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

3D Object Detection Evaluation 2017


The 3D object detection 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 3D object 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 an 3D bounding box overlap of 70%, while for pedestrians and cyclists we require a 3D bounding box 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 MMLab PV-RCNN
This method makes use of Velodyne laser scans.
 81.43 % 90.25 % 76.82 % 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.
2 SA-SSD 79.79 % 88.75 % 74.16 % 0.04 s 1 core @ 2.5 Ghz (Python)
3 STD 79.71 % 87.95 % 75.09 % 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 3DSSD 79.57 % 88.36 % 74.55 % 0.04 s GPU @ 2.5 Ghz (Python + C/C++)
5 Point-GNN
This method makes use of Velodyne laser scans.
 79.47 % 88.33 % 72.29 % 0.6 s GPU @ 2.5 Ghz (Python)
6 EPNet 79.28 % 89.81 % 74.59 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
7 Noah CV Lab - SSL 78.99 % 85.50 % 71.75 % 0.1 s GPU @ 2.5 Ghz (Python)
8 CPRCCNN 78.96 % 87.74 % 74.30 % 0.1 s 1 core @ 2.5 Ghz (Python)
9 ORP 78.50 % 87.38 % 71.49 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
10 MMLab-PartA^2
This method makes use of Velodyne laser scans.
 78.49 % 87.81 % 73.51 % 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 Patches - EMP
This method makes use of Velodyne laser scans.
 78.41 % 89.84 % 73.15 % 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.
12 ELE 78.35 % 86.95 % 73.33 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
13 OHS-Dense 78.34 % 88.12 % 73.49 % 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.
14 HRI-FusionRCNN 78.29 % 88.46 % 70.75 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
15 CP
This method makes use of Velodyne laser scans.
 78.11 % 86.40 % 71.63 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
16 MLF_SecCas 77.97 % 86.53 % 67.90 % 0.05 s 1 core @ 2.5 Ghz (Python)
17 OHS-Direct 77.74 % 86.40 % 72.97 % 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.
18 deprecated 77.62 % 86.21 % 67.68 % 0.05 s GPU @ 2.0 Ghz (Python)
19 UberATG-MMF
This method makes use of Velodyne laser scans.
 77.43 % 88.40 % 70.22 % 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 Associate-3Ddet
This method makes use of Velodyne laser scans.
 77.40 % 85.99 % 70.53 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
21 Fast Point R-CNN
This method makes use of Velodyne laser scans.
 77.40 % 85.29 % 70.24 % 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.
22 3D-CVF
This method makes use of Velodyne laser scans.
 77.31 % 86.44 % 70.91 % 0.05 s GPU @ >3.5 Ghz (Python)
23 Patches
This method makes use of Velodyne laser scans.
 77.20 % 88.67 % 71.82 % 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.
24 deprecated 77.17 % 86.27 % 70.83 % 0.05 s GPU @ >3.5 Ghz (Python)
25 DEFT 77.15 % 86.34 % 70.76 % 1 s GPU @ 2.5 Ghz (Python)
26 HRI-VoxelFPN 76.70 % 85.64 % 69.44 % 0.02 s GPU @ 2.5 Ghz (Python + C/C++)
B. Wang, J. An and J. Cao: Voxel-FPN: multi-scale voxel feature aggregation in 3D object detection from point clouds. arXiv preprint arXiv:1907.05286v2 2019.
27 MLF_PointCas 76.68 % 87.50 % 71.21 % 0.1 s GPU @ 2.5 Ghz (Python)
28 SARPNET 76.64 % 85.63 % 71.31 % 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.
29 SRF 76.61 % 86.63 % 71.28 % 0.05 s GPU @ 2.5 Ghz (Python + C/C++)
30 3D IoU Loss
This method makes use of Velodyne laser scans.
 76.50 % 86.16 % 71.39 % 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.
31 F-ConvNet
This method makes use of Velodyne laser scans.
 code 76.39 % 87.36 % 66.69 % 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.
32 Alibaba-AILabsX
This method makes use of Velodyne laser scans.
 76.29 % 84.71 % 69.18 % 0.05 s 1 core @ >3.5 Ghz (C/C++)
33 RGB3D
This method makes use of Velodyne laser scans.
