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

Object Detection Evaluation 2012


The object detection and object orientation estimation benchmark consists of 7481 training images and 7518 test images, comprising a total of 80.256 labeled objects. All images are color and saved as png. For evaluation, we compute precision-recall curves for object detection and orientation-similarity-recall curves for joint object detection and orientation estimation. In the latter case not only the object 2D bounding box has to be located correctly, but also the orientation estimate in bird's eye view is evaluated. To rank the methods we compute average precision and average orientation similiarity. 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 object detection performance using the PASCAL criteria and object detection and orientation estimation performance using the measure discussed in our CVPR 2012 publication. For cars we require an overlap of 70%, while for pedestrians and cyclists we require an overlap of 50% for a detection. Detections in don't care areas or detections which are smaller than the minimum size do not count as false positive. 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 that for the hard evaluation ~2 % of the provided bounding boxes have not been recognized by humans, thereby upper bounding recall at 98 %. Hence, the hard evaluation is only given for reference.
Note 1: On 25.04.2017, we have fixed a bug in the object detection evaluation script. As of now, the submitted detections are filtered based on the min. bounding box height for the respective category which we have been done before only for the ground truth detections, thus leading to false positives for the category "Easy" when bounding boxes of height 25-39 Px were submitted (and to false positives for all categories if bounding boxes smaller than 25 Px were submitted). We like to thank Amy Wu, Matt Wilder, Pekka Jänis and Philippe Vandermersch for their feedback. The last leaderboards right before the changes can be found here!

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 PC-CNN-V2
This method makes use of Velodyne laser scans.
 95.20 % 96.06 % 89.37 % 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.
2 F-PointNet
This method makes use of Velodyne laser scans.
 code 95.17 % 95.85 % 85.42 % 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.
3 SA-SSD 95.16 % 97.92 % 90.15 % 0.04 s 1 core @ 2.5 Ghz (Python)
4 3DSSD 95.10 % 97.69 % 92.18 % 0.04 s GPU @ 2.5 Ghz (Python + C/C++)
5 MVRA + I-FRCNN+ 94.98 % 95.87 % 82.52 % 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.
6 MMLab PV-RCNN
This method makes use of Velodyne laser scans.
 94.70 % 98.17 % 92.04 % 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.
7 BM-NET 94.49 % 95.09 % 85.06 % 0.5 s GPU @ 2.5 Ghz (Python + C/C++)
8 TuSimple code 94.47 % 95.12 % 86.45 % 1.6 s GPU @ 2.5 Ghz (Python + C/C++)
F. Yang, W. Choi and Y. Lin: Exploit all the layers: Fast and accurate cnn object detector with scale dependent pooling and cascaded rejection classifiers. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition 2016.
K. He, X. Zhang, S. Ren and J. Sun: Deep residual learning for image recognition. Proceedings of the IEEE conference on computer vision and pattern recognition 2016.
9 EPNet 94.44 % 96.15 % 89.99 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
10 CPRCCNN 94.42 % 96.33 % 89.96 % 0.1 s 1 core @ 2.5 Ghz (Python)
11 UberATG-MMF
This method makes use of Velodyne laser scans.
 94.25 % 97.41 % 89.87 % 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.
12 DGIST-CellBox 93.90 % 95.86 % 88.26 % 0.1 s GPU @ 2.5 Ghz (Java + C/C++)
13 Patches - EMP
This method makes use of Velodyne laser scans.
 93.75 % 97.91 % 90.56 % 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.
14 Noah CV Lab - SSL 93.65 % 94.02 % 86.02 % 0.1 s GPU @ 2.5 Ghz (Python)
15 THICV-YDM 93.60 % 96.26 % 81.08 % 0.06 s GPU @ 2.5 Ghz (Python)
16 MLF_PointCas 93.55 % 96.69 % 86.16 % 0.1 s GPU @ 2.5 Ghz (Python)
17 MonoPair 93.55 % 96.61 % 83.55 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
18 Deep MANTA 93.50 % 98.89 % 83.21 % 0.7 s GPU @ 2.5 Ghz (Python + C/C++)
F. Chabot, M. Chaouch, J. Rabarisoa, C. Teulière and T. Chateau: Deep MANTA: A Coarse-to-fine Many-Task Network for joint 2D and 3D vehicle analysis from monocular image. CVPR 2017.
19 Point-GNN
This method makes use of Velodyne laser scans.
 93.50 % 96.58 % 88.35 % 0.6 s GPU @ 2.5 Ghz (Python)
20 FichaDL 93.46 % 96.00 % 84.39 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
21 RRC code 93.40 % 95.68 % 87.37 % 3.6 s GPU @ 2.5 Ghz (C/C++)
J. Ren, X. Chen, J. Liu, W. Sun, J. Pang, Q. Yan, Y. Tai and L. Xu: Accurate Single Stage Detector Using Recurrent Rolling Convolution. CVPR 2017.
22 ORP 93.27 % 96.76 % 85.86 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
23 CFENet 93.26 % 93.91 % 86.99 % 4 s GPU @ 2.5 Ghz (Python + C/C++)
24 STD 93.22 % 96.14 % 90.53 % 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.
25 SARPNET 93.21 % 96.07 % 88.09 % 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.
26 Fast Point R-CNN
This method makes use of Velodyne laser scans.
 93.18 % 96.13 % 87.68 % 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.
27 sensekitti code 93.17 % 94.79 % 84.38 % 4.5 s GPU @ 2.5 Ghz (Python + C/C++)
B. Yang, J. Yan, Z. Lei and S. Li: Craft Objects from Images. CVPR 2016.
28 ELE 93.14 % 98.44 % 90.32 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
29 SJTU-HW 93.11 % 96.30 % 82.21 % 0.85s GPU @ 1.5 Ghz (Python + C/C++)
S. Zhang, X. Zhao, L. Fang, F. Haiping and S. Haitao: LED: LOCALIZATION-QUALITY ESTIMATION EMBEDDED DETECTOR. IEEE International Conference on Image Processing 2018.
L. Fang, X. Zhao and S. Zhang: Small-objectness sensitive detection based on shifted single shot detector. Multimedia Tools and Applications 2018.
30 RGB3D
This method makes use of Velodyne laser scans.
 93.07 % 96.54 % 88.04 % 0.39 s GPU @ 2.5 Ghz (Python)
31 PointRCNN-deprecated
This method makes use of Velodyne laser scans.
 92.96 % 96.72 % 85.81 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
32 SerialR-FCN+SG-NMS 92.93 % 95.72 % 82.92 % 0.2 s 1 core @ 2.5 Ghz (Python)
33 PointCSE 92.78 % 95.99 % 87.66 % 0.02 s 1 core @ 2.5 Ghz (C/C++)
34 MRF 92.74 % 95.74 % 87.64 % 0.05 s GPU @ 1.5 Ghz (Python + C/C++)
35 SegVoxelNet 92.73 % 96.00 % 87.60 % 0.04 s 1 core @ 2.5 Ghz (Python)
36 Patches
This method makes use of Velodyne laser scans.
 92.72 % 96.34 % 87.63 % 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.
37 PPFNet code 92.68 % 96.32 % 87.66 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
38 R-GCN 92.67 % 96.19 % 87.66 % 0.16 s GPU @ 2.5 Ghz (Python)
39 PI-RCNN 92.66 % 96.17 % 87.68 % 0.1 s 1 core @ 2.5 Ghz (Python)
40 OHS-Direct 92.65 % 96.09 % 89.72 % 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.
41 NU-optim 92.63 % 95.67 % 87.37 % 0.04 s GPU @ >3.5 Ghz (Python)
42 3D-CVF
This method makes use of Velodyne laser scans.
 92.60 % 96.20 % 89.60 % 0.05 s GPU @ >3.5 Ghz (Python)
43 deprecated 92.59 % 96.21 % 89.58 % 0.05 s GPU @ >3.5 Ghz (Python)
44 PointPainting
This method makes use of Velodyne laser scans.
