The website uses cookies. By using this site, you agree to our use of cookies as described in the Privacy Policy.
I Agree

The KITTI Vision Benchmark Suite

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
  • Additional training data: Use of additional data sources for training (see details)

Car

Method Setting Code Moderate Easy Hard Runtime Environment
1 BorderAtt 82.33 % 87.77 % 77.37 % 0.08 s 1 core @ 2.5 Ghz (C/C++)
2 HUAWEI Octopus 82.13 % 88.26 % 77.41 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
3 ADLAB 82.08 % 90.92 % 77.36 % 0.08 s 1 core @ >3.5 Ghz (C/C++)
4 RangeRCNN-LV 81.85 % 88.76 % 77.18 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
5 PVGNet 81.81 % 89.94 % 77.09 % 0.05 s 1 core @ >3.5 Ghz (C/C++)
6 Voxel R-CNN 81.62 % 90.90 % 77.06 % 0.04 s GPU @ 3.0 Ghz (C/C++)
7 IC-PVRCNN 81.57 % 88.60 % 77.09 % 0.08 s 1 core @ 2.5 Ghz (C/C++)
8 Deformable PV-RCNN
This method makes use of Velodyne laser scans.
code 81.46 % 88.25 % 76.96 % 0.08 s 1 core @ 2.5 Ghz (Python + C/C++)
P. Bhattacharyya and K. Czarnecki: Deformable PV-RCNN: Improving 3D Object Detection with Learned Deformations. ECCV 2020 Perception for Autonomous Driving Workshop.
9 MMLab PV-RCNN
This method makes use of Velodyne laser scans.
code 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. CVPR 2020.
10 PC-RGNN 81.38 % 87.94 % 76.88 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
11 RangeRCNN
This method makes use of Velodyne laser scans.
81.33 % 88.47 % 77.09 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
Z. Liang, M. Zhang, Z. Zhang, X. Zhao and S. Pu: RangeRCNN: Towards Fast and Accurate 3D Object Detection with Range Image Representation. arXiv preprint arXiv:2009.00206 2020.
12 OneCoLab SicNet V2 80.94 % 87.55 % 76.36 % 0.08 s 1 core @ 2.5 Ghz (C/C++)
13 Associate-3Ddet_v2 80.77 % 91.53 % 75.23 % 0.04 s 1 core @ 2.5 Ghz (Python)
14 CIA-SSD v2
This method makes use of Velodyne laser scans.
80.71 % 89.61 % 75.06 % 0.03 s 1 core @ 2.5 Ghz (C/C++)
15 CLOCs_PVCas 80.67 % 88.94 % 77.15 % 0.1 s 1 core @ 2.5 Ghz (Python)
S. Pang, D. Morris and H. Radha: CLOCs: Camera-LiDAR Object Candidates Fusion for 3D Object Detection . 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2020.
16 AIMC-RUC 80.63 % 89.90 % 75.32 % 0.08 s 1 core @ 2.5 Ghz (C/C++)
17 OAP 80.63 % 89.18 % 73.04 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
18 HRI-MSP-L
This method makes use of Velodyne laser scans.
80.62 % 87.61 % 76.29 % 0.07 s 1 core @ 2.5 Ghz (C/C++)
19 IC-SECOND 80.61 % 88.25 % 75.83 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
20 CVIS-DF3D_v2 80.48 % 87.20 % 76.01 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
21 SVGA-Net
This method makes use of Velodyne laser scans.
80.38 % 87.73 % 76.27 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
22 SPANet 80.34 % 91.05 % 74.89 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
23 CVRS_PF 80.33 % 88.04 % 75.21 % 0.09 s 1 core @ 2.5 Ghz (C/C++)
24 CIA-SSD
This method makes use of Velodyne laser scans.
80.28 % 89.59 % 72.87 % 0.03 s 1 core @ 2.5 Ghz (Python + C/C++)
25 Baseline of CA RCNN 80.28 % 87.45 % 76.21 % 0.1 s GPU @ 2.5 Ghz (Python)
26 CVIS-DF3D 80.28 % 87.45 % 76.21 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
27 CBi-GNN 80.18 % 91.50 % 74.76 % 0.03 s 1 core @ 2.5 Ghz (C/C++)
28 deprecated 80.16 % 89.48 % 72.75 % deprecated deprecated
29 3D-CVF at SPA
This method makes use of Velodyne laser scans.
80.05 % 89.20 % 73.11 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
J. Yoo, Y. Kim, J. Kim and J. Choi: 3D-CVF: Generating Joint Camera and LiDAR Features Using Cross-View Spatial Feature Fusion for 3D Object Detection. ECCV 2020.
30 CN 79.89 % 90.55 % 76.31 % 0.04 s GPU @ 2.5 Ghz (Python + C/C++)
31 VAL 79.87 % 89.35 % 70.27 % 0.03 s 1 core @ 2.5 Ghz (C/C++)
32 SA-SSD code 79.79 % 88.75 % 74.16 % 0.04 s 1 core @ 2.5 Ghz (Python)
C. He, H. Zeng, J. Huang, X. Hua and L. Zhang: Structure Aware Single-stage 3D Object Detection from Point Cloud. CVPR 2020.
33 CJJ 79.72 % 88.98 % 74.71 % 0.04 s 1 core @ 2.5 Ghz (C/C++)
34 STD code 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.
35 AF_V1 79.68 % 88.16 % 72.39 % 0.1 s 1 core @ 2.5 Ghz (Python)
36 FCY
This method makes use of Velodyne laser scans.
79.67 % 89.19 % 74.35 % 0.02 s GPU @ 2.5 Ghz (Python)
37 scssd-normal(0.3) 79.59 % 88.97 % 72.51 % 0.05 s GPU @ 2.5 Ghz (Python)
P. An, J. Liang, J. Ma, K. Yu and B. Fang: SCSSD for multi sensor fusion 3d object detection method. Submitted to Information Science 2020.
38 3DSSD code 79.57 % 88.36 % 74.55 % 0.04 s GPU @ 2.5 Ghz (Python + C/C++)
Z. Yang, Y. Sun, S. Liu and J. Jia: 3DSSD: Point-based 3D Single Stage Object Detector. CVPR 2020.
39 PointRes
This method makes use of Velodyne laser scans.
79.55 % 88.73 % 74.17 % 0.013 s 1 core @ 2.5 Ghz (Python + C/C++)
40 Cas-SSD 79.50 % 88.73 % 72.46 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
41 scssd-normal(0.4) 79.49 % 88.70 % 74.25 % 0.05 s 1 core @ 2.5 Ghz (Python)
P. An, J. Liang, J. Ma, K. Yu and B. Fang: SCSSD for multi sensor fusion 3d object detection method. Submitted to Information Science 2020.
