.5 Ghz (Python)
35 SCANet 57.20 % 72.86 % 51.16 % 0.17 s >8 cores @ 2.5 Ghz (Python)
36 AVOD-FPN
This method makes use of Velodyne laser scans.
 code 57.12 % 69.39 % 51.09 % 0.1 s Titan X (Pascal)
J. Ku, M. Mozifian, J. Lee, A. Harakeh and S. Waslander: Joint 3D Proposal Generation and Object Detection from View Aggregation. IROS 2018.
37 DDB
This method makes use of Velodyne laser scans.
 57.01 % 73.70 % 50.71 % 0.05 s GPU @ 2.5 Ghz (Python)
38 deprecated 56.42 % 81.02 % 49.28 % 0.05 s GPU @ 2.0 Ghz (Python)
39 SCNet
This method makes use of Velodyne laser scans.
 56.39 % 73.73 % 49.99 % 0.04 s GPU @ 3.0 Ghz (Python)
Z. Wang, H. Fu, L. Wang, L. Xiao and B. Dai: SCNet: Subdivision Coding Network for Object Detection Based on 3D Point Cloud. IEEE Access 2019.
40 PP-3D 55.06 % 71.94 % 48.10 % 0.1 s 1 core @ 2.5 Ghz (Python)
41 MLOD
This method makes use of Velodyne laser scans.
 code 55.06 % 73.03 % 48.21 % 0.12 s GPU @ 1.5 Ghz (Python)
J. Deng and K. Czarnecki: MLOD: A multi-view 3D object detection based on robust feature fusion method. arXiv preprint arXiv:1909.04163 2019.
42 AVOD
This method makes use of Velodyne laser scans.
 code 48.15 % 64.11 % 42.37 % 0.08 s Titan X (pascal)
J. Ku, M. Mozifian, J. Lee, A. Harakeh and S. Waslander: Joint 3D Proposal Generation and Object Detection from View Aggregation. IROS 2018.
43 BirdNet
This method makes use of Velodyne laser scans.
 41.56 % 58.64 % 36.94 % 0.11 s Titan Xp GPU
J. Beltrán, C. Guindel, F. Moreno, D. Cruzado, F. García and A. Escalera: BirdNet: A 3D Object Detection Framework from LiDAR Information. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
44 SparsePool code 40.74 % 56.52 % 36.68 % 0.13 s 8 cores @ 2.5 Ghz (Python)
Z. Wang, W. Zhan and M. Tomizuka: Fusing bird view lidar point cloud and front view camera image for deep object detection. arXiv preprint arXiv:1711.06703 2017.
45 anm 38.56 % 56.94 % 34.06 % 3 s 1 core @ 2.5 Ghz (C/C++)
46 TopNet-Retina
This method makes use of Velodyne laser scans.
 36.83 % 47.48 % 33.58 % 52ms GeForce 1080Ti (tensorflow-gpu, v1.12)
S. Wirges, T. Fischer, C. Stiller and J. Frias: Object Detection and Classification in Occupancy Grid Maps Using Deep Convolutional Networks. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
47 SparsePool code 35.24 % 43.55 % 30.15 % 0.13 s 8 cores @ 2.5 Ghz (Python)
Z. Wang, W. Zhan and M. Tomizuka: Fusing bird view lidar point cloud and front view camera image for deep object detection. arXiv preprint arXiv:1711.06703 2017.
48 Complexer-YOLO
This method makes use of Velodyne laser scans.
 25.43 % 32.00 % 22.88 % 0.06 s GPU @ 3.5 Ghz (C/C++)
M. Simon, K. Amende, A. Kraus, J. Honer, T. Samann, H. Kaulbersch, S. Milz and H. Michael Gross: Complexer-YOLO: Real-Time 3D Object Detection and Tracking on Semantic Point Clouds. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops 2019.
49 DSGN
This method uses stereo information.
 code 21.04 % 31.23 % 18.93 % 0.67 s NVIDIA Tesla V100
Y. Chen, S. Liu, X. Shen and J. Jia: DSGN: Deep Stereo Geometry Network for 3D Object Detection. arXiv preprint arXiv:2001.03398 2020.
50 OC Stereo
This method uses stereo information.
 19.23 % 32.47 % 17.11 % 0.35 s 1 core @ 2.5 Ghz (Python + C/C++)
A. Pon, J. Ku, C. Li and S. Waslander: Object-Centric Stereo Matching for 3D Object Detection. arXiv preprint arXiv:1909.07566 2019.
51 TopNet-DecayRate
This method makes use of Velodyne laser scans.
 16.00 % 23.02 % 13.24 % 92 ms NVIDIA GeForce 1080 Ti (tensorflow-gpu)
S. Wirges, T. Fischer, C. Stiller and J. Frias: Object Detection and Classification in Occupancy Grid Maps Using Deep Convolutional Networks. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