 92.58 % 98.39 % 89.71 % 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.
45 SPA 92.56 % 95.96 % 87.60 % 0.1 s 1 core @ 2.5 Ghz (Python)
46 DEFT 92.55 % 96.17 % 89.51 % 1 s GPU @ 2.5 Ghz (Python)
47 Associate-3Ddet
This method makes use of Velodyne laser scans.
 92.45 % 95.61 % 87.32 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
48 CP
This method makes use of Velodyne laser scans.
 92.44 % 96.14 % 87.58 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
49 YOLOv3.5 92.42 % 95.22 % 82.32 % 0.05 s GPU @ 2.5 Ghz (Python)
50 OHS-Dense 92.39 % 95.84 % 89.51 % 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.
51 PointRGCN 92.33 % 97.51 % 87.07 % 0.26 s GPU @ V100 (Python)
52 F-ConvNet
This method makes use of Velodyne laser scans.
 code 92.19 % 95.85 % 80.09 % 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.
53 IE-PointRCNN 92.08 % 96.01 % 87.05 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
54 SDP+RPN 92.03 % 95.16 % 79.16 % 0.4 s GPU @ 2.5 Ghz (Python + C/C++)
F. Yang, W. Choi and Y. Lin: Exploit All the Layers: Fast and Accurate CNN Object Detector with Scale Dependent Pooling and Cascaded Rejection Classifiers. Proceedings of the IEEE International Conference on Computer Vision and Pattern Recognition 2016.
S. Ren, K. He, R. Girshick and J. Sun: Faster R-CNN: Towards real-time object detection with region proposal networks. Advances in Neural Information Processing Systems 2015.
55 AB3DMOT
This method makes use of Velodyne laser scans.
This is an online method (no batch processing).
 code 92.00 % 95.88 % 86.98 % 0.0047s 1 core @ 2.5 Ghz (Python)
X. Weng and K. Kitani: A Baseline for 3D Multi-Object Tracking. arXiv:1907.03961 2019.
56 MMLab-PointRCNN
This method makes use of Velodyne laser scans.
 code 91.90 % 95.92 % 87.11 % 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.
57 MMLab-PartA^2
This method makes use of Velodyne laser scans.
 91.86 % 95.03 % 89.06 % 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.
58 MBR-SSD 91.83 % 93.46 % 84.97 % 4.0 s GPU @ 2.5 Ghz (Python + C/C++)
59 epBRM
This method makes use of Velodyne laser scans.
 code 91.77 % 94.59 % 88.45 % 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.
60 MLF_SecCas 91.76 % 96.53 % 83.90 % 0.05 s 1 core @ 2.5 Ghz (Python)
61 ITVD code 91.73 % 95.85 % 79.31 % 0.3 s GPU @ 2.5 Ghz (C/C++)
Y. Wei Liu: Improving Tiny Vehicle Detection in Complex Scenes. IEEE International Conference on Multimedia and Expo (ICME) 2018.
62 HRI-FusionRCNN 91.70 % 94.61 % 84.10 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
63 PiP 91.67 % 94.35 % 88.35 % 0.05 s 1 core @ 2.5 Ghz (Python)
64 SINet+ code 91.67 % 94.17 % 78.60 % 0.3 s TITAN X GPU
X. Hu, X. Xu, Y. Xiao, H. Chen, S. He, J. Qin and P. Heng: SINet: A Scale-insensitive Convolutional Neural Network for Fast Vehicle Detection. IEEE Transactions on Intelligent Transportation Systems 2019.
65 Faster RCNN + A 91.60 % 94.77 % 81.43 % 0.19 s GPU @ 2.5 Ghz (Python + C/C++)
66 Cascade MS-CNN code 91.60 % 94.26 % 78.84 % 0.25 s GPU @ 2.5 Ghz (C/C++)
Z. Cai and N. Vasconcelos: Cascade R-CNN: High Quality Object Detection and Instance Segmentation. arXiv preprint arXiv:1906.09756 2019.
Z. Cai, Q. Fan, R. Feris and N. Vasconcelos: A unified multi-scale deep convolutional neural network for fast object detection. European conference on computer vision 2016.
67 deprecated 91.59 % 94.34 % 79.14 % 0.05 s GPU @ 2.0 Ghz (Python)
68 Det-RGBD
This method uses stereo information.
 91.49 % 94.30 % 79.41 % 0.58 s GPU @ 2.5 Ghz (Python + C/C++)
69 HRI-VoxelFPN 91.44 % 96.65 % 86.18 % 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.
70 TBA 91.43 % 93.99 % 88.51 % 0.07 s 1 core @ 2.5 Ghz (Python)
71 RUC 91.40 % 95.02 % 88.41 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
72 Faster RCNN + G 91.28 % 94.34 % 81.02 % 1.1 s GPU @ 2.5 Ghz (Python)
73 Faster RCNN + Gr + A 91.25 % 94.09 % 81.25 % 1.29 s GPU @ 2.5 Ghz (Python)
74 Alibaba-AILabsX
This method makes use of Velodyne laser scans.
 91.23 % 96.33 % 83.75 % 0.05 s 1 core @ >3.5 Ghz (C/C++)
75 OACV 91.21 % 94.23 % 83.07 % 0.23 s GPU @ 2.5 Ghz (Python)
76 CentrNet-v1
This method makes use of Velodyne laser scans.
 91.21 % 94.22 % 88.36 % 0.03 s GPU @ 2.5 Ghz (Python)
77 PointPillars
This method makes use of Velodyne laser scans.
 code 91.19 % 94.00 % 88.17 % 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.
78 Faster RCNN + A 91.19 % 94.43 % 80.99 % 0.19 s GPU @ 2.5 Ghz (Python)
79 LTN 91.18 % 94.68 % 81.51 % 0.4 s GPU @ >3.5 Ghz (Python)
T. Wang, X. He, Y. Cai and G. Xiao: Learning a Layout Transfer Network for Context Aware Object Detection. IEEE Transactions on Intelligent Transportation Systems 2019.
80 PointPiallars_SECA 91.12 % 93.66 % 87.94 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
81 DDB
This method makes use of Velodyne laser scans.
 91.12 % 93.71 % 87.34 % 0.05 s GPU @ 2.5 Ghz (Python)
82 Aston-EAS 91.02 % 93.91 % 77.93 % 0.24 s GPU @ 2.5 Ghz (Python + C/C++)
J. Wei, J. He, Y. Zhou, K. Chen, Z. Tang and Z. Xiong: Enhanced Object Detection With Deep Convolutional Neural Networks for Advanced Driving Assistance. IEEE Transactions on Intelligent Transportation Systems 2019.
83 ARPNET 90.99 % 94.00 % 83.49 % 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.
84 PCSC-Net 90.97 % 94.20 % 87.38 % 0.04 s 1 core @ 2.5 Ghz (C/C++)
85 MMV 90.91 % 94.16 % 83.36 % 0.4 s GPU @ 2.5 Ghz (C/C++)
86 JSU-NET 90.90 % 96.41 % 80.67 % 0.1 s 1 core @ 2.5 Ghz (Python)
87 GAFM 90.90 % 96.46 % 80.70 % 0.5 s 1 core @ 2.5 Ghz (Python)
88 GA_BALANCE 90.86 % 96.19 % 78.40 % 1 s 1 core @ 2.5 Ghz (Python)
89 A-VoxelNet 90.86 % 93.84 % 83.27 % 0.029 s GPU @ 2.5 Ghz (Python)
90 MV3D
This method makes use of Velodyne laser scans.
 90.83 % 96.47 % 78.63 % 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.
91 MVSLN 90.81 % 96.12 % 83.39 % 0.1s s 1 core @ 2.5 Ghz (C/C++)
92 MPNet
This method makes use of Velodyne laser scans.