42 Point-GNN
This method makes use of Velodyne laser scans.
code 79.47 % 88.33 % 72.29 % 0.6 s GPU @ 2.5 Ghz (Python)
W. Shi and R. Rajkumar: Point-GNN: Graph Neural Network for 3D Object Detection in a Point Cloud. CVPR 2020.
43 PP-3D 79.47 % 88.33 % 72.29 % 0.1 s 1 core @ 2.5 Ghz (Python)
44 nonet 79.42 % 88.28 % 75.77 % 0.08 s 1 core @ 2.5 Ghz (C/C++)
45 RoIFusion code 79.41 % 88.43 % 72.58 % 0.22 s 1 core @ 3.0 Ghz (Python)
46 EPNet code 79.28 % 89.81 % 74.59 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
T. Huang, Z. Liu, X. Chen and X. Bai: EPNet: Enhancing Point Features with Image Semantics for 3D Object Detection. ECCV 2020.
47 MGACNet 79.18 % 86.20 % 74.58 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
48 D3D 79.15 % 87.07 % 73.79 % 0.02 s 1 core @ 2.5 Ghz (C/C++)
49 NLK-ALL code 79.13 % 87.23 % 74.30 % 0.04 s 1 core @ 2.5 Ghz (Python + C/C++)
50 3D IoU-Net 79.03 % 87.96 % 72.78 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
J. Li, S. Luo, Z. Zhu, H. Dai, S. Krylov, Y. Ding and L. Shao: 3D IoU-Net: IoU Guided 3D Object Detector for Point Clouds. arXiv preprint arXiv:2004.04962 2020.
51 Noah CV Lab - SSL 78.99 % 85.50 % 71.75 % 0.1 s GPU @ 2.5 Ghz (Python)
52 SERCNN
This method makes use of Velodyne laser scans.
78.96 % 87.74 % 74.30 % 0.1 s 1 core @ 2.5 Ghz (Python)
D. Zhou, J. Fang, X. Song, L. Liu, J. Yin, Y. Dai, H. Li and R. Yang: Joint 3D Instance Segmentation and Object Detection for Autonomous Driving. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition 2020.
53 deprecated 78.83 % 87.89 % 73.52 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
54 FLID 78.78 % 86.73 % 71.24 % 0.04 s GPU @ 2.5 Ghz (Python)
55 OneCoLab SicNet 78.65 % 87.41 % 74.16 % 0.08 s 1 core @ 2.5 Ghz (C/C++)
56 PVF-NET 78.58 % 87.05 % 71.68 % 0.1 s 1 core @ 2.5 Ghz (Python)
57 BLPNet_V2 78.57 % 87.10 % 71.67 % 0.04 s 1 core @ 2.5 Ghz (Python)
58 Discrete-PointDet 78.51 % 88.53 % 71.29 % 0.02 s 1 core @ 2.5 Ghz (C/C++)
59 MMLab-PartA^2
This method makes use of Velodyne laser scans.
code 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. IEEE Transactions on Pattern Analysis and Machine Intelligence 2020.
60 SRDL
This method uses stereo information.
This method makes use of Velodyne laser scans.
78.49 % 86.75 % 74.04 % 0.15 s GPU @ 2.5 Ghz (Python + C/C++)
61 F-3DNet 78.48 % 85.48 % 71.62 % 0.5 s GPU @ 2.5 Ghz (Python)
62 CLOCs_SecCas 78.45 % 86.38 % 72.45 % 0.1 s 1 core @ 2.5 Ghz (Python)
S. Pang, D. Morris and H. Radha: CLOCs: Camera-LiDAR Object Candidates Fusion for 3D Object Detection. 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2020.
63 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.
64 LZY_RCNN 78.41 % 85.38 % 74.04 % 0.08 s 1 core @ 2.5 Ghz (C/C++)
65 deprecated 78.32 % 89.34 % 71.21 % 0.06 s GPU @ >3.5 Ghz (Python)
66 HotSpotNet 78.31 % 87.60 % 73.34 % 0.04 s 1 core @ 2.5 Ghz (Python + C/C++)
Q. Chen, L. Sun, Z. Wang, K. Jia and A. Yuille: object as hotspots. Proceedings of the European Conference on Computer Vision (ECCV) 2020.
67 cvMax 78.28 % 86.60 % 71.60 % 0.04 s GPU @ >3.5 Ghz (Python)
68 KNN-GCNN 78.26 % 86.37 % 71.14 % 0.4 s 1 core @ 2.5 Ghz (Python + C/C++)
69 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++)
70 CenterNet3DV1.5 code 78.08 % 86.22 % 73.21 % 0.04 s 1 core @ 2.5 Ghz (Python + C/C++)
G. Wang, B. Tian, Y. Ai, T. Xu, L. Chen and D. Cao: CenterNet3D:An Anchor free Object Detector for Autonomous Driving. 2020.
71 Chovy 78.02 % 86.86 % 73.20 % 0.04 s GPU @ 2.5 Ghz (Python)
72 deprecated 77.97 % 86.76 % 73.00 % 0.04 s GPU @ 2.5 Ghz (Python)
73 deprecated 77.97 % 86.53 % 67.90 % 0.05 s 1 core @ 2.5 Ghz (Python)
74 CenterNet3D 77.90 % 86.20 % 73.03 % 0.04 s GPU @ 1.5 Ghz (Python)
G. Wang, B. Tian, Y. Ai, T. Xu, L. Chen and D. Cao: CenterNet3D:An Anchor free Object Detector for Autonomous Driving. 2020.
75 tbd code 77.72 % 86.09 % 72.53 % 0.08 s 1 core @ 2.5 Ghz (C/C++)
76 PPBA 77.65 % 84.16 % 71.21 % NA s GPU @ 2.5 Ghz (Python)
77 TBU 77.65 % 84.16 % 71.21 % NA s GPU @ 2.5 Ghz (Python)
78 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.
79 Associate-3Ddet code 77.40 % 85.99 % 70.53 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
L. Du, X. Ye, X. Tan, J. Feng, Z. Xu, E. Ding and S. Wen: Associate-3Ddet: Perceptual-to-Conceptual Association for 3D Point Cloud Object Detection. The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2020.
80 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.
81 deprecated 77.31 % 86.44 % 70.91 % - -
82 Dccnet 77.22 % 86.67 % 69.97 % 0.05 s 1 core @ 2.5 Ghz (C/C++)
83 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.
84 deprecated 77.17 % 86.27 % 70.83 % 0.05 s GPU @ >3.5 Ghz (Python)
85 DEFT 77.15 % 86.34 % 70.76 % 1 s GPU @ 2.5 Ghz (Python)
86 VAR 77.08 % 84.92 % 72.21 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
87 CU-PointRCNN 76.87 % 86.55 % 73.17 % 0.1 s GPU @ 1.5 Ghz (Python + C/C++)
88 HRI-VoxelFPN 76.70 % 85.64 % 69.44 % 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.
89 CLOCs_PointCas 76.68 % 87.50 % 71.21 % 0.1 s GPU @ 2.5 Ghz (Python)
90 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.