 76.26 % 87.26 % 71.16 % 0.39 s GPU @ 2.5 Ghz (Python)
34 PiP 76.24 % 85.30 % 70.45 % 0.05 s 1 core @ 2.5 Ghz (Python)
35 SegVoxelNet 76.13 % 86.04 % 70.76 % 0.04 s 1 core @ 2.5 Ghz (Python)
36 PTS
This method makes use of Velodyne laser scans.
 code 76.04 % 84.51 % 70.77 % 0.01 s 1 core @ 2.5 Ghz (C/C++)
37 SECOND-V1.5
This method makes use of Velodyne laser scans.
 code 75.96 % 84.65 % 68.71 % 0.04 s GPU @ 2.0 Ghz (Python + C/C++)
38 TANet code 75.94 % 84.39 % 68.82 % 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.
39 MMV 75.91 % 84.46 % 68.69 % 0.4 s GPU @ 2.5 Ghz (C/C++)
40 PointCSE 75.82 % 86.46 % 70.47 % 0.02 s 1 core @ 2.5 Ghz (C/C++)
41 PointRCNN-deprecated
This method makes use of Velodyne laser scans.
 75.81 % 86.23 % 68.99 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
42 CentrNet-v1
This method makes use of Velodyne laser scans.
 75.76 % 85.40 % 70.29 % 0.03 s GPU @ 2.5 Ghz (Python)
43 PointRGCN 75.73 % 85.97 % 70.60 % 0.26 s GPU @ V100 (Python)
44 IE-PointRCNN 75.67 % 86.26 % 70.47 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
45 MMLab-PointRCNN
This method makes use of Velodyne laser scans.
 code 75.64 % 86.96 % 70.70 % 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.
46 AB3DMOT
This method makes use of Velodyne laser scans.
This is an online method (no batch processing).
 code 75.43 % 86.10 % 68.88 % 0.0047s 1 core @ 2.5 Ghz (Python)
X. Weng and K. Kitani: A Baseline for 3D Multi-Object Tracking. arXiv:1907.03961 2019.
47 PPFNet code 75.43 % 85.91 % 68.88 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
48 HR-SECOND code 75.32 % 84.78 % 68.70 % 0.11 s 1 core @ 2.5 Ghz (Python + C/C++)
49 NU-optim 75.30 % 85.72 % 69.80 % 0.04 s GPU @ >3.5 Ghz (Python)
50 R-GCN 75.26 % 83.42 % 68.73 % 0.16 s GPU @ 2.5 Ghz (Python)
51 SPA 75.25 % 85.35 % 70.46 % 0.1 s 1 core @ 2.5 Ghz (Python)
52 epBRM
This method makes use of Velodyne laser scans.
 code 75.15 % 85.00 % 69.84 % 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.
53 MRF 75.11 % 84.81 % 69.99 % 0.05 s GPU @ 1.5 Ghz (Python + C/C++)
54 PI-RCNN 74.82 % 84.37 % 70.03 % 0.1 s 1 core @ 2.5 Ghz (Python)
55 TBA 74.37 % 83.36 % 69.57 % 0.07 s 1 core @ 2.5 Ghz (Python)
56 MPNet
This method makes use of Velodyne laser scans.
 74.34 % 85.42 % 68.59 % 0.02 s GPU @ 2.5 Ghz (Python + C/C++)
57 PointPillars
This method makes use of Velodyne laser scans.
 code 74.31 % 82.58 % 68.99 % 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.
58 ARPNET 74.04 % 84.69 % 68.64 % 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.
59 PCSC-Net 74.03 % 83.18 % 68.39 % 0.04 s 1 core @ 2.5 Ghz (C/C++)
60 MVSLN 74.00 % 85.19 % 66.81 % 0.1s s 1 core @ 2.5 Ghz (C/C++)
61 PointPiallars_SECA 73.99 % 82.62 % 69.98 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
62 VOXEL_FPN_HR 73.98 % 85.33 % 68.67 % 0.12 s 8 cores @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
63 A-VoxelNet 73.82 % 84.01 % 66.46 % 0.029 s GPU @ 2.5 Ghz (Python)
64 PC-CNN-V2
This method makes use of Velodyne laser scans.
 73.79 % 85.57 % 65.65 % 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.
65 FOFNet
This method makes use of Velodyne laser scans.
 73.70 % 84.56 % 68.09 % 0.04 s GPU @ 2.5 Ghz (Python)
66 3DBN
This method makes use of Velodyne laser scans.