 90.80 % 94.68 % 87.30 % 0.02 s GPU @ 2.5 Ghz (Python + C/C++)
93 3D IoU Loss
This method makes use of Velodyne laser scans.
 90.79 % 95.92 % 85.65 % 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 SINet_VGG code 90.79 % 93.59 % 77.53 % 0.2 s TITAN X GPU
X. Hu, X. Xu, Y. Xiao, H. Chen, S. He, J. Qin and P. Heng: SINet: A Scale-insensitive Convolutional Neural Network for Fast Vehicle Detection. IEEE Transactions on Intelligent Transportation Systems 2019.
95 Tencent_ADlab_Lidar
This method makes use of Velodyne laser scans.
 90.74 % 93.80 % 86.75 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
96 GA_FULLDATA 90.73 % 96.31 % 78.22 % 1 s 4 cores @ 2.5 Ghz (Python)
97 SRF 90.69 % 95.88 % 85.52 % 0.05 s GPU @ 2.5 Ghz (Python + C/C++)
98 HR-SECOND code 90.68 % 93.72 % 85.63 % 0.11 s 1 core @ 2.5 Ghz (Python + C/C++)
99 GA2500 90.68 % 95.86 % 80.29 % 0.2 s 1 core @ 2.5 Ghz (Python)
100 GA_rpn500 90.68 % 95.86 % 80.29 % 1 s 1 core @ 2.5 Ghz (Python)
101 TANet code 90.67 % 93.67 % 85.31 % 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.
102 SFB-SECOND 90.67 % 96.17 % 85.43 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
103 SECOND-V1.5
This method makes use of Velodyne laser scans.
 code 90.65 % 95.96 % 85.35 % 0.04 s GPU @ 2.0 Ghz (Python + C/C++)
104 PTS
This method makes use of Velodyne laser scans.
 code 90.64 % 95.74 % 85.41 % 0.01 s 1 core @ 2.5 Ghz (C/C++)
105 VOXEL_FPN_HR 90.55 % 93.76 % 85.42 % 0.12 s 8 cores @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
106 FOFNet
This method makes use of Velodyne laser scans.
 90.52 % 94.00 % 85.20 % 0.04 s GPU @ 2.5 Ghz (Python)
107 MP 90.50 % 93.86 % 85.17 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
108 Sogo_MM 90.46 % 94.31 % 80.62 % 1.5 s GPU @ 2.5 Ghz (C/C++)
109 bigger_ga 90.38 % 95.76 % 77.92 % 1 s 1 core @ 2.5 Ghz (Python)
110 AtrousDet 90.35 % 95.94 % 77.94 % 0.05 s TITAN X
111 SCNet
This method makes use of Velodyne laser scans.
 90.30 % 95.59 % 85.09 % 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.
112 RUC code 90.24 % 92.60 % 86.55 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
113 Deep3DBox 90.19 % 94.71 % 76.82 % 1.5 s GPU @ 2.5 Ghz (C/C++)
A. Mousavian, D. Anguelov, J. Flynn and J. Kosecka: 3D Bounding Box Estimation Using Deep Learning and Geometry. CVPR 2017.
114 FQNet 90.17 % 94.72 % 76.78 % 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.
115 BVVF 90.15 % 95.65 % 84.95 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
116 SAANet 90.14 % 95.93 % 82.95 % 0.10 s 1 core @ 2.5 Ghz (Python)
117 DeepStereoOP 90.06 % 95.15 % 79.91 % 3.4 s GPU @ 3.5 Ghz (Matlab + C/C++)
C. Pham and J. Jeon: Robust Object Proposals Re-ranking for Object Detection in Autonomous Driving Using Convolutional Neural Networks. Signal Processing: Image Communiation 2017.
118 SubCNN 89.98 % 94.26 % 79.78 % 2 s GPU @ 3.5 Ghz (Python + C/C++)
Y. Xiang, W. Choi, Y. Lin and S. Savarese: Subcategory-aware Convolutional Neural Networks for Object Proposals and Detection. IEEE Winter Conference on Applications of Computer Vision (WACV) 2017.
119 MLOD
This method makes use of Velodyne laser scans.
 code 89.97 % 94.88 % 84.98 % 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.
120 GPP code 89.96 % 94.02 % 81.13 % 0.23 s GPU @ 1.5 Ghz (Python + C/C++)
A. Rangesh and M. Trivedi: Ground plane polling for 6dof pose estimation of objects on the road. arXiv preprint arXiv:1811.06666 2018.
121 RUC code 89.93 % 93.12 % 85.44 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
122 ZRNet(ResNet-50) 89.92 % 95.24 % 79.69 % 0.04 s GPU @ 2.5 Ghz (Python)
123 AVOD
This method makes use of Velodyne laser scans.
 code 89.88 % 95.17 % 82.83 % 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.
124 SINet_PVA code 89.86 % 92.72 % 76.47 % 0.11 s TITAN X GPU
X. Hu, X. Xu, Y. Xiao, H. Chen, S. He, J. Qin and P. Heng: SINet: A Scale-insensitive Convolutional Neural Network for Fast Vehicle Detection. IEEE Transactions on Intelligent Transportation Systems 2019.
125 ZRNet 89.72 % 93.97 % 79.47 % 0.04 s GPU @ 2.5 Ghz (Python)
126 3DOP
This method uses stereo information.
 code 89.55 % 92.96 % 79.38 % 3s GPU @ 2.5 Ghz (Matlab + C/C++)
X. Chen, K. Kundu, Y. Zhu, A. Berneshawi, H. Ma, S. Fidler and R. Urtasun: 3D Object Proposals for Accurate Object Class Detection. NIPS 2015.
127 PAD 89.49 % 93.43 % 85.85 % 0.15 s 1 core @ 2.5 Ghz (Python)
128 Mono3D code 89.37 % 94.52 % 79.15 % 4.2 s GPU @ 2.5 Ghz (Matlab + C/C++)
X. Chen, K. Kundu, Z. Zhang, H. Ma, S. Fidler and R. Urtasun: Monocular 3D Object Detection for Autonomous Driving. CVPR 2016.
129 4D-MSCNN+CRL
This method uses stereo information.
 89.37 % 92.40 % 77.00 % 0.2 s GPU @ 2.5 Ghz (Matlab + C/C++)
130 MonoDIS 89.15 % 94.61 % 78.37 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
131 cas+res+soft 89.14 % 94.54 % 78.37 % 0.2 s 4 cores @ 2.5 Ghz (Python)
132 merge12-12 88.96 % 94.58 % 78.22 % 0.2 s 4 cores @ 2.5 Ghz (Python)
133 AVOD-FPN
This method makes use of Velodyne laser scans.
 code 88.92 % 94.70 % 84.13 % 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.
134 AM3D 88.71 % 92.55 % 77.78 % 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.
135 SS3D_HW 88.68 % 94.49 % 68.79 % 0.4 s GPU @ 2.5 Ghz (Python)
136 MS-CNN code 88.68 % 93.87 % 76.11 % 0.4 s GPU @ 2.5 Ghz (C/C++)
Z. Cai, Q. Fan, R. Feris and N. Vasconcelos: A Unified Multi-scale Deep Convolutional Neural Network for Fast Object Detection. ECCV 2016.
137 CRCNNA 88.59 % 94.82 % 76.74 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
138 3DNN 88.56 % 94.52 % 81.51 % 0.09 s GPU @ 2.5 Ghz (Python)
139 CSFADet 88.54 % 93.75 % 78.62 % 0.05 s GPU @ 2.5 Ghz (Python)
140 MonoPSR code 88.50 % 93.63 % 73.36 % 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.
141 Shift R-CNN (mono) code 88.48 % 94.07 % 78.34 % 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.
142 MM-MRFC
This method uses optical flow information.
This method makes use of Velodyne laser scans.
 88.46 % 95.54 % 78.14 % 0.05 s GPU @ 2.5 Ghz (C/C++)
A. Costea, R. Varga and S. Nedevschi: Fast Boosting based Detection using Scale Invariant Multimodal Multiresolution Filtered Features. CVPR 2017.