91 TBD 76.57 % 85.33 % 72.05 % 0.05 s GPU @ 2.5 Ghz (Python)
92 IGRP+ 76.54 % 86.90 % 71.77 % 0.18 s 1 core @ 2.5 Ghz (Python + C/C++)
93 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.
94 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.
95 PiP 76.24 % 85.30 % 70.45 % 0.033 s 1 core @ 2.5 Ghz (Python)
96 VICNet 76.18 % 85.21 % 70.60 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
97 SegVoxelNet 76.13 % 86.04 % 70.76 % 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.
98 NLK-3D 76.08 % 84.47 % 70.93 % 0.04 s 1 core @ 2.5 Ghz (Python + C/C++)
99 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.
100 IGRP 75.90 % 86.27 % 69.31 % 0.18 s 1 core @ 2.5 Ghz (Python + C/C++)
101 MVX-Net++ 75.86 % 85.99 % 70.70 % 0.15 s 1 core @ 2.5 Ghz (C/C++)
102 PointCSE 75.82 % 86.46 % 70.47 % 0.02 s 1 core @ 2.5 Ghz (C/C++)
103 CentrNet-v1
This method makes use of Velodyne laser scans.
75.76 % 85.40 % 70.29 % 0.03 s GPU @ 2.5 Ghz (Python)
104 PointRGCN 75.73 % 85.97 % 70.60 % 0.26 s GPU @ V100 (Python)
J. Zarzar, S. Giancola and B. Ghanem: PointRGCN: Graph Convolution Networks for 3D Vehicles Detection Refinement. ArXiv 2019.
105 IE-PointRCNN 75.67 % 86.26 % 70.47 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
106 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.
107 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.
108 PPFNet code 75.43 % 85.91 % 68.88 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
109 MDA 75.39 % 83.72 % 71.98 % 0.03 s 1 core @ 2.5 Ghz (C/C++)
110 HR-SECOND code 75.32 % 84.78 % 68.70 % 0.11 s 1 core @ 2.5 Ghz (Python + C/C++)
111 R-GCN 75.26 % 83.42 % 68.73 % 0.16 s GPU @ 2.5 Ghz (Python)
J. Zarzar, S. Giancola and B. Ghanem: PointRGCN: Graph Convolution Networks for 3D Vehicles Detection Refinement. ArXiv 2019.
112 SPA 75.25 % 85.35 % 70.46 % 0.1 s 1 core @ 2.5 Ghz (Python)
113 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.
114 MuRF 75.11 % 84.81 % 69.99 % 0.05 s GPU @ 1.5 Ghz (Python + C/C++)
115 3DBN_2 75.06 % 84.90 % 72.10 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
ERROR: Wrong syntax in BIBTEX file.
116 MAFF-Net(DAF-Pillar) 75.04 % 85.52 % 67.61 % 0.04 s 1 core @ 2.5 Ghz (Python + C/C++)
Z. Zhang, Z. Liang, M. Zhang, X. Zhao, Y. Ming, T. Wenming and S. Pu: MAFF-Net: Filter False Positive for 3D Vehicle Detection with Multi-modal Adaptive Feature Fusion. arXiv preprint arXiv:2009.10945 2020.
117 PBASN code 75.02 % 83.16 % 69.72 % NA s GPU @ 2.5 Ghz (Python)
118 PI-RCNN 74.82 % 84.37 % 70.03 % 0.1 s 1 core @ 2.5 Ghz (Python)
L. Xie, C. Xiang, Z. Yu, G. Xu, Z. Yang, D. Cai and X. He: PI-RCNN: An Efficient Multi-sensor 3D Object Detector with Point-based Attentive Cont-conv Fusion Module. AAAI 2020 : The Thirty-Fourth AAAI Conference on Artificial Intelligence 2020.
119 EPENet 74.72 % 85.19 % 70.05 % 0.04 s 1 core @ 2.5 Ghz (C/C++)
120 Pointpillar_TV 74.55 % 83.08 % 69.13 % 0.05 s 1 core @ 2.5 Ghz (C/C++)
121 CentrNet-FG 74.47 % 83.67 % 69.94 % 0.03 s 1 core @ 2.5 Ghz (C/C++)
122 RethinkDet3D 74.35 % 82.81 % 67.90 % 0.15 s 1 core @ 2.5 Ghz (Python)
123 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.
124 Bit 74.30 % 82.67 % 68.73 % 0.11 s 1 core @ 2.5 Ghz (C/C++)
125 Prune 74.28 % 85.03 % 67.16 % 0.11 s 1 core @ 2.5 Ghz (C/C++)
126 autoRUC 74.08 % 84.54 % 67.04 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
127 Simple3D Net 74.06 % 83.06 % 69.17 % 0.02 s 1 core @ 2.5 Ghz (C/C++)
128 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.
129 PointPiallars_SECA 73.99 % 82.62 % 69.98 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
130 VOXEL_FPN_HR 73.98 % 85.33 % 68.67 % 0.12 s 8 cores @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
131 tt code 73.92 % 84.14 % 69.15 % 0.08 s 1 core @ 2.5 Ghz (C/C++)
132 autonet 73.83 % 82.66 % 67.93 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
133 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.
134 C-GCN 73.62 % 83.49 % 67.01 % 0.147 s GPU @ V100 (Python)
J. Zarzar, S. Giancola and B. Ghanem: PointRGCN: Graph Convolution Networks for 3D Vehicles Detection Refinement. ArXiv 2019.
135 baseline 73.55 % 82.92 % 67.42 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
136 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.
137 DDB
This method makes use of Velodyne laser scans.
73.49 % 82.45 % 67.82 % 0.05 s GPU @ 2.5 Ghz (Python)
138 BVVF 73.34 % 80.19 % 67.34 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
139 MP 73.32 % 84.00 % 67.82 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
140 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.
141 SFB-SECOND 73.07 % 83.66 % 68.12 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
142 DPointNet 73.02 % 79.25 % 68.53 % 0.09 s 1 core @ 2.5 Ghz (C/C++)
143 PFF3D
This method makes use of Velodyne laser scans.
72.93 % 81.11 % 67.24 % 0.05 s GPU @ 3.0 Ghz (Python + C/C++)
144 RUC 72.65 % 80.76 % 68.74 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
145 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.
146 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. CVPR 2020.
147 RUC code 71.40 % 80.98 % 65.98 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
148 RUC code 71.32 % 81.07 % 64.69 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
149 WS3D
This method makes use of Velodyne laser scans.
70.59 % 80.99 % 64.23 % 0.1 s GPU @ 2.5 Ghz (Python)
Q. Meng, W. Wang, T. Zhou, J. Shen, L. Van Gool and D. Dai: Weakly Supervised 3D Object Detection from Lidar Point Cloud. 2020.
150 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.
151 RuiRUC 69.32 % 81.45 % 57.64 % 0.12 s 1 core @ 2.5 Ghz (Python)
152 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.
153 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.
154 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.