 73.53 % 83.77 % 66.23 % 0.13s 1080Ti (Python+C/C++)
X. Li, J. Guivant, N. Kwok and Y. Xu: 3D Backbone Network for 3D Object Detection. CoRR 2019.
67 DDB
This method makes use of Velodyne laser scans.
 73.49 % 82.45 % 67.82 % 0.05 s GPU @ 2.5 Ghz (Python)
68 BVVF 73.34 % 80.19 % 67.34 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
69 MP 73.32 % 84.00 % 67.82 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
70 SCNet
This method makes use of Velodyne laser scans.
 73.17 % 83.34 % 67.93 % 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.
71 SAANet 73.14 % 84.30 % 66.28 % 0.10 s 1 core @ 2.5 Ghz (Python)
72 SFB-SECOND 73.07 % 83.66 % 68.12 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
73 Tencent_ADlab_Lidar
This method makes use of Velodyne laser scans.
 72.88 % 82.16 % 66.36 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
74 RUC 72.65 % 80.76 % 68.74 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
75 PP-3D 72.20 % 80.35 % 63.54 % 0.1 s 1 core @ 2.5 Ghz (Python)
76 AVOD-FPN
This method makes use of Velodyne laser scans.
 code 71.76 % 83.07 % 65.73 % 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.
77 PointPainting
This method makes use of Velodyne laser scans.
 71.70 % 82.11 % 67.08 % 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.
78 RUC code 71.40 % 80.98 % 65.98 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
79 RUC code 71.32 % 81.07 % 64.69 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
80 FCY
This method makes use of Velodyne laser scans.
 71.11 % 80.63 % 66.11 % 0.02 s GPU @ 2.5 Ghz (Python)
81 PAD 70.33 % 78.94 % 64.83 % 0.15 s 1 core @ 2.5 Ghz (Python)
82 F-PointNet
This method makes use of Velodyne laser scans.
 code 69.79 % 82.19 % 60.59 % 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.
83 RuiRUC 69.32 % 81.45 % 57.64 % 0.12 s 1 core @ 2.5 Ghz (Python)
84 UberATG-ContFuse
This method makes use of Velodyne laser scans.
 68.78 % 83.68 % 61.67 % 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.
85 SCANet 68.12 % 78.65 % 61.44 % 0.17 s >8 cores @ 2.5 Ghz (Python)
86 RADNet-Fusion
This method makes use of Velodyne laser scans.
 68.05 % 79.67 % 63.32 % 0.1 s 1 core @ 2.5 Ghz (Python)
87 RTL3D 67.79 % 80.72 % 61.34 % 0.02 s GPU @ 2.5 Ghz (Python)
88 MLOD
This method makes use of Velodyne laser scans.
 code 67.76 % 77.24 % 62.05 % 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.
89 RADNet-LIDAR
This method makes use of Velodyne laser scans.
 67.29 % 79.71 % 61.03 % 0.1 s 1 core @ 2.5 Ghz (Python)
90 SECA 66.51 % 79.04 % 60.18 % 1 s GPU @ 2.5 Ghz (Python)
91 AVOD
This method makes use of Velodyne laser scans.
 code 66.47 % 76.39 % 60.23 % 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 Multi-3D
This method makes use of Velodyne laser scans.
 66.35 % 78.45 % 55.93 % 0.15 s 1 core @ 2.5 Ghz (C/C++)
93 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).
 66.23 % 81.91 % 60.67 % 0.013 s 1 core @ 2.5 Ghz (Python + C/C++)
94 FailNet-Fusion
This method makes use of Velodyne laser scans.
 65.07 % 79.50 % 58.86 % 0.1 s 1 core @ 2.5 Ghz (Python)
95 3DNN 64.74 % 76.32 % 58.10 % 0.09 s GPU @ 2.5 Ghz (Python)
96 NLK 64.49 % 76.78 % 59.37 % 0.02 s 1 core @ 2.5 Ghz (Python)
97 VoxelNet(Unofficial) 64.17 % 77.82 % 57.51 % 0.5 s GPU @ 2.0 Ghz (Python)
98 MV3D
This method makes use of Velodyne laser scans.
 63.63 % 74.97 % 54.00 % 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.
99 FailNet-LIDAR
This method makes use of Velodyne laser scans.