143 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).
 88.41 % 95.38 % 84.22 % 0.013 s 1 core @ 2.5 Ghz (Python + C/C++)
144 FCY
This method makes use of Velodyne laser scans.
 88.41 % 93.72 % 83.28 % 0.02 s GPU @ 2.5 Ghz (Python)
145 TridentNet 88.37 % 90.33 % 80.57 % 0.2 s GPU @ 2.5 Ghz (Python)
146 PP-3D 88.35 % 93.71 % 80.84 % 0.1 s 1 core @ 2.5 Ghz (Python)
147 3DBN
This method makes use of Velodyne laser scans.
 88.29 % 93.74 % 80.74 % 0.13s 1080Ti (Python+C/C++)
X. Li, J. Guivant, N. Kwok and Y. Xu: 3D Backbone Network for 3D Object Detection. CoRR 2019.
148 NLK 87.89 % 91.65 % 83.32 % 0.02 s 1 core @ 2.5 Ghz (Python)
149 Multi-3D
This method makes use of Velodyne laser scans.
 87.87 % 93.70 % 76.07 % 0.15 s 1 core @ 2.5 Ghz (C/C++)
150 ga50 87.65 % 95.76 % 75.14 % 1 s 1 core @ 2.5 Ghz (Python)
151 cas_retina 87.64 % 93.87 % 75.30 % 0.2 s 4 cores @ 2.5 Ghz (Python)
152 SMOKE 87.51 % 93.21 % 77.66 % 0.03 s GPU @ 2.5 Ghz (Python)
153 MonoSS 87.46 % 93.15 % 77.58 % 0.03 s GPU @ 2.5 Ghz (Python + C/C++)
154 cascadercnn 87.36 % 89.37 % 73.42 % 0.36 s 4 cores @ 2.5 Ghz (Python)
155 SCANet 87.28 % 92.91 % 81.99 % 0.17 s >8 cores @ 2.5 Ghz (Python)
156 RTM3D code 86.92 % 91.85 % 77.41 % 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.
157 anm 86.52 % 94.88 % 76.46 % 3 s 1 core @ 2.5 Ghz (C/C++)
158 DSGN
This method uses stereo information.
 code 86.43 % 95.53 % 78.75 % 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.
159 ReSqueeze 86.12 % 90.35 % 76.53 % 0.03 s GPU @ >3.5 Ghz (Python)
160 Stereo R-CNN
This method uses stereo information.
 code 85.98 % 93.98 % 71.25 % 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.
161 StereoFENet
This method uses stereo information.
 85.70 % 91.48 % 77.62 % 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.
162 ResNet-RRC w/RGBD 85.58 % 91.32 % 74.80 % 0.057 s GPU @ 1.5 Ghz (Python + C/C++)
163 cas_retina_1_13 85.48 % 91.54 % 74.60 % 0.03 s 4 cores @ 2.5 Ghz (Python)
164 NEUAV 85.42 % 89.67 % 77.28 % 0.06 s GPU @ 2.5 Ghz (Python)
165 ResNet-RRC 85.33 % 91.45 % 74.27 % 0.06 s GPU @ 1.5 Ghz (Python + C/C++)
H. Jeon and . others: High-Speed Car Detection Using ResNet- Based Recurrent Rolling Convolution. Proceedings of the IEEE conference on systems, man, and cybernetics 2018.
166 Cmerge 85.32 % 93.40 % 70.57 % 0.2 s 4 cores @ 2.5 Ghz (Python)
167 PL++ (SDN+GDC)
This method uses stereo information.
This method makes use of Velodyne laser scans.
 85.15 % 94.95 % 77.78 % 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.
168 RAR-Net 85.08 % 89.04 % 69.26 % 0.5 s 1 core @ 2.5 Ghz (C/C++)
169 M3D-RPN code 85.08 % 89.04 % 69.26 % 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 .
170 SDP+CRC (ft) 85.00 % 92.06 % 71.71 % 0.6 s GPU @ 2.5 Ghz (C/C++)
F. Yang, W. Choi and Y. Lin: Exploit All the Layers: Fast and Accurate CNN Object Detector with Scale Dependent Pooling and Cascaded Rejection Classifiers. Proceedings of the IEEE International Conference on Computer Vision and Pattern Recognition 2016.
171 SS3D 84.92 % 92.72 % 70.35 % 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.
172 LPN 84.77 % 89.19 % 74.08 % 0.2 s GPU @ 2.5 Ghz (Python + C/C++)
173 MonoFENet 84.63 % 91.68 % 76.71 % 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.
174 SECA 84.60 % 92.51 % 79.53 % 1 s GPU @ 2.5 Ghz (Python)
175 PG-MonoNet 84.42 % 88.61 % 68.59 % 0.19 s GPU @ 2.5 Ghz (Python)
176 MV3D (LIDAR)
This method makes use of Velodyne laser scans.
 84.39 % 93.08 % 79.27 % 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.
177 Complexer-YOLO
This method makes use of Velodyne laser scans.
 84.16 % 91.92 % 79.62 % 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.
178 ZoomNet
This method uses stereo information.
 code 83.92 % 94.22 % 69.00 % 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.
179 D4LCN code 83.67 % 90.34 % 65.33 % 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.
180 ASOD 83.52 % 94.09 % 68.68 % 0.28 s GPU @ 2.5 Ghz (Python)
181 softretina 83.30 % 93.55 % 70.59 % 0.16 s 4 cores @ 2.5 Ghz (Python)
182 Faster R-CNN code 83.16 % 88.97 % 72.62 % 2 s GPU @ 3.5 Ghz (Python + C/C++)
S. Ren, K. He, R. Girshick and J. Sun: Faster R-CNN: Towards Real- Time Object Detection with Region Proposal Networks. NIPS 2015.
183 ZKNet 82.96 % 92.17 % 72.43 % 0.01 s GPU @ 2.0 Ghz (Python)
184 Pseudo-LiDAR++
This method uses stereo information.
 code 82.90 % 94.46 % 75.45 % 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.
185 Retinanet100 82.73 % 93.97 % 68.37 % 0.2 s 4 cores @ 2.5 Ghz (Python)
186 BS3D 82.72 % 95.35 % 70.01 % 22 ms Titan Xp
N. Gählert, J. Wan, M. Weber, J. Zöllner, U. Franke and J. Denzler: Beyond Bounding Boxes: Using Bounding Shapes for Real-Time 3D Vehicle Detection from Monocular RGB Images. 2019 IEEE Intelligent Vehicles Symposium (IV) 2019.
187 Pseudo-LiDAR E2E
This method uses stereo information.
 82.54 % 94.00 % 75.31 % 0.4 s GPU @ 2.5 Ghz (Python)
188 Disp R-CNN
This method uses stereo information.
 82.47 % 93.15 % 70.35 % 0.42 s GPU @ 2.5 Ghz (Python + C/C++)
189 Disp R-CNN (velo)
This method uses stereo information.
 82.40 % 93.11 % 70.26 % 0.42 s GPU @ 2.5 Ghz (Python + C/C++)
190 cascade_gw 82.35 % 85.98 % 71.60 % 0.2 s 4 cores @ 2.5 Ghz (Python)
191 FRCNN+Or code 82.00 % 92.91 % 68.79 % 0.09 s Titan Xp GPU
C. Guindel, D. Martin and J. Armingol: Fast Joint Object Detection and Viewpoint Estimation for Traffic Scene Understanding. IEEE Intelligent Transportation Systems Magazine 2018.
C. Guindel, D. Martin and J. Armingol: Joint Object Detection and Viewpoint Estimation using CNN features. IEEE International Conference on Vehicular Electronics and Safety (ICVES) 2017.