155 seivl 66.40 % 77.00 % 63.48 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
156 DAMNET code 65.52 % 76.25 % 59.54 % 1 s 1 core @ 2.5 Ghz (C/C++)
157 voxelrcnn 64.77 % 73.60 % 60.05 % 15 s 1 core @ 2.5 Ghz (C/C++)
158 3DNN 64.74 % 76.32 % 58.10 % 0.09 s GPU @ 2.5 Ghz (Python)
159 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.
160 RCD 60.56 % 70.54 % 55.58 % 0.1 s GPU @ 2.5 Ghz (Python)
161 ANM 59.07 % 74.99 % 47.64 % 0.12 s 1 core @ 2.5 Ghz (Python)
162 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.
163 PL++ (SDN+GDC)
This method uses stereo information.
This method makes use of Velodyne laser scans.
code 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.
164 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.
165 CDN
This method uses stereo information.
54.22 % 74.52 % 46.36 % 0.6 s GPU @ 2.5 Ghz (Python)
D. Garg, Y. Wang, B. Hariharan, M. Campbell, K. Weinberger and W. Chao: Wasserstein Distances for Stereo Disparity Estimation. Advances in Neural Information Processing Systems (NeurIPS) 2020.
166 CG-Stereo
This method uses stereo information.
53.58 % 74.39 % 46.50 % 0.57 s GeForce RTX 2080 Ti
C. Li, J. Ku and S. Waslander: Confidence Guided Stereo 3D Object Detection with Split Depth Estimation. IROS 2020.
167 tiny-stereo-v2
This method uses stereo information.
53.36 % 71.12 % 47.73 % 0.4 s 1 core @ 2.5 Ghz (Python + C/C++)
168 tiny-stereo-v1
This method uses stereo information.
52.99 % 74.27 % 45.24 % 0.3 s GPU @ 2.5 Ghz (Python + C/C++)
169 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. CVPR 2020.
170 SF
This method uses stereo information.
This method makes use of Velodyne laser scans.
51.92 % 58.88 % 44.59 % 0.5 s 1 core @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
171 BirdNet+
This method makes use of Velodyne laser scans.
code 51.85 % 70.14 % 50.03 % 0.1 s Titan Xp (PyTorch)
A. Barrera, C. Guindel, J. Beltrán and F. García: BirdNet+: End-to-End 3D Object Detection in LiDAR Bird's Eye View. arXiv:2003.04188 [cs.CV] 2020.
172 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.
173 CDN-PL++
This method uses stereo information.
44.86 % 64.31 % 38.11 % 0.4 s GPU @ 2.5 Ghz (C/C++)
D. Garg, Y. Wang, B. Hariharan, M. Campbell, K. Weinberger and W. Chao: Wasserstein Distances for Stereo Disparity Estimation. arXiv preprint arXiv:2007.03085 2020.
174 Pseudo-LiDAR E2E
This method uses stereo information.
43.92 % 64.75 % 38.14 % 0.4 s GPU @ 2.5 Ghz (Python)
175 PB3D
This method uses stereo information.
43.27 % 64.78 % 37.13 % 0.42 s 1 core @ 2.5 Ghz (C/C++)
176 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.
177 Stereo3D
This method uses stereo information.
41.25 % 65.68 % 30.42 % 0.1 s GPU 1080Ti
178 Disp R-CNN (velo)
This method uses stereo information.
code 39.36 % 59.61 % 32.01 % 0.42 s GPU @ 2.5 Ghz (Python + C/C++)
J. Sun, L. Chen, Y. Xie, S. Zhang, Q. Jiang, X. Zhou and H. Bao: Disp R-CNN: Stereo 3D Object Detection via Shape Prior Guided Instance Disparity Estimation. CVPR 2020.
179 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.
180 Disp R-CNN
This method uses stereo information.
code 37.93 % 58.55 % 31.95 % 0.42 s GPU @ 2.5 Ghz (Python + C/C++)
J. Sun, L. Chen, Y. Xie, S. Zhang, Q. Jiang, X. Zhou and H. Bao: Disp R-CNN: Stereo 3D Object Detection via Shape Prior Guided Instance Disparity Estimation. CVPR 2020.
181 stereo_sa
This method uses stereo information.
37.92 % 58.70 % 31.99 % 0.3 s GPU @ 2.5 Ghz (Python)
182 OC Stereo
This method uses stereo information.
code 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. ICRA 2020.
183 RTS3D 37.38 % 58.51 % 31.12 % 0.03 s GPU @ 2.5 Ghz (Python)
184 Pseudo-Lidar
This method uses stereo information.
code 34.05 % 54.53 % 28.25 % 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.
185 m-prcnn
This method uses stereo information.
31.21 % 53.96 % 24.52 % 0.43 s 1 core @ 2.5 Ghz (Python)
186 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.
187 IDA-3D
This method uses stereo information.
29.32 % 45.09 % 23.13 % 0.08 s 1 core @ 2.5 Ghz (Python + C/C++)
188 BirdNet
This method makes use of Velodyne laser scans.
27.26 % 40.99 % 25.32 % 0.11 s Titan Xp (Caffe)
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.
189 RT3D-GMP
This method uses stereo information.
23.83 % 32.44 % 17.91 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
190 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.
191 ASOD 22.37 % 38.42 % 17.01 % 0.28 s GPU @ 2.5 Ghz (Python)
192 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.
193 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.
194 ITS-MDPL 14.21 % 23.81 % 12.11 % 0.16 s GPU @ 2.5 Ghz (Python)
195 PSMD 13.57 % 21.37 % 10.89 % 0.1 s GPU @ 2.5 Ghz (Python)
196 deprecated 13.30 % 14.81 % 11.04 % 1 core @ 2.5 Ghz (C/C++)
197 Det3D 13.26 % 24.00 % 9.94 % 0.5 s 1 core @ 2.5 Ghz (C/C++)
198 S3D 12.75 % 14.58 % 10.72 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
199 Kinematic3D code 12.72 % 19.07 % 9.17 % 0.12 s 1 core @ 1.5 Ghz (C/C++)
G. Brazil, G. Pons-Moll, X. Liu and B. Schiele: Kinematic 3D Object Detection in Monocular Video. ECCV 2020 .
200 MTMono3d 12.44 % 18.54 % 10.09 % 0.05 s 1 core @ 2.5 Ghz (C/C++)
201 DP3D 12.24 % 18.84 % 8.96 % 0.07 s GPU @ 1.5 Ghz (Python + C/C++)
202 YoloMono3D 12.06 % 18.28 % 8.42 % 0.05 s GPU @ 2.5 Ghz (Python)
203 IAFA 12.01 % 17.81 % 10.61 % 0.04 s 1 core @ 2.5 Ghz (C/C++)
204 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. CVPR 2020.