 62.07 % 76.07 % 55.89 % 0.1 s 1 core @ 2.5 Ghz (Python)
100 ANM 59.07 % 74.99 % 47.64 % 0.12 s 1 core @ 2.5 Ghz (Python)
101 A3DODWTDA
This method makes use of Velodyne laser scans.
 code 56.82 % 62.84 % 48.12 % 0.08 s GPU @ 3.0 Ghz (Python)
F. Gustafsson and E. Linder-Norén: Automotive 3D Object Detection Without Target Domain Annotations. 2018.
102 anm 56.17 % 70.34 % 48.11 % 3 s 1 core @ 2.5 Ghz (C/C++)
103 PL++ (SDN+GDC)
This method uses stereo information.
This method makes use of Velodyne laser scans.
 54.88 % 68.38 % 49.16 % 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.
104 MV3D (LIDAR)
This method makes use of Velodyne laser scans.
 54.54 % 68.35 % 49.16 % 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.
105 avodC 54.03 % 67.80 % 47.95 % 0.1 s GPU @ 2.5 Ghz (Python)
106 E-VoxelNet 52.39 % 66.35 % 46.74 % 0.1 s GPU @ 2.5 Ghz (Python)
107 DSGN
This method uses stereo information.
 code 52.18 % 73.50 % 45.14 % 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.
108 Complexer-YOLO
This method makes use of Velodyne laser scans.
 47.34 % 55.93 % 42.60 % 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.
109 Pseudo-LiDAR E2E
This method uses stereo information.
 43.92 % 64.75 % 38.14 % 0.4 s GPU @ 2.5 Ghz (Python)
110 Pseudo-LiDAR++
This method uses stereo information.
 code 42.43 % 61.11 % 36.99 % 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.
111 Disp R-CNN (velo)
This method uses stereo information.
 39.34 % 59.58 % 31.99 % 0.42 s GPU @ 2.5 Ghz (Python + C/C++)
112 ZoomNet
This method uses stereo information.
 code 38.64 % 55.98 % 30.97 % 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.
113 stereo_sa
This method uses stereo information.
 37.92 % 58.70 % 31.99 % 0.3 s GPU @ 2.5 Ghz (Python)
114 Disp R-CNN
This method uses stereo information.
 37.91 % 58.53 % 31.93 % 0.42 s GPU @ 2.5 Ghz (Python + C/C++)
115 OC Stereo
This method uses stereo information.
 37.60 % 55.15 % 30.25 % 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.
116 m-prcnn
This method uses stereo information.
 31.21 % 53.96 % 24.52 % 0.43 s 1 core @ 2.5 Ghz (Python)
117 Stereo R-CNN
This method uses stereo information.
 code 30.23 % 47.58 % 23.72 % 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.
118 RT3DStereo
This method uses stereo information.
 23.28 % 29.90 % 18.96 % 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.
119 ASOD 22.37 % 38.42 % 17.01 % 0.28 s GPU @ 2.5 Ghz (Python)
120 RT3D
This method makes use of Velodyne laser scans.
 19.14 % 23.74 % 18.86 % 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.
121 StereoFENet
This method uses stereo information.
 18.41 % 29.14 % 14.20 % 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.
122 SAIC-SA-3D
This method makes use of Velodyne laser scans.
 17.79 % 24.78 % 16.56 % 0.05 s GPU @ 2.5 Ghz (Python)
123 Licar
This method makes use of Velodyne laser scans.
 12.10 % 15.23 % 11.39 % 0.09 s GPU @ 2.0 Ghz (Python)
124 D4LCN code 11.72 % 16.65 % 9.51 % 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.
125 RefinedMPL 11.14 % 18.09 % 8.94 % 0.1 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.
126 AM3D 10.74 % 16.50 % 9.52 % 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.
127 RTM3D code 10.09 % 13.61 % 8.18 % 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.
128 MonoPair 9.99 % 13.04 % 8.65 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
129 SMOKE 9.76 % 14.03 % 7.84 % 0.03 s GPU @ 2.5 Ghz (Python)
130 M3D-RPN code 9.71 % 14.76 % 7.42 % 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 .
131 SS3D_HW 9.70 % 14.74 % 7.22 % 0.4 s GPU @ 2.5 Ghz (Python)
132 MonoSS 9.61 % 13.74 % 7.75 % 0.03 s GPU @ 2.5 Ghz (Python + C/C++)
133 BirdNet
This method makes use of Velodyne laser scans.
 9.47 % 13.53 % 8.49 % 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-HighRes
This method makes use of Velodyne laser scans.
 9.28 % 12.67 % 7.95 % 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.