192 CBNet 81.70 % 91.47 % 72.02 % 1 s 4 cores @ 2.5 Ghz (Python)
193 Resnet101Faster rcnn 81.44 % 91.08 % 71.52 % 1 s 1 core @ 2.5 Ghz (Python)
194 A3DODWTDA (image) code 81.25 % 78.96 % 70.56 % 0.8 s GPU @ 3.0 Ghz (Python)
F. Gustafsson and E. Linder-Norén: Automotive 3D Object Detection Without Target Domain Annotations. 2018.
195 RefineNet 81.01 % 91.91 % 65.67 % 0.20 s GPU @ 2.5 Ghz (Matlab + C++)
R. Rajaram, E. Bar and M. Trivedi: RefineNet: Refining Object Detectors for Autonomous Driving. IEEE Transactions on Intelligent Vehicles 2016.
R. Rajaram, E. Bar and M. Trivedi: RefineNet: Iterative Refinement for Accurate Object Localization. Intelligent Transportation Systems Conference 2016.
196 MTDP 80.97 % 89.03 % 66.91 % 0.15 s GPU @ 2.0 Ghz (Python)
197 RFCN_RFB 80.89 % 88.07 % 69.66 % 0.2 s 4 cores @ 2.5 Ghz (Python)
198 Manhnet 80.85 % 89.06 % 64.29 % 26 ms 1 core @ 2.5 Ghz (C/C++)
199 centernet 80.78 % 90.29 % 70.53 % 0.01 s GPU @ 2.5 Ghz (Python)
200 3D GCK 80.19 % 89.55 % 68.08 % 24 ms Tesla V100
201 RADNet-Fusion
This method makes use of Velodyne laser scans.
 80.04 % 76.72 % 76.78 % 0.1 s 1 core @ 2.5 Ghz (Python)
202 NM code 79.98 % 90.71 % 68.98 % 0.01 s GPU @ 2.5 Ghz (Python)
203 RADNet-LIDAR
This method makes use of Velodyne laser scans.
 79.59 % 75.20 % 76.03 % 0.1 s 1 core @ 2.5 Ghz (Python)
204 MMRetina
This method uses stereo information.
This method uses optical flow information.
This method makes use of Velodyne laser scans.
 79.53 % 89.66 % 69.52 % 0.38 s GPU @ 2.5 Ghz (Python)
205 SceneNet 79.26 % 90.70 % 67.98 % 0.03 s GPU @ 2.5 Ghz (C/C++)
206 A3DODWTDA
This method makes use of Velodyne laser scans.
 code 79.15 % 82.98 % 68.30 % 0.08 s GPU @ 3.0 Ghz (Python)
F. Gustafsson and E. Linder-Norén: Automotive 3D Object Detection Without Target Domain Annotations. 2018.
207 spLBP 78.66 % 81.66 % 61.69 % 1.5 s 8 cores @ 2.5 Ghz (Matlab + C/C++)
Q. Hu, S. Paisitkriangkrai, C. Shen, A. Hengel and F. Porikli: Fast Detection of Multiple Objects in Traffic Scenes With a Common Detection Framework. IEEE Trans. Intelligent Transportation Systems 2016.
208 dgist_multiDetNet 78.26 % 93.58 % 70.04 % 0.05 s 1 core @ 2.5 Ghz (C/C++)
209 3D-SSMFCNN code 78.19 % 77.92 % 69.19 % 0.1 s GPU @ 1.5 Ghz (C/C++)
L. Novak: Vehicle Detection and Pose Estimation for Autonomous Driving. 2017.
210 MonoGRNet code 77.94 % 88.65 % 63.31 % 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.
211 yolov3_warp 77.61 % 92.24 % 65.70 % 0.5 s 1 core @ 2.5 Ghz (C/C++)
ERROR: Wrong syntax in BIBTEX file.
212 Reinspect code 77.48 % 90.27 % 66.73 % 2s 1 core @ 2.5 Ghz (C/C++)
R. Stewart, M. Andriluka and A. Ng: End-to-End People Detection in Crowded Scenes. CVPR 2016.
213 multi-task CNN 77.18 % 86.12 % 68.09 % 25.1 ms GPU @ 2.0 Ghz (Python)
M. Oeljeklaus, F. Hoffmann and T. Bertram: A Fast Multi-Task CNN for Spatial Understanding of Traffic Scenes. IEEE Intelligent Transportation Systems Conference 2018.
214 Regionlets 76.99 % 88.75 % 60.49 % 1 s >8 cores @ 2.5 Ghz (C/C++)
X. Wang, M. Yang, S. Zhu and Y. Lin: Regionlets for Generic Object Detection. T-PAMI 2015.
W. Zou, X. Wang, M. Sun and Y. Lin: Generic Object Detection with Dense Neural Patterns and Regionlets. British Machine Vision Conference 2014.
C. Long, X. Wang, G. Hua, M. Yang and Y. Lin: Accurate Object Detection with Location Relaxation and Regionlets Relocalization. Asian Conference on Computer Vision 2014.
215 3DVP code 76.98 % 84.95 % 65.78 % 40 s 8 cores @ 3.5 Ghz (Matlab + C/C++)
Y. Xiang, W. Choi, Y. Lin and S. Savarese: Data-Driven 3D Voxel Patterns for Object Category Recognition. IEEE Conference on Computer Vision and Pattern Recognition 2015.
216 FailNet-Fusion
This method makes use of Velodyne laser scans.
 76.90 % 74.55 % 71.94 % 0.1 s 1 core @ 2.5 Ghz (Python)
217 RTL3D 76.74 % 79.68 % 72.56 % 0.02 s GPU @ 2.5 Ghz (Python)
218 avodC 76.58 % 87.30 % 71.65 % 0.1 s GPU @ 2.5 Ghz (Python)
219 SubCat code 76.36 % 84.10 % 60.56 % 0.7 s 6 cores @ 3.5 Ghz (Matlab + C/C++)
E. Ohn-Bar and M. Trivedi: Learning to Detect Vehicles by Clustering Appearance Patterns. T-ITS 2015.
220 GS3D 76.35 % 86.23 % 62.67 % 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.
221 FailNet-LIDAR
This method makes use of Velodyne laser scans.
 76.26 % 74.16 % 71.24 % 0.1 s 1 core @ 2.5 Ghz (Python)
222 AOG code 76.24 % 86.08 % 61.51 % 3 s 4 cores @ 2.5 Ghz (Matlab)
T. Wu, B. Li and S. Zhu: Learning And-Or Models to Represent Context and Occlusion for Car Detection and Viewpoint Estimation. TPAMI 2016.
B. Li, T. Wu and S. Zhu: Integrating Context and Occlusion for Car Detection by Hierarchical And-Or Model. ECCV 2014.
223 bin 76.16 % 78.73 % 63.39 % 15ms s GPU @ >3.5 Ghz (Python)
224 Pose-RCNN 75.83 % 89.59 % 64.06 % 2 s >8 cores @ 2.5 Ghz (Python)
M. Braun, Q. Rao, Y. Wang and F. Flohr: Pose-RCNN: Joint object detection and pose estimation using 3D object proposals. Intelligent Transportation Systems (ITSC), 2016 IEEE 19th International Conference on 2016.
225 VoxelNet(Unofficial) 75.22 % 81.37 % 68.74 % 0.5 s GPU @ 2.0 Ghz (Python)
226 RFCN 75.14 % 83.04 % 61.55 % 0.2 s 4 cores @ 2.5 Ghz (Python)
227 myfaster-rcnn-v1.5 74.93 % 89.85 % 62.56 % 0.1 s 1 core @ 2.5 Ghz (Python)
228 3D FCN
This method makes use of Velodyne laser scans.
 74.65 % 86.74 % 67.85 % >5 s 1 core @ 2.5 Ghz (C/C++)
B. Li: 3D Fully Convolutional Network for Vehicle Detection in Point Cloud. IROS 2017.
229 OC Stereo
This method uses stereo information.