205 MP-Mono 11.65 % 16.78 % 9.01 % 0.16 s GPU @ 2.5 Ghz (Python)
206 MCA 11.63 % 18.46 % 10.24 % 0.04 s 1 core @ 2.5 Ghz (C/C++)
207 PG-MonoNet 11.51 % 15.91 % 9.01 % 0.19 s GPU @ 2.5 Ghz (Python)
208 DA-3Ddet 11.50 % 16.77 % 8.93 % 0.4 s GPU @ 2.5 Ghz (Python)
209 NL_M3D 11.46 % 17.54 % 8.98 % 0.2 s 1 core @ 2.5 Ghz (Python)
210 SSL-RTM3D 11.45 % 16.73 % 9.92 % 0.03 s 1 core @ 2.5 Ghz (Python)
211 IMA 11.34 % 16.24 % 9.44 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
212 CDI3D 11.32 % 15.70 % 9.26 % 0.03 s GPU @ 2.5 Ghz (Python)
213 LAPNet 11.29 % 18.02 % 8.50 % 0.03 s 1 core @ 2.5 Ghz (Python)
214 DP3D 11.22 % 17.27 % 8.54 % 0.05 s GPU @ 1.5 Ghz (Python + C/C++)
215 LNET 11.21 % 12.79 % 9.94 % 0.05 s 1 core @ 2.5 Ghz (C/C++)
216 RefinedMPL 11.14 % 18.09 % 8.94 % 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.
217 UM3D_TUM 11.13 % 15.30 % 9.31 % 0.05 s 1 core @ 2.5 Ghz (Python)
218 PatchNet code 11.12 % 15.68 % 10.17 % 0.4 s 1 core @ 2.5 Ghz (C/C++)
X. Ma, S. Liu, Z. Xia, H. Zhang, X. Zeng and W. Ouyang: Rethinking Pseudo-LiDAR Representation. Proceedings of the European Conference on Computer Vision (ECCV) 2020.
219 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.
220 OCM3D 10.44 % 17.48 % 7.87 % 0.5 s 1 core @ 2.5 Ghz (C/C++)
221 RTM3D code 10.34 % 14.41 % 8.77 % 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.
222 MA 10.21 % 14.90 % 8.78 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
223 MonoPair 9.99 % 13.04 % 8.65 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
Y. Chen, L. Tai, K. Sun and M. Li: MonoPair: Monocular 3D Object Detection Using Pairwise Spatial Relationships. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2020.
224 SMOKE code 9.76 % 14.03 % 7.84 % 0.03 s GPU @ 2.5 Ghz (Python)
Z. Liu, Z. Wu and R. Tóth: SMOKE: Single-Stage Monocular 3D Object Detection via Keypoint Estimation. 2020.
225 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 .
226 SS3D_HW 9.70 % 14.74 % 7.22 % 0.4 s GPU @ 2.5 Ghz (Python)
227 Center3D 9.31 % 12.01 % 8.06 % 0.05 s GPU @ 3.5 Ghz (Python)
228 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.
229 Mono3CN 9.17 % 12.73 % 7.91 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
230 LCD3D 9.04 % 13.77 % 7.23 % 0.03 s GPU @ 2.5 Ghz (Python)
231 RAR-Net 8.73 % 13.70 % 6.92 % 0.5 s 1 core @ 2.5 Ghz (C/C++)
232 SSL-RTM3D Res18 8.39 % 12.65 % 7.12 % 0.02 s GPU @ 2.5 Ghz (Python)
233 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.
234 anonymous 7.66 % 15.21 % 6.24 % 1 s 1 core @ 2.5 Ghz (C/C++)
235 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.
236 Decoupled-3D v2 7.28 % 11.68 % 5.69 % 0.08 s GPU @ 2.5 Ghz (C/C++)
237 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.
238 Decoupled-3D 7.02 % 11.08 % 5.63 % 0.08 s GPU @ 2.5 Ghz (C/C++)
Y. Cai, B. Li, Z. Jiao, H. Li, X. Zeng and X. Wang: Monocular 3D Object Detection with Decoupled Structured Polygon Estimation and Height-Guided Depth Estimation. AAAI 2020.
239 anonymous 6.77 % 13.18 % 5.63 % 1 s 1 core @ 2.5 Ghz (C/C++)
240 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.
241 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.
242 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.
243 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.
244 OACV 4.77 % 8.13 % 3.78 % 0.23 s GPU @ 2.5 Ghz (Python)
245 TLNet (Stereo)
This method uses stereo information.
code 4.37 % 7.64 % 3.74 % 0.1 s 1 core @ 2.5 Ghz (Python)
Z. Qin, J. Wang and Y. Lu: Triangulation Learning Network: from Monocular to Stereo 3D Object Detection. IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2019.
246 AACL 4.18 % 5.62 % 3.34 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
247 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.
248 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.
249 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.
250 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.
251 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.
252 3D-GCK 2.52 % 3.27 % 2.11 % 24 ms Tesla V100
N. Gählert, J. Wan, N. Jourdan, J. Finkbeiner, U. Franke and J. Denzler: Single-Shot 3D Detection of Vehicles from Monocular RGB Images via Geometrically Constrained Keypoints in Real-Time. 2020 IEEE Intelligent Vehicles Symposium (IV) 2020.
253 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.
254 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.
255 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.
256 UDI-mono3D 0.41 % 0.51 % 0.43 % 0.05 s 1 core @ 2.5 Ghz (Python)
257 PVNet 0.00 % 0.00 % 0.00 % 0,1 s 1 core @ 2.5 Ghz (Python)
258 ANM 0.00 % 0.00 % 0.00 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
259 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 HotSpotNet 45.37 % 53.10 % 41.47 % 0.04 s 1 core @ 2.5 Ghz (Python + C/C++)
Q. Chen, L. Sun, Z. Wang, K. Jia and A. Yuille: object as hotspots. Proceedings of the European Conference on Computer Vision (ECCV) 2020.
2 Noah CV Lab - SSL 45.23 % 52.85 % 41.28 % 0.1 s GPU @ 2.5 Ghz (Python)
3 VICNet 44.80 % 54.00 % 41.11 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
4 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.
5 3DSSD code 44.27 % 54.64 % 40.23 % 0.04 s GPU @ 2.5 Ghz (Python + C/C++)
Z. Yang, Y. Sun, S. Liu and J. Jia: 3DSSD: Point-based 3D Single Stage Object Detector. CVPR 2020.
6 PPBA 44.08 % 52.65 % 41.54 % NA s GPU @ 2.5 Ghz (Python)
7 CentrNet-FG 44.02 % 53.51 % 40.53 % 0.03 s 1 core @ 2.5 Ghz (C/C++)
8 Point-GNN
This method makes use of Velodyne laser scans.
code 43.77 % 51.92 % 40.14 % 0.6 s GPU @ 2.5 Ghz (Python)
W. Shi and R. Rajkumar: Point-GNN: Graph Neural Network for 3D Object Detection in a Point Cloud. CVPR 2020.
9 PP-3D 43.77 % 51.92 % 40.14 % 0.1 s 1 core @ 2.5 Ghz (Python)
10 MVX-Net++ 43.73 % 50.90 % 39.96 % 0.15 s 1 core @ 2.5 Ghz (C/C++)
11 KNN-GCNN 43.57 % 51.82 % 40.02 % 0.4 s 1 core @ 2.5 Ghz (Python + C/C++)
12 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.