135 RAR-Net 8.73 % 13.70 % 6.92 % 0.5 s 1 core @ 2.5 Ghz (C/C++)
136 PG-MonoNet 8.52 % 13.24 % 6.73 % 0.19 s GPU @ 2.5 Ghz (Python)
137 MonoDIS 7.94 % 10.37 % 6.40 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
138 SS3D 7.68 % 10.78 % 6.51 % 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.
139 Mono3D_PLiDAR code 7.50 % 10.76 % 6.10 % 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.
140 Decoupled-3D v2 7.28 % 11.68 % 5.69 % 0.08 s GPU @ 2.5 Ghz (C/C++)
141 MonoPSR code 7.25 % 10.76 % 5.85 % 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.
142 Decoupled-3D 7.02 % 11.08 % 5.63 % 0.08 s GPU @ 2.5 Ghz (C/C++)
143 VoxelJones code 6.35 % 7.39 % 5.80 % .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.
144 RADNet-Mono 5.80 % 8.42 % 4.62 % 0.1 s 1 core @ 2.5 Ghz (Python)
145 MonoGRNet code 5.74 % 9.61 % 4.25 % 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.
146 A3DODWTDA (image) code 5.27 % 6.88 % 4.45 % 0.8 s GPU @ 3.0 Ghz (Python)
F. Gustafsson and E. Linder-Norén: Automotive 3D Object Detection Without Target Domain Annotations. 2018.
147 MonoFENet 5.14 % 8.35 % 4.10 % 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.
148 mylsi-faster-rcnn 5.07 % 8.36 % 4.21 % 0.3 s 1 core @ 2.5 Ghz (Python)
149 OACV 4.77 % 8.13 % 3.78 % 0.23 s GPU @ 2.5 Ghz (Python)
150 mymask-rcnn 4.21 % 8.54 % 3.44 % 0.3 s 1 core @ 2.5 Ghz (Python)
151 FailNet-Mono 4.19 % 6.84 % 3.28 % 0.1 s 1 core @ 2.5 Ghz (Python)
152 CSoR
This method makes use of Velodyne laser scans.
 4.06 % 5.61 % 3.17 % 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.
153 Shift R-CNN (mono) code 3.87 % 6.88 % 2.83 % 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.
154 MVRA + I-FRCNN+ 3.27 % 5.19 % 2.49 % 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.
155 TopNet-UncEst
This method makes use of Velodyne laser scans.
 3.02 % 3.24 % 2.26 % 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.
156 GS3D 2.90 % 4.47 % 2.47 % 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.
157 3D GCK 2.52 % 3.27 % 2.11 % 24 ms Tesla V100
158 ROI-10D 2.02 % 4.32 % 1.46 % 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.
159 FQNet 1.51 % 2.77 % 1.01 % 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.
160 3D-SSMFCNN code 1.41 % 1.88 % 1.11 % 0.1 s GPU @ 1.5 Ghz (C/C++)
L. Novak: Vehicle Detection and Pose Estimation for Autonomous Driving. 2017.
161 3DVSSD 0.73 % 0.87 % 0.63 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
162 monoref3d 0.04 % 0.08 % 0.04 % 0.1 s 1 core @ 2.5 Ghz (Python)
163 ref3D 0.00 % 0.00 % 0.00 % 0.1 s 1 core @ 2.5 Ghz (Python)
164 ref3D 0.00 % 0.00 % 0.00 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
165 ANM 0.00 % 0.00 % 0.00 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
166 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 Noah CV Lab - SSL 45.23 % 52.85 % 41.28 % 0.1 s GPU @ 2.5 Ghz (Python)
2 OHS-Direct 44.81 % 51.29 % 41.13 % 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.
3 TANet code 44.34 % 53.72 % 40.49 % 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 A-VoxelNet 44.30 % 53.66 % 40.43 % 0.029 s GPU @ 2.5 Ghz (Python)
5 Point-GNN
This method makes use of Velodyne laser scans.
 43.77 % 51.92 % 40.14 % 0.6 s GPU @ 2.5 Ghz (Python)
6 F-ConvNet
This method makes use of Velodyne laser scans.
 code 43.38 % 52.16 % 38.80 % 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.
7 MMLab-PartA^2
This method makes use of Velodyne laser scans.
 43.35 % 53.10 % 40.06 % 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.
8 VMVS
This method makes use of Velodyne laser scans.
 43.27 % 53.44 % 39.51 % 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.