 74.60 % 87.39 % 62.56 % 0.35 s 1 core @ 2.5 Ghz (Python + C/C++)
230 yolo800 74.31 % 78.93 % 63.83 % 0.13 s 4 cores @ 2.5 Ghz (Python)
231 3DVSSD 74.11 % 86.99 % 63.57 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
232 Multi-task DG 74.07 % 91.06 % 64.48 % 0.06 s GPU @ 2.5 Ghz (Python)
233 FD2 73.93 % 88.65 % 64.62 % 0.01 s GPU @ >3.5 Ghz (Python + C/C++)
234 BdCost+DA+BB+MS 73.72 % 85.18 % 57.79 % TBD s 4 cores @ 2.5 Ghz (Matlab + C/C++)
235 m-prcnn
This method uses stereo information.
 73.64 % 87.64 % 57.03 % 0.43 s 1 core @ 2.5 Ghz (Python)
236 BdCost+DA+MS 73.62 % 85.03 % 58.94 % TBD s 4 cores @ 2.5 Ghz (Matlab/C++)
237 Int-YOLO code 73.23 % 75.81 % 63.59 % 0.03 s 1 core @ 2.5 Ghz (C/C++)
ERROR: Wrong syntax in BIBTEX file.
238 stereo_sa
This method uses stereo information.
 72.99 % 87.88 % 63.49 % 0.3 s GPU @ 2.5 Ghz (Python)
239 RuiRUC 72.08 % 87.48 % 55.28 % 0.12 s 1 core @ 2.5 Ghz (Python)
240 ANM 71.97 % 87.17 % 55.19 % 0.12 s 1 core @ 2.5 Ghz (Python)
241 RFBnet 71.66 % 87.25 % 63.00 % 0.2 s 4 cores @ 2.5 Ghz (Python)
242 AOG-View 71.26 % 85.01 % 55.73 % 3 s 1 core @ 2.5 Ghz (Matlab, C/C++)
B. Li, T. Wu and S. Zhu: Integrating Context and Occlusion for Car Detection by Hierarchical And-Or Model. ECCV 2014.
243 GPVL 71.06 % 81.67 % 54.96 % 10 s 1 core @ 2.5 Ghz (C/C++)
244 BdCost+DA+BB 70.86 % 85.52 % 56.19 % TBD s 4 cores @ 2.5 Ghz (Matlab + C/C++)
245 MV-RGBD-RF
This method makes use of Velodyne laser scans.
 70.70 % 77.89 % 57.41 % 4 s 4 cores @ 2.5 Ghz (C/C++)
A. Gonzalez, D. Vazquez, A. Lopez and J. Amores: On-Board Object Detection: Multicue, Multimodal, and Multiview Random Forest of Local Experts.. IEEE Trans. on Cybernetics 2016.
A. Gonzalez, G. Villalonga, J. Xu, D. Vazquez, J. Amores and A. Lopez: Multiview Random Forest of Local Experts Combining RGB and LIDAR data for Pedestrian Detection. IEEE Intelligent Vehicles Symposium (IV) 2015.
246 Vote3Deep
This method makes use of Velodyne laser scans.
 70.30 % 78.95 % 63.12 % 1.5 s 4 cores @ 2.5 Ghz (C/C++)
M. Engelcke, D. Rao, D. Zeng Wang, C. Hay Tong and I. Posner: Vote3Deep: Fast Object Detection in 3D Point Clouds Using Efficient Convolutional Neural Networks. ArXiv e-prints 2016.
247 ROI-10D 70.16 % 76.56 % 61.15 % 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.
248 fasterrcnn 69.45 % 74.76 % 60.20 % 0.2 s 4 cores @ 2.5 Ghz (Python)
249 myfaster-rcnn 68.38 % 90.54 % 55.97 % 0.01 s 1 core @ 2.5 Ghz (Python)
250 Decoupled-3D v2 68.17 % 88.64 % 54.74 % 0.08 s GPU @ 2.5 Ghz (C/C++)
251 Decoupled-3D 67.92 % 87.78 % 54.53 % 0.08 s GPU @ 2.5 Ghz (C/C++)
252 OC-DPM 67.06 % 79.07 % 52.61 % 10 s 8 cores @ 2.5 Ghz (Matlab)
B. Pepik, M. Stark, P. Gehler and B. Schiele: Occlusion Patterns for Object Class Detection. IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2013.
253 mymask-rcnn 66.82 % 88.60 % 52.18 % 0.3 s 1 core @ 2.5 Ghz (Python)
254 Fast-SSD 66.79 % 85.19 % 57.89 % 0.06 s GTX650Ti
255 DPM-VOC+VP 66.72 % 82.15 % 49.01 % 8 s 1 core @ 2.5 Ghz (C/C++)
B. Pepik, M. Stark, P. Gehler and B. Schiele: Multi-view and 3D Deformable Part Models. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015.
256 BdCost48LDCF code 66.63 % 81.38 % 52.20 % 0.5 s 8 cores @ 3.5 Ghz (Matlab + C/C++)
A. Fernández-Baldera, J. Buenaposada and L. Baumela: BAdaCost: Multi-class Boosting with Costs . Pattern Recognition 2018.
257 E-VoxelNet 65.33 % 68.00 % 57.84 % 0.1 s GPU @ 2.5 Ghz (Python)
258 RefinedMPL 65.24 % 88.29 % 53.20 % 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.
259 BdCost48-25C 64.63 % 81.42 % 52.22 % 4 s 1 core @ 2.5 Ghz (C/C++)
260 MDPM-un-BB 64.06 % 79.74 % 49.07 % 60 s 4 core @ 2.5 Ghz (MATLAB)
P. Felzenszwalb, R. Girshick, D. McAllester and D. Ramanan: Object Detection with Discriminatively Trained Part-Based Models. PAMI 2010.
261 PDV-Subcat 63.24 % 78.27 % 47.67 % 7 s 1 core @ 2.5 Ghz (C/C++)
J. Shen, X. Zuo, J. Li, W. Yang and H. Ling: A novel pixel neighborhood differential statistic feature for pedestrian and face detection . Pattern Recognition 2017.
262 MODet
This method makes use of Velodyne laser scans.
 62.54 % 66.06 % 60.04 % 0.05 s GTX1080Ti
263 yl_net 61.78 % 66.00 % 60.36 % 0.03 s GPU @ 2.5 Ghz (Python)
264 Lidar_ROI+Yolo(UJS) 61.71 % 73.32 % 53.65 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
265 GNN 61.48 % 79.09 % 51.06 % 0.2 s 1 core @ 2.5 Ghz (Python)
266 SubCat48LDCF code 61.16 % 78.86 % 44.69 % 0.5 s 8 cores @ 3.5 Ghz (Matlab + C/C++)
A. Fernández-Baldera, J. Buenaposada and L. Baumela: BAdaCost: Multi-class Boosting with Costs . Pattern Recognition 2018.
267 DPM-C8B1
This method uses stereo information.
 60.21 % 75.24 % 44.73 % 15 s 4 cores @ 2.5 Ghz (Matlab + C/C++)
J. Yebes, L. Bergasa and M. García-Garrido: Visual Object Recognition with 3D-Aware Features in KITTI Urban Scenes. Sensors 2015.
J. Yebes, L. Bergasa, R. Arroyo and A. Lázaro: Supervised learning and evaluation of KITTI's cars detector with DPM. IV 2014.
268 tiny_rfdet code 59.94 % 65.51 % 57.20 % 0.01 s GPU @ 2.5 Ghz (Python)
269 RADNet-Mono 59.85 % 67.47 % 54.14 % 0.1 s 1 core @ 2.5 Ghz (Python)
270 monoref3d 58.97 % 78.11 % 47.72 % 0.1 s 1 core @ 2.5 Ghz (Python)
271 ref3D 58.97 % 78.11 % 47.72 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
272 100Frcnn 58.92 % 82.09 % 49.04 % 2 s 4 cores @ 2.5 Ghz (Python + C/C++)
273 SAMME48LDCF code 58.38 % 77.47 % 44.43 % 0.5 s 8 cores @ 3.5 Ghz (Matlab + C/C++)
A. Fernández-Baldera, J. Buenaposada and L. Baumela: BAdaCost: Multi-class Boosting with Costs . Pattern Recognition 2018.