13 MMLab-PartA^2
This method makes use of Velodyne laser scans.
code 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. IEEE Transactions on Pattern Analysis and Machine Intelligence 2020.
14 MMLab PV-RCNN
This method makes use of Velodyne laser scans.
code 43.29 % 52.17 % 40.29 % 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. CVPR 2020.
15 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.
16 RethinkDet3D 43.25 % 53.13 % 40.58 % 0.15 s 1 core @ 2.5 Ghz (Python)
17 STD code 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.
18 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.
19 SemanticVoxels 42.19 % 50.90 % 39.52 % 0.04 s GPU @ 2.5 Ghz (Python + C/C++)
J. Fei, W. Chen, P. Heidenreich, S. Wirges and C. Stiller: SemanticVoxels: Sequential Fusion for 3D Pedestrian Detection using LiDAR Point Cloud and Semantic Segmentation. MFI 2020.
20 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.
21 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.
22 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.
23 CentrNet-v1
This method makes use of Velodyne laser scans.
41.50 % 50.86 % 38.24 % 0.03 s GPU @ 2.5 Ghz (Python)
24 TBU 41.16 % 49.33 % 38.84 % NA s GPU @ 2.5 Ghz (Python)
25 PiP 41.01 % 49.01 % 37.90 % 0.033 s 1 core @ 2.5 Ghz (Python)
26 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. CVPR 2020.
27 Deformable PV-RCNN
This method makes use of Velodyne laser scans.
code 40.89 % 46.97 % 38.80 % 0.08 s 1 core @ 2.5 Ghz (Python + C/C++)
P. Bhattacharyya and K. Czarnecki: Deformable PV-RCNN: Improving 3D Object Detection with Learned Deformations. ECCV 2020 Perception for Autonomous Driving Workshop.
28 DDB
This method makes use of Velodyne laser scans.
40.40 % 49.03 % 37.04 % 0.05 s GPU @ 2.5 Ghz (Python)
29 Simple3D Net 40.20 % 48.41 % 37.50 % 0.02 s GPU @ 2.5 Ghz (Python)
30 PPFNet code 40.11 % 48.36 % 37.00 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
31 IC-PVRCNN 39.46 % 45.19 % 37.16 % 0.08 s 1 core @ 2.5 Ghz (C/C++)
32 SVGA-Net
This method makes use of Velodyne laser scans.
39.43 % 47.30 % 36.99 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
33 Baseline of CA RCNN 39.42 % 47.30 % 36.97 % 0.1 s GPU @ 2.5 Ghz (Python)
34 CVIS-DF3D 39.42 % 47.30 % 36.97 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
35 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.
36 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.
37 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.
38 CVIS-DF3D_v2 38.31 % 45.10 % 36.15 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
39 3DBN_2 38.23 % 46.79 % 35.57 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
ERROR: Wrong syntax in BIBTEX file.
40 IGRP+ 38.05 % 46.26 % 34.53 % 0.18 s 1 core @ 2.5 Ghz (Python + C/C++)
41 MGACNet 37.50 % 43.55 % 35.33 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
42 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.
43 TBD 37.37 % 43.60 % 34.36 % 0.05 s GPU @ 2.5 Ghz (Python)
44 IC-SECOND 37.18 % 43.82 % 35.35 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
45 VOXEL_FPN_HR 37.01 % 46.32 % 34.67 % 0.12 s 8 cores @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
46 PFF3D
This method makes use of Velodyne laser scans.
36.07 % 43.93 % 32.86 % 0.05 s GPU @ 3.0 Ghz (Python + C/C++)
47 NLK-3D 35.86 % 45.17 % 32.24 % 0.04 s 1 core @ 2.5 Ghz (Python + C/C++)
48 HR-SECOND code 35.52 % 45.31 % 33.14 % 0.11 s 1 core @ 2.5 Ghz (Python + C/C++)
49 SRDL
This method uses stereo information.
This method makes use of Velodyne laser scans.
35.28 % 42.66 % 33.26 % 0.15 s GPU @ 2.5 Ghz (Python + C/C++)
50 deprecated 35.21 % 41.32 % 33.32 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
51 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.
52 PBASN code 34.48 % 41.28 % 32.24 % NA s GPU @ 2.5 Ghz (Python)
53 NLK-ALL code 34.46 % 44.30 % 30.83 % 0.04 s 1 core @ 2.5 Ghz (Python + C/C++)
54 MP 33.89 % 43.04 % 31.46 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
55 DAMNET code 33.66 % 43.32 % 30.12 % 1 s 1 core @ 2.5 Ghz (C/C++)
56 BirdNet+
This method makes use of Velodyne laser scans.
code 31.46 % 37.99 % 29.46 % 0.1 s Titan Xp (PyTorch)
A. Barrera, C. Guindel, J. Beltrán and F. García: BirdNet+: End-to-End 3D Object Detection in LiDAR Bird's Eye View. arXiv:2003.04188 [cs.CV] 2020.
57 Pointpillar_TV 30.79 % 38.56 % 28.57 % 0.05 s 1 core @ 2.5 Ghz (C/C++)
58 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.
59 FCY
This method makes use of Velodyne laser scans.
29.38 % 37.28 % 26.19 % 0.02 s GPU @ 2.5 Ghz (Python)
60 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.
61 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.
62 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.
63 SF
This method uses stereo information.
This method makes use of Velodyne laser scans.
24.84 % 31.61 % 21.96 % 0.5 s 1 core @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
64 CG-Stereo
This method uses stereo information.
24.31 % 33.22 % 20.95 % 0.57 s GeForce RTX 2080 Ti
C. Li, J. Ku and S. Waslander: Confidence Guided Stereo 3D Object Detection with Split Depth Estimation. IROS 2020.
65 Disp R-CNN (velo)
This method uses stereo information.
code 21.98 % 30.98 % 18.68 % 0.42 s GPU @ 2.5 Ghz (Python + C/C++)
J. Sun, L. Chen, Y. Xie, S. Zhang, Q. Jiang, X. Zhou and H. Bao: Disp R-CNN: Stereo 3D Object Detection via Shape Prior Guided Instance Disparity Estimation. CVPR 2020.
66 Disp R-CNN
This method uses stereo information.
code 21.98 % 31.05 % 18.67 % 0.42 s GPU @ 2.5 Ghz (Python + C/C++)
J. Sun, L. Chen, Y. Xie, S. Zhang, Q. Jiang, X. Zhou and H. Bao: Disp R-CNN: Stereo 3D Object Detection via Shape Prior Guided Instance Disparity Estimation. CVPR 2020.
67 PB3D
This method uses stereo information.