9 STD 42.47 % 53.29 % 38.35 % 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.
10 AVOD-FPN
This method makes use of Velodyne laser scans.
 code 42.27 % 50.46 % 39.04 % 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.
11 F-PointNet
This method makes use of Velodyne laser scans.
 code 42.15 % 50.53 % 38.08 % 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 PointPillars
This method makes use of Velodyne laser scans.
 code 41.92 % 51.45 % 38.89 % 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.
13 epBRM
This method makes use of Velodyne laser scans.
 code 41.52 % 49.17 % 39.08 % 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.
14 CentrNet-v1
This method makes use of Velodyne laser scans.
 41.50 % 50.86 % 38.24 % 0.03 s GPU @ 2.5 Ghz (Python)
15 PiP 41.01 % 49.01 % 37.90 % 0.05 s 1 core @ 2.5 Ghz (Python)
16 PointPainting
This method makes use of Velodyne laser scans.
 40.97 % 50.32 % 37.87 % 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.
17 Multi-3D
This method makes use of Velodyne laser scans.
 40.72 % 49.50 % 36.22 % 0.15 s 1 core @ 2.5 Ghz (C/C++)
18 DDB
This method makes use of Velodyne laser scans.
 40.40 % 49.03 % 37.04 % 0.05 s GPU @ 2.5 Ghz (Python)
19 PPFNet code 40.11 % 48.36 % 37.00 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
20 PP-3D 39.76 % 49.59 % 36.49 % 0.1 s 1 core @ 2.5 Ghz (Python)
21 OHS-Dense 39.72 % 47.14 % 37.25 % 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.
22 LDAM 39.55 % 45.15 % 37.27 % 0.05 s GPU @ 2.5 Ghz (C/C++)
23 Tencent_ADlab_Lidar
This method makes use of Velodyne laser scans.
 39.47 % 47.87 % 36.35 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
24 MMLab-PointRCNN
This method makes use of Velodyne laser scans.
 code 39.37 % 47.98 % 36.01 % 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.
25 ARPNET 39.31 % 48.32 % 35.93 % 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.
26 SCNet
This method makes use of Velodyne laser scans.
 38.66 % 47.83 % 35.70 % 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.
27 SCANet 37.93 % 48.41 % 34.10 % 0.17 s >8 cores @ 2.5 Ghz (Python)
28 FOFNet
This method makes use of Velodyne laser scans.
 37.56 % 47.45 % 34.00 % 0.04 s GPU @ 2.5 Ghz (Python)
29 MLOD
This method makes use of Velodyne laser scans.
 code 37.47 % 47.58 % 35.07 % 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.
30 VOXEL_FPN_HR 37.01 % 46.32 % 34.67 % 0.12 s 8 cores @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
31 deprecated 36.25 % 47.69 % 32.18 % 0.05 s GPU @ 2.0 Ghz (Python)
32 HR-SECOND code 35.52 % 45.31 % 33.14 % 0.11 s 1 core @ 2.5 Ghz (Python + C/C++)
33 AB3DMOT
This method makes use of Velodyne laser scans.
This is an online method (no batch processing).
 code 34.59 % 42.27 % 31.37 % 0.0047s 1 core @ 2.5 Ghz (python)
X. Weng and K. Kitani: A Baseline for 3D Multi-Object Tracking. arXiv:1907.03961 2019.
34 MP 33.89 % 43.04 % 31.46 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
35 FCY
This method makes use of Velodyne laser scans.
 33.74 % 45.82 % 29.58 % 0.02 s GPU @ 2.5 Ghz (Python)
36 anm 32.98 % 43.55 % 29.12 % 3 s 1 core @ 2.5 Ghz (C/C++)
37 SAANet 30.61 % 38.50 % 27.35 % 0.10 s 1 core @ 2.5 Ghz (Python)
38 SparsePool code 30.38 % 37.84 % 26.94 % 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.
39 SparsePool code 27.92 % 35.52 % 25.87 % 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.
40 AVOD
This method makes use of Velodyne laser scans.
 code 27.86 % 36.10 % 25.76 % 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.
41 27.75 % 35.85 % 25.09 %
42 CSW3D
This method makes use of Velodyne laser scans.
 26.64 % 33.75 % 23.34 % 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.
43 OC Stereo
This method uses stereo information.
 17.58 % 24.48 % 15.60 % 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.