274 LSVM-MDPM-sv 58.36 % 71.11 % 43.22 % 10 s 4 cores @ 3.0 Ghz (C/C++)
P. Felzenszwalb, R. Girshick, D. McAllester and D. Ramanan: Object Detection with Discriminatively Trained Part-Based Models. PAMI 2010.
A. Geiger, C. Wojek and R. Urtasun: Joint 3D Estimation of Objects and Scene Layout. NIPS 2011.
275 ref3D 57.16 % 77.96 % 45.99 % 0.1 s 1 core @ 2.5 Ghz (Python)
276 BirdNet
This method makes use of Velodyne laser scans.
 57.02 % 78.91 % 55.08 % 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.
277 ACF-SC 56.60 % 69.90 % 43.61 % <0.3 s 1 core @ >3.5 Ghz (Matlab + C/C++)
C. Cadena, A. Dick and I. Reid: A Fast, Modular Scene Understanding System using Context-Aware Object Detection. Robotics and Automation (ICRA), 2015 IEEE International Conference on 2015.
278 LSVM-MDPM-us code 55.95 % 68.94 % 41.45 % 10 s 4 cores @ 3.0 Ghz (C/C++)
P. Felzenszwalb, R. Girshick, D. McAllester and D. Ramanan: Object Detection with Discriminatively Trained Part-Based Models. PAMI 2010.
279 mylsi-faster-rcnn 55.81 % 80.45 % 47.38 % 0.3 s 1 core @ 2.5 Ghz (Python)
280 ACF 54.09 % 63.05 % 41.81 % 0.2 s 1 core @ >3.5 Ghz (Matlab + C/C++)
P. Doll\'ar, R. Appel, S. Belongie and P. Perona: Fast Feature Pyramids for Object Detection. PAMI 2014.
P. Doll\'ar: Piotr's Image and Video Matlab Toolbox (PMT). .
281 Mono3D_PLiDAR code 53.36 % 80.85 % 44.80 % 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.
282 VeloFCN
This method makes use of Velodyne laser scans.
 51.82 % 70.53 % 45.70 % 1 s GPU @ 2.5 Ghz (Python + C/C++)
B. Li, T. Zhang and T. Xia: Vehicle Detection from 3D Lidar Using Fully Convolutional Network. RSS 2016 .
283 FailNet-Mono 47.95 % 59.59 % 41.33 % 0.1 s 1 core @ 2.5 Ghz (Python)
284 softyolo 45.97 % 66.08 % 38.02 % 0.16 s 4 cores @ 2.5 Ghz (Python)
285 Vote3D
This method makes use of Velodyne laser scans.
 45.94 % 54.38 % 40.48 % 0.5 s 4 cores @ 2.8 Ghz (C/C++)
D. Wang and I. Posner: Voting for Voting in Online Point Cloud Object Detection. Proceedings of Robotics: Science and Systems 2015.
286 TopNet-HighRes
This method makes use of Velodyne laser scans.
 45.85 % 58.04 % 41.11 % 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.
287 RT3DStereo
This method uses stereo information.
 45.81 % 56.53 % 37.63 % 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.
288 Multimodal Detection
This method makes use of Velodyne laser scans.
 code 45.46 % 63.91 % 37.25 % 0.06 s GPU @ 3.5 Ghz (Matlab + C/C++)
A. Asvadi, L. Garrote, C. Premebida, P. Peixoto and U. Nunes: Multimodal vehicle detection: fusing 3D- LIDAR and color camera data. Pattern Recognition Letters 2017.
289 RT3D
This method makes use of Velodyne laser scans.
 39.69 % 50.33 % 40.04 % 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.
290 VoxelJones code 36.31 % 43.89 % 34.16 % .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.
291 Licar
This method makes use of Velodyne laser scans.
 35.19 % 42.34 % 33.97 % 0.09 s GPU @ 2.0 Ghz (Python)
292 KD53-20 34.76 % 51.76 % 29.39 % 0.19 s 4 cores @ 2.5 Ghz (Python)
293 SAIC-SA-3D
This method makes use of Velodyne laser scans.
 31.16 % 41.51 % 29.83 % 0.05 s GPU @ 2.5 Ghz (Python)
294 FCN-Depth code 25.05 % 52.32 % 18.07 % 1 s GPU @ 1.5 Ghz (Matlab + C/C++)
295 CSoR
This method makes use of Velodyne laser scans.
 code 21.66 % 31.52 % 17.99 % 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.
296 mBoW
This method makes use of Velodyne laser scans.
 21.59 % 35.22 % 16.89 % 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.
297 R-CNN_VGG 21.36 % 29.38 % 16.61 % 10 s GPU @ 2.5 Ghz (Matlab + C/C++)
298 DepthCN
This method makes use of Velodyne laser scans.
 code 21.18 % 37.45 % 16.08 % 2.3 s GPU @ 3.5 Ghz (Matlab)
A. Asvadi, L. Garrote, C. Premebida, P. Peixoto and U. Nunes: DepthCN: vehicle detection using 3D- LIDAR and convnet. IEEE ITSC 2017.
299 YOLOv2 code 14.31 % 26.74 % 10.94 % 0.02 s GPU @ 3.5 Ghz (C/C++)
J. Redmon, S. Divvala, R. Girshick and A. Farhadi: You only look once: Unified, real-time object detection. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition 2016.
J. Redmon and A. Farhadi: YOLO9000: Better, Faster, Stronger. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition 2017.
300 TopNet-UncEst
This method makes use of Velodyne laser scans.
 6.24 % 7.24 % 5.42 % 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.
301 TopNet-Retina
This method makes use of Velodyne laser scans.
 5.00 % 6.82 % 4.52 % 52ms GeForce 1080Ti (tensorflow-gpu, v1.12)
S. Wirges, T. Fischer, C. Stiller and J. Frias: Object Detection and Classification in Occupancy Grid Maps Using Deep Convolutional Networks. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
302 FCPP 0.07 % 0.01 % 0.07 % 0.02 s 1 core @ 2.0 Ghz (Python + C/C++)
303 ANM 0.01 % 0.01 % 0.02 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
304 TopNet-DecayRate
This method makes use of Velodyne laser scans.
 0.01 % 0.00 % 0.01 % 92 ms NVIDIA GeForce 1080 Ti (tensorflow-gpu)
S. Wirges, T. Fischer, C. Stiller and J. Frias: Object Detection and Classification in Occupancy Grid Maps Using Deep Convolutional Networks. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
305 LaserNet 0.00 % 0.00 % 0.00 % 12 ms GPU @ 2.5 Ghz (C/C++)
G. Meyer, A. Laddha, E. Kee, C. Vallespi-Gonzalez and C. Wellington: LaserNet: An Efficient Probabilistic 3D Object Detector for Autonomous Driving. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2019.
306 JSyolo 0.00 % 0.00 % 0.00 % 0.16 s 4 cores @ 2.5 Ghz (Python)
Table as LaTeX | Only published Methods

Pedestrian


Method Setting Code Moderate Easy Hard Runtime Environment
1 FichaDL 82.50 % 90.75 % 75.66 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
2 DGIST-CellBox 81.29 % 90.04 % 76.92 % 0.1 s GPU @ 2.5 Ghz (Java + C/C++)
3 Alibaba-CityBrain 81.19 % 90.99 % 74.68 % 1.5 s GPU @ 2.5 Ghz (Python + C/C++)
4 ExtAtt 81.05 % 90.60 % 76.08 % 1.2 s GPU @ 2.5 Ghz (Python + C/C++)
5 F-PointNet
This method makes use of Velodyne laser scans.
 code 80.13 % 89.83 % 75.05 % 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.