20.65 % 28.68 % 17.65 % 0.42 s 1 core @ 2.5 Ghz (C/C++)
68 Stereo3D
This method uses stereo information.
19.75 % 28.49 % 16.48 % 0.1 s GPU 1080Ti
69 OC Stereo
This method uses stereo information.
code 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. ICRA 2020.
70 BirdNet
This method makes use of Velodyne laser scans.
17.08 % 22.04 % 15.82 % 0.11 s Titan Xp (Caffe)
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.
71 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. CVPR 2020.
72 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.
73 RefinedMPL 7.18 % 11.14 % 5.84 % 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.
74 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.
75 MonoPair 6.68 % 10.02 % 5.53 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
Y. Chen, L. Tai, K. Sun and M. Li: MonoPair: Monocular 3D Object Detection Using Pairwise Spatial Relationships. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2020.
76 SS3D_HW 5.00 % 7.77 % 4.03 % 0.4 s GPU @ 2.5 Ghz (Python)
77 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.
78 PG-MonoNet 4.50 % 5.76 % 3.93 % 0.19 s GPU @ 2.5 Ghz (Python)
79 CDI3D 4.03 % 5.64 % 3.29 % 0.03 s GPU @ 2.5 Ghz (Python)
80 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.
81 NL_M3D 3.87 % 5.16 % 3.08 % 0.2 s 1 core @ 2.5 Ghz (Python)
82 MP-Mono 3.79 % 5.30 % 3.15 % 0.16 s GPU @ 2.5 Ghz (Python)
83 DP3D 3.54 % 4.75 % 2.88 % 0.05 s GPU @ 1.5 Ghz (Python + C/C++)
84 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 .
85 Mono3CN 3.44 % 5.13 % 3.00 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
86 Center3D 3.43 % 4.86 % 2.78 % 0.05 s GPU @ 3.5 Ghz (Python)
87 RT3D-GMP
This method uses stereo information.
3.42 % 4.51 % 2.77 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
88 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. CVPR 2020.
89 DP3D 3.37 % 4.77 % 2.77 % 0.07 s GPU @ 1.5 Ghz (Python + C/C++)
90 LAPNet 3.16 % 4.41 % 2.70 % 0.03 s 1 core @ 2.5 Ghz (Python)
91 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.
92 MTMono3d 2.05 % 2.40 % 1.68 % 0.05 s 1 core @ 2.5 Ghz (C/C++)
93 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.
94 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.
95 UM3D_TUM 1.74 % 3.49 % 1.74 % 0.05 s 1 core @ 2.5 Ghz (Python)
96 UDI-mono3D 1.01 % 1.81 % 0.99 % 0.05 s 1 core @ 2.5 Ghz (Python)
97 PVNet 0.00 % 0.00 % 0.00 % 0,1 s 1 core @ 2.5 Ghz (Python)
98 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 HRI-MSP-L
This method makes use of Velodyne laser scans.
71.86 % 87.77 % 63.57 % 0.07 s 1 core @ 2.5 Ghz (C/C++)
2 Noah CV Lab - SSL 71.53 % 84.24 % 62.20 % 0.1 s GPU @ 2.5 Ghz (Python)
3 Deformable PV-RCNN
This method makes use of Velodyne laser scans.
code 68.54 % 82.19 % 61.33 % 0.08 s 1 core @ 2.5 Ghz (Python + C/C++)
P. Bhattacharyya and K. Czarnecki: Deformable PV-RCNN: Improving 3D Object Detection with Learned Deformations. ECCV 2020 Perception for Autonomous Driving Workshop.
4 HotSpotNet 65.95 % 82.59 % 59.00 % 0.04 s 1 core @ 2.5 Ghz (Python + C/C++)
Q. Chen, L. Sun, Z. Wang, K. Jia and A. Yuille: object as hotspots. Proceedings of the European Conference on Computer Vision (ECCV) 2020.
5 TBD 65.64 % 82.29 % 57.98 % 0.05 s GPU @ 2.5 Ghz (Python)
6 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.
7 IC-PVRCNN 64.99 % 81.47 % 58.62 % 0.08 s 1 core @ 2.5 Ghz (C/C++)
8 3DSSD code 64.10 % 82.48 % 56.90 % 0.04 s GPU @ 2.5 Ghz (Python + C/C++)
Z. Yang, Y. Sun, S. Liu and J. Jia: 3DSSD: Point-based 3D Single Stage Object Detector. CVPR 2020.
9 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. CVPR 2020.
10 MMLab PV-RCNN
This method makes use of Velodyne laser scans.
code 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. CVPR 2020.
11 IC-SECOND 63.65 % 78.29 % 57.27 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
12 NLK-ALL code 63.65 % 79.94 % 57.28 % 0.04 s 1 core @ 2.5 Ghz (Python + C/C++)
13 MMLab-PartA^2
This method makes use of Velodyne laser scans.
code 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. IEEE Transactions on Pattern Analysis and Machine Intelligence 2020.
14 Point-GNN
This method makes use of Velodyne laser scans.
code 63.48 % 78.60 % 57.08 % 0.6 s GPU @ 2.5 Ghz (Python)
W. Shi and R. Rajkumar: Point-GNN: Graph Neural Network for 3D Object Detection in a Point Cloud. CVPR 2020.
15 PP-3D 63.48 % 78.60 % 57.08 % 0.1 s 1 core @ 2.5 Ghz (Python)
16 CVIS-DF3D_v2 63.05 % 77.46 % 55.43 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
17 KNN-GCNN 62.91 % 80.24 % 56.49 % 0.4 s 1 core @ 2.5 Ghz (Python + C/C++)
18 deprecated 62.16 % 75.45 % 56.00 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
19 SRDL
This method uses stereo information.
This method makes use of Velodyne laser scans.
62.03 % 80.66 % 55.73 % 0.15 s GPU @ 2.5 Ghz (Python + C/C++)
20 SVGA-Net
This method makes use of Velodyne laser scans.
62.02 % 77.35 % 55.52 % 0.08 s GPU @ 2.5 Ghz (Python + C/C++)
21 Baseline of CA RCNN 62.02 % 77.33 % 55.52 % 0.1 s GPU @ 2.5 Ghz (Python)
22 CVIS-DF3D 62.02 % 77.33 % 55.52 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
23 MGACNet 62.00 % 78.73 % 55.18 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
24 PPBA 61.99 % 79.42 % 55.34 % NA s GPU @ 2.5 Ghz (Python)
25 TBU 61.99 % 79.42 % 55.34 % NA s GPU @ 2.5 Ghz (Python)
26 VOXEL_FPN_HR 61.91 % 78.29 % 55.54 % 0.12 s 8 cores @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
27 STD code 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.
28 RethinkDet3D 61.10 % 79.31 % 54.47 % 0.15 s 1 core @ 2.5 Ghz (Python)
29 MVX-Net++ 61.03 % 76.07 % 53.41 % 0.15 s 1 core @ 2.5 Ghz (C/C++)
30 3DBN_2 60.88 % 78.10 % 54.10 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
ERROR: Wrong syntax in BIBTEX file.