52 TopNet-UncEst
This method makes use of Velodyne laser scans.
 9.18 % 12.31 % 8.14 % 0.09 s NVIDIA GeForce 1080 Ti (tensorflow-gpu)
S. Wirges, M. Braun, M. Lauer and C. Stiller: Capturing Object Detection Uncertainty in Multi-Layer Grid Maps. 2019.
53 TopNet-HighRes
This method makes use of Velodyne laser scans.
 6.48 % 9.99 % 6.76 % 101ms NVIDIA GeForce 1080 Ti (tensorflow-gpu)
S. Wirges, T. Fischer, C. Stiller and J. Frias: Object Detection and Classification in Occupancy Grid Maps Using Deep Convolutional Networks. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
54 MonoPSR code 5.78 % 9.87 % 4.57 % 0.2 s GPU @ 3.5 Ghz (Python)
J. Ku*, A. Pon* and S. Waslander: Monocular 3D Object Detection Leveraging Accurate Proposals and Shape Reconstruction. CVPR 2019.
55 RT3DStereo
This method uses stereo information.
 4.10 % 7.03 % 3.88 % 0.08 s GPU @ 2.5 Ghz (C/C++)
H. Königshof, N. Salscheider and C. Stiller: Realtime 3D Object Detection for Automated Driving Using Stereo Vision and Semantic Information. Proc. IEEE Intl. Conf. Intelligent Transportation Systems 2019.
56 MonoPair 2.87 % 4.76 % 2.42 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
57 SS3D_HW 2.78 % 5.03 % 2.36 % 0.4 s GPU @ 2.5 Ghz (Python)
58 Mono3CN 2.69 % 3.92 % 2.19 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
59 RTM3D code 2.55 % 4.35 % 2.37 % 0.05 s GPU @ 1.0 Ghz (Python)
P. Li, H. Zhao, P. Liu and F. Cao: RTM3D: Real-time Monocular 3D Detection from Object Keypoints for Autonomous Driving. 2020.
60 RefinedMPL 2.42 % 4.23 % 2.14 % 0.15 s GPU @ 2.5 Ghz (Python + C/C++)
J. Vianney, S. Aich and B. Liu: RefinedMPL: Refined Monocular PseudoLiDAR for 3D Object Detection in Autonomous Driving. arXiv preprint arXiv:1911.09712 2019.
61 SS3D 1.89 % 3.45 % 1.44 % 48 ms Tesla V100 (Python)
E. Jörgensen, C. Zach and F. Kahl: Monocular 3D Object Detection and Box Fitting Trained End-to-End Using Intersection-over-Union Loss. CoRR 2019.
62 D4LCN code 1.82 % 2.72 % 1.79 % 0.2 s GPU @ 2.5 Ghz (Python + C/C++)
M. Ding, Y. Huo, H. Yi, Z. Wang, J. Shi, Z. Lu and P. Luo: Learning Depth-Guided Convolutions for Monocular 3D Object Detection. arXiv preprint arXiv:1912.04799 2019.
63 mylsi-faster-rcnn 1.54 % 2.41 % 1.41 % 0.3 s 1 core @ 2.5 Ghz (Python)
64 PG-MonoNet 1.24 % 1.96 % 1.01 % 0.19 s GPU @ 2.5 Ghz (Python)
65 M3D-RPN code 0.81 % 1.25 % 0.78 % 0.16 s GPU @ 1.5 Ghz (Python)
G. Brazil and X. Liu: M3D-RPN: Monocular 3D Region Proposal Network for Object Detection . ICCV 2019 .
66 mymask-rcnn 0.71 % 1.39 % 0.69 % 0.3 s 1 core @ 2.5 Ghz (Python)
67 Shift R-CNN (mono) code 0.38 % 0.76 % 0.41 % 0.25 s GPU @ 1.5 Ghz (Python)
A. Naiden, V. Paunescu, G. Kim, B. Jeon and M. Leordeanu: Shift R-CNN: Deep Monocular 3D Object Detection With Closed-form Geometric Constraints. ICIP 2019.
68 mBoW
This method makes use of Velodyne laser scans.
 0.00 % 0.00 % 0.00 % 10 s 1 core @ 2.5 Ghz (C/C++)
J. Behley, V. Steinhage and A. Cremers: Laser-based Segment Classification Using a Mixture of Bag-of-Words. Proc. of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2013.
Table as LaTeX | Only published Methods

Related Datasets

Citation

When using this dataset in your research, we will be happy if you cite us:
@INPROCEEDINGS{Geiger2012CVPR,
  author = {Andreas Geiger and Philip Lenz and Raquel Urtasun},
  title = {Are we ready for Autonomous Driving? The KITTI Vision Benchmark Suite},
  booktitle = {Conference on Computer Vision and Pattern Recognition (CVPR)},
  year = {2012}
}



eXTReMe Tracker