44 DSGN
This method uses stereo information.
 code 15.55 % 20.53 % 14.15 % 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.
45 Complexer-YOLO
This method makes use of Velodyne laser scans.
 13.96 % 17.60 % 12.70 % 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.
46 BirdNet
This method makes use of Velodyne laser scans.
 8.99 % 12.25 % 8.06 % 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.
47 RefinedMPL 7.18 % 11.14 % 5.84 % 0.1 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.
48 TopNet-HighRes
This method makes use of Velodyne laser scans.
 6.92 % 10.40 % 6.63 % 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.
49 MonoPair 6.68 % 10.02 % 5.53 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
50 SS3D_HW 5.00 % 7.77 % 4.03 % 0.4 s GPU @ 2.5 Ghz (Python)
51 Shift R-CNN (mono) code 4.66 % 7.95 % 4.16 % 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.
52 MonoPSR code 4.00 % 6.12 % 3.30 % 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.
53 RTM3D code 3.52 % 5.60 % 3.30 % 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.
54 M3D-RPN code 3.48 % 4.92 % 2.94 % 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 .
55 D4LCN code 3.42 % 4.55 % 2.83 % 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.
56 PG-MonoNet 2.58 % 3.61 % 2.36 % 0.19 s GPU @ 2.5 Ghz (Python)
57 RT3DStereo
This method uses stereo information.
 2.45 % 3.28 % 2.35 % 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.
58 TopNet-UncEst
This method makes use of Velodyne laser scans.
 1.87 % 3.42 % 1.73 % 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.
59 SS3D 1.78 % 2.31 % 1.48 % 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.
60 mylsi-faster-rcnn 0.96 % 1.36 % 0.62 % 0.3 s 1 core @ 2.5 Ghz (Python)
61 mymask-rcnn 0.80 % 1.16 % 0.71 % 0.3 s 1 core @ 2.5 Ghz (Python)
62 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 71.53 % 84.24 % 62.20 % 0.1 s GPU @ 2.5 Ghz (Python)
2 F-ConvNet
This method makes use of Velodyne laser scans.
 code 65.07 % 81.98 % 56.54 % 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.
3 PointPainting
This method makes use of Velodyne laser scans.
 63.78 % 77.63 % 55.89 % 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.
4 MMLab PV-RCNN
This method makes use of Velodyne laser scans.
 63.71 % 78.60 % 57.65 % 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 MMLab-PartA^2
This method makes use of Velodyne laser scans.
 63.52 % 79.17 % 56.93 % 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.
6 Point-GNN
This method makes use of Velodyne laser scans.
 63.48 % 78.60 % 57.08 % 0.6 s GPU @ 2.5 Ghz (Python)
7 OHS-Direct 63.16 % 77.70 % 57.16 % 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.
8 OHS-Dense 62.72 % 79.09 % 56.76 % 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.
9 VOXEL_FPN_HR 61.91 % 78.29 % 55.54 % 0.12 s 8 cores @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
10 STD 61.59 % 78.69 % 55.30 % 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.
11 HR-SECOND code 60.82 % 75.83 % 53.67 % 0.11 s 1 core @ 2.5 Ghz (Python + C/C++)
12 AB3DMOT
This method makes use of Velodyne laser scans.
This is an online method (no batch processing).
 code 60.30 % 75.42 % 53.81 % 0.0047s 1 core @ 2.5 Ghz (Python)
X. Weng and K. Kitani: A Baseline for 3D Multi-Object Tracking. arXiv:1907.03961 2019.
13 FOFNet
This method makes use of Velodyne laser scans.
 59.73 % 76.23 % 53.44 % 0.04 s GPU @ 2.5 Ghz (Python)
14 PiP 59.54 % 75.43 % 53.37 % 0.05 s 1 core @ 2.5 Ghz (Python)
15 TANet code 59.44 % 75.70 % 52.53 % 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.
16 Multi-3D
This method makes use of Velodyne laser scans.
 59.04 % 76.77 % 50.45 % 0.15 s 1 core @ 2.5 Ghz (C/C++)
17 MMLab-PointRCNN
This method makes use of Velodyne laser scans.
 code 58.82 % 74.96 % 52.53 % 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.
18 PointPillars
This method makes use of Velodyne laser scans.
 code 58.65 % 77.10 % 51.92 % 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 ARPNET 58.20 % 74.21 % 52.13 % 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.
20 Tencent_ADlab_Lidar
This method makes use of Velodyne laser scans.