31 HR-SECOND code 60.82 % 75.83 % 53.67 % 0.11 s 1 core @ 2.5 Ghz (Python + C/C++)
32 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.
33 VICNet 59.99 % 78.75 % 52.37 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
34 FCY
This method makes use of Velodyne laser scans.
59.54 % 76.30 % 52.29 % 0.02 s GPU @ 2.5 Ghz (Python)
35 PiP 59.54 % 75.43 % 53.37 % 0.033 s 1 core @ 2.5 Ghz (Python)
36 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.
37 PBASN code 59.43 % 76.80 % 52.77 % NA s GPU @ 2.5 Ghz (Python)
38 NLK-3D 59.30 % 76.45 % 51.82 % 0.04 s 1 core @ 2.5 Ghz (Python + C/C++)
39 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.
40 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.
41 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.
42 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.
43 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.
44 CentrNet-FG 55.54 % 72.07 % 49.03 % 0.03 s 1 core @ 2.5 Ghz (C/C++)
45 MP 55.36 % 72.99 % 49.36 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
46 Pointpillar_TV 54.69 % 71.61 % 48.22 % 0.05 s 1 core @ 2.5 Ghz (C/C++)
47 CentrNet-v1
This method makes use of Velodyne laser scans.
54.64 % 72.03 % 48.03 % 0.03 s GPU @ 2.5 Ghz (Python)
48 Simple3D Net 54.49 % 70.79 % 48.21 % 0.02 s GPU @ 2.5 Ghz (Python)
49 IGRP+ 53.22 % 69.87 % 47.55 % 0.18 s 1 core @ 2.5 Ghz (Python + C/C++)
50 DDB
This method makes use of Velodyne laser scans.
51.38 % 68.83 % 45.15 % 0.05 s GPU @ 2.5 Ghz (Python)
51 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.
52 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.
53 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.
54 BirdNet+
This method makes use of Velodyne laser scans.
code 47.72 % 67.38 % 42.89 % 0.1 s Titan Xp (PyTorch)
A. Barrera, C. Guindel, J. Beltrán and F. García: BirdNet+: End-to-End 3D Object Detection in LiDAR Bird's Eye View. arXiv:2003.04188 [cs.CV] 2020.
55 PFF3D
This method makes use of Velodyne laser scans.
46.78 % 63.27 % 41.37 % 0.05 s GPU @ 3.0 Ghz (Python + C/C++)
56 DAMNET code 42.82 % 58.71 % 38.74 % 1 s 1 core @ 2.5 Ghz (C/C++)
57 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.
58 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.
59 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.
60 CG-Stereo
This method uses stereo information.
30.89 % 47.40 % 27.23 % 0.57 s GeForce RTX 2080 Ti
C. Li, J. Ku and S. Waslander: Confidence Guided Stereo 3D Object Detection with Split Depth Estimation. IROS 2020.
61 BirdNet
This method makes use of Velodyne laser scans.
30.25 % 43.98 % 27.21 % 0.11 s Titan Xp (Caffe)
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.
62 Disp R-CNN (velo)
This method uses stereo information.
code 23.75 % 39.72 % 20.47 % 0.42 s GPU @ 2.5 Ghz (Python + C/C++)
J. Sun, L. Chen, Y. Xie, S. Zhang, Q. Jiang, X. Zhou and H. Bao: Disp R-CNN: Stereo 3D Object Detection via Shape Prior Guided Instance Disparity Estimation. CVPR 2020.
63 Disp R-CNN
This method uses stereo information.
code 23.75 % 39.72 % 20.47 % 0.42 s GPU @ 2.5 Ghz (Python + C/C++)
J. Sun, L. Chen, Y. Xie, S. Zhang, Q. Jiang, X. Zhou and H. Bao: Disp R-CNN: Stereo 3D Object Detection via Shape Prior Guided Instance Disparity Estimation. CVPR 2020.
64 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.
65 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. CVPR 2020.
66 PB3D
This method uses stereo information.
17.28 % 28.50 % 15.56 % 0.42 s 1 core @ 2.5 Ghz (C/C++)
67 OC Stereo
This method uses stereo information.
code 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. ICRA 2020.
68 RT3D-GMP
This method uses stereo information.
4.90 % 7.75 % 4.18 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
69 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.
70 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.
71 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.
72 CDI3D 2.69 % 4.15 % 2.45 % 0.03 s GPU @ 2.5 Ghz (Python)
73 Center3D 2.35 % 4.32 % 2.06 % 0.05 s GPU @ 3.5 Ghz (Python)
74 Mono3CN 2.17 % 3.68 % 2.02 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
75 SS3D_HW 2.17 % 4.29 % 2.00 % 0.4 s GPU @ 2.5 Ghz (Python)
76 MonoPair 2.12 % 3.79 % 1.83 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
Y. Chen, L. Tai, K. Sun and M. Li: MonoPair: Monocular 3D Object Detection Using Pairwise Spatial Relationships. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2020.
77 RefinedMPL 1.82 % 3.23 % 1.77 % 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.
78 UDI-mono3D 1.74 % 3.29 % 1.37 % 0.05 s 1 core @ 2.5 Ghz (Python)
79 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.
80 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. CVPR 2020.
81 DP3D 1.66 % 2.77 % 1.31 % 0.07 s GPU @ 1.5 Ghz (Python + C/C++)
82 NL_M3D 1.51 % 2.10 % 1.58 % 0.2 s 1 core @ 2.5 Ghz (Python)
83 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.
84 PG-MonoNet 1.43 % 2.41 % 1.23 % 0.19 s GPU @ 2.5 Ghz (Python)
85 MP-Mono 1.42 % 1.89 % 1.29 % 0.16 s GPU @ 2.5 Ghz (Python)
86 DP3D 1.39 % 2.04 % 1.12 % 0.05 s GPU @ 1.5 Ghz (Python + C/C++)
87 MTMono3d 0.90 % 1.59 % 0.96 % 0.05 s 1 core @ 2.5 Ghz (C/C++)
88 LAPNet 0.89 % 1.37 % 0.62 % 0.03 s 1 core @ 2.5 Ghz (Python)
89 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 .
90 UM3D_TUM 0.62 % 0.45 % 0.62 % 0.05 s 1 core @ 2.5 Ghz (Python)
91 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.
92 PVNet 0.00 % 0.00 % 0.00 % 0,1 s 1 core @ 2.5 Ghz (Python)
93 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

Measure
Measure
Summary | 5 Annotations
objects in don't car areas do not count as false positives
2020/10/12 09:44
For cars we require an 3D bounding box overlap of 70%
2020/10/12 09:44
pedestrians and cyclists we require a 3D bounding box overlap of 50%
2020/10/12 09:44
ranked based on the moderately difficult results
2020/10/12 09:45
SECOND
2020/10/12 09:45