 57.15 % 73.69 % 50.17 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
21 LDAM 56.79 % 71.66 % 50.82 % 0.05 s GPU @ 2.5 Ghz (C/C++)
22 epBRM
This method makes use of Velodyne laser scans.
 code 56.13 % 72.08 % 49.91 % 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.
23 F-PointNet
This method makes use of Velodyne laser scans.
 code 56.12 % 72.27 % 49.01 % 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.
24 A-VoxelNet 55.86 % 72.58 % 49.13 % 0.029 s GPU @ 2.5 Ghz (Python)
25 deprecated 55.58 % 77.86 % 48.49 % 0.05 s GPU @ 2.0 Ghz (Python)
26 MP 55.36 % 72.99 % 49.36 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
27 CentrNet-v1
This method makes use of Velodyne laser scans.
 54.64 % 72.03 % 48.03 % 0.03 s GPU @ 2.5 Ghz (Python)
28 SCANet 53.38 % 68.71 % 47.59 % 0.17 s >8 cores @ 2.5 Ghz (Python)
29 DDB
This method makes use of Velodyne laser scans.
 51.38 % 68.83 % 45.15 % 0.05 s GPU @ 2.5 Ghz (Python)
30 SCNet
This method makes use of Velodyne laser scans.
 50.79 % 67.98 % 45.15 % 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.
31 AVOD-FPN
This method makes use of Velodyne laser scans.
 code 50.55 % 63.76 % 44.93 % 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.
32 MLOD
This method makes use of Velodyne laser scans.
 code 49.43 % 68.81 % 42.84 % 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 PP-3D 49.19 % 66.54 % 42.68 % 0.1 s 1 core @ 2.5 Ghz (Python)
34 SAANet 48.67 % 62.76 % 43.45 % 0.10 s 1 core @ 2.5 Ghz (Python)
35 AVOD
This method makes use of Velodyne laser scans.
 code 42.08 % 57.19 % 38.29 % 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.
36 FCY
This method makes use of Velodyne laser scans.
 38.15 % 53.92 % 34.37 % 0.02 s GPU @ 2.5 Ghz (Python)
37 SparsePool code 37.33 % 52.61 % 33.39 % 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.
38 anm 33.28 % 49.27 % 28.90 % 3 s 1 core @ 2.5 Ghz (C/C++)
39 SparsePool code 32.61 % 40.87 % 29.05 % 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.
40 Complexer-YOLO
This method makes use of Velodyne laser scans.
 18.53 % 24.27 % 17.31 % 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.
41 DSGN
This method uses stereo information.
 code 18.17 % 27.76 % 16.21 % 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.
42 OC Stereo
This method uses stereo information.
 16.63 % 29.40 % 14.72 % 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.
43 BirdNet
This method makes use of Velodyne laser scans.
 10.46 % 16.63 % 9.53 % 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 MonoPSR code 4.74 % 8.37 % 3.68 % 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.
45 TopNet-UncEst
This method makes use of Velodyne laser scans.
 4.54 % 7.13 % 3.81 % 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.
46 RT3DStereo
This method uses stereo information.
 3.37 % 5.29 % 2.57 % 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.
47 RTM3D code 2.38 % 3.97 % 2.22 % 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.
48 SS3D_HW 2.17 % 4.29 % 2.00 % 0.4 s GPU @ 2.5 Ghz (Python)
49 MonoPair 2.12 % 3.79 % 1.83 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
50 RefinedMPL 1.82 % 3.23 % 1.77 % 0.1 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.
51 TopNet-HighRes
This method makes use of Velodyne laser scans.
 1.67 % 2.49 % 1.88 % 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.
52 D4LCN code 1.67 % 2.45 % 1.36 % 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.
53 SS3D 1.45 % 2.80 % 1.35 % 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.
54 mylsi-faster-rcnn 1.07 % 1.71 % 0.85 % 0.3 s 1 core @ 2.5 Ghz (Python)
55 PG-MonoNet 0.90 % 1.59 % 0.95 % 0.19 s GPU @ 2.5 Ghz (Python)
56 M3D-RPN code 0.65 % 0.94 % 0.47 % 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 .
57 mymask-rcnn 0.30 % 0.71 % 0.28 % 0.3 s 1 core @ 2.5 Ghz (Python)
58 Shift R-CNN (mono) code 0.29 % 0.48 % 0.31 % 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.
59 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