6 HWFD 79.06 % 88.94 % 73.98 % 0.21 s one 1080Ti
7 TuSimple code 78.40 % 88.87 % 73.66 % 1.6 s GPU @ 2.5 Ghz (Python + C/C++)
F. Yang, W. Choi and Y. Lin: Exploit all the layers: Fast and accurate cnn object detector with scale dependent pooling and cascaded rejection classifiers. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition 2016.
K. He, X. Zhang, S. Ren and J. Sun: Deep residual learning for image recognition. Proceedings of the IEEE conference on computer vision and pattern recognition 2016.
8 Argus_detection_v1 77.01 % 84.86 % 72.15 % 0.25 s GPU @ 1.5 Ghz (C/C++)
9 RRC code 76.61 % 85.98 % 71.47 % 3.6 s GPU @ 2.5 Ghz (C/C++)
J. Ren, X. Chen, J. Liu, W. Sun, J. Pang, Q. Yan, Y. Tai and L. Xu: Accurate Single Stage Detector Using Recurrent Rolling Convolution. CVPR 2017.
10 ECP Faster R-CNN 76.25 % 85.96 % 70.55 % 0.25 s GPU @ 2.5 Ghz (Python)
M. Braun, S. Krebs, F. Flohr and D. Gavrila: The EuroCity Persons Dataset: A Novel Benchmark for Object Detection. CoRR 2018.
11 Aston-EAS 76.07 % 86.71 % 70.02 % 0.24 s GPU @ 2.5 Ghz (Python + C/C++)
J. Wei, J. He, Y. Zhou, K. Chen, Z. Tang and Z. Xiong: Enhanced Object Detection With Deep Convolutional Neural Networks for Advanced Driving Assistance. IEEE Transactions on Intelligent Transportation Systems 2019.
12 MHN 75.99 % 87.21 % 69.50 % 0.39 s GPU @ 2.5 Ghz (Python)
J. Cao, Y. Pang, S. Zhao and X. Li: High-Level Semantic Networks for Multi- Scale Object Detection. IEEE Transactions on Circuits and Systems for Video Technology 2019.
13 FFNet code 75.81 % 87.17 % 69.86 % 1.07 s GPU @ 1.5 Ghz (Python)
C. Zhao, Y. Qian and M. Yang: Monocular Pedestrian Orientation Estimation Based on Deep 2D-3D Feedforward. Pattern Recognition 2019.
14 SJTU-HW 75.81 % 87.17 % 69.86 % 0.85s GPU @ 1.5 Ghz (Python + C/C++)
S. Zhang, X. Zhao, L. Fang, F. Haiping and S. Haitao: LED: LOCALIZATION-QUALITY ESTIMATION EMBEDDED DETECTOR. IEEE International Conference on Image Processing 2018.
L. Fang, X. Zhao and S. Zhang: Small-objectness sensitive detection based on shifted single shot detector. Multimedia Tools and Applications 2018.
15 THICV-YDM 75.65 % 89.04 % 68.72 % 0.06 s GPU @ 2.5 Ghz (Python)
16 Noah CV Lab - SSL 75.64 % 86.57 % 70.53 % 0.1 s GPU @ 2.5 Ghz (Python)
17 Multi-3D
This method makes use of Velodyne laser scans.
 75.32 % 86.00 % 69.55 % 0.15 s 1 core @ 2.5 Ghz (C/C++)
18 Faster RCNN + Gr + A 74.95 % 86.95 % 69.50 % 1.29 s GPU @ 2.5 Ghz (Python)
19 MS-CNN code 74.89 % 85.71 % 68.99 % 0.4 s GPU @ 2.5 Ghz (C/C++)
Z. Cai, Q. Fan, R. Feris and N. Vasconcelos: A Unified Multi-scale Deep Convolutional Neural Network for Fast Object Detection. ECCV 2016.
20 Faster RCNN + G 73.75 % 85.51 % 68.54 % 1.1 s GPU @ 2.5 Ghz (Python)
21 F-ConvNet
This method makes use of Velodyne laser scans.
 code 72.91 % 83.63 % 67.18 % 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.
22 Sogo_MM 72.82 % 84.99 % 67.42 % 1.5 s GPU @ 2.5 Ghz (C/C++)
23 Faster RCNN + A 72.67 % 86.21 % 67.55 % 0.19 s GPU @ 2.5 Ghz (Python)
24 GN 72.29 % 82.93 % 65.56 % 1 s GPU @ 2.5 Ghz (Matlab + C/C++)
S. Jung and K. Hong: Deep network aided by guiding network for pedestrian detection. Pattern Recognition Letters 2017.
25 SubCNN 72.27 % 84.88 % 66.82 % 2 s GPU @ 3.5 Ghz (Python + C/C++)
Y. Xiang, W. Choi, Y. Lin and S. Savarese: Subcategory-aware Convolutional Neural Networks for Object Proposals and Detection. IEEE Winter Conference on Applications of Computer Vision (WACV) 2017.
26 FRCNN-WS 72.26 % 84.20 % 67.47 % 0.22 s 1 core @ 3.0 Ghz (Python)
27 Faster RCNN + A 72.09 % 85.35 % 66.87 % 0.19 s GPU @ 2.5 Ghz (Python + C/C++)
28 VMVS
This method makes use of Velodyne laser scans.
 71.82 % 82.80 % 66.85 % 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.
29 IVA code 71.37 % 84.61 % 64.90 % 0.4 s GPU @ 2.5 Ghz (C/C++)
Y. Zhu, J. Wang, C. Zhao, H. Guo and H. Lu: Scale-adaptive Deconvolutional Regression Network for Pedestrian Detection. ACCV 2016.
S. Ren, K. He, R. Girshick and J. Sun: Faster R-CNN: Towards real-time object detection with region proposal networks. Advances in neural information processing systems 2015.
30 MM-MRFC
This method uses optical flow information.
This method makes use of Velodyne laser scans.
 70.76 % 83.79 % 64.81 % 0.05 s GPU @ 2.5 Ghz (C/C++)
A. Costea, R. Varga and S. Nedevschi: Fast Boosting based Detection using Scale Invariant Multimodal Multiresolution Filtered Features. CVPR 2017.
31 SDP+RPN 70.42 % 82.07 % 65.09 % 0.4 s GPU @ 2.5 Ghz (Python + C/C++)
F. Yang, W. Choi and Y. Lin: Exploit All the Layers: Fast and Accurate CNN Object Detector with Scale Dependent Pooling and Cascaded Rejection Classifiers. Proceedings of the IEEE International Conference on Computer Vision and Pattern Recognition 2016.
S. Ren, K. He, R. Girshick and J. Sun: Faster R-CNN: Towards real-time object detection with region proposal networks. Advances in Neural Information Processing Systems 2015.
32 TridentNet 70.07 % 83.42 % 65.28 % 0.2 s GPU @ 2.5 Ghz (Python)
33 CSFADet 70.07 % 84.72 % 64.81 % 0.05 s GPU @ 2.5 Ghz (Python)
34 3DOP
This method uses stereo information.
 code 69.57 % 83.17 % 63.48 % 3s GPU @ 2.5 Ghz (Matlab + C/C++)
X. Chen, K. Kundu, Y. Zhu, A. Berneshawi, H. Ma, S. Fidler and R. Urtasun: 3D Object Proposals for Accurate Object Class Detection. NIPS 2015.
35 MonoPSR code 68.56 % 85.60 % 63.34 % 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.
36 DeepStereoOP 68.46 % 83.00 % 63.35 % 3.4 s GPU @ 3.5 Ghz (Matlab + C/C++)
C. Pham and J. Jeon: Robust Object Proposals Re-ranking for Object Detection in Autonomous Driving Using Convolutional Neural Networks. Signal Processing: Image Communiation 2017.
37 sensekitti code 68.41 % 82.72 % 62.72 % 4.5 s GPU @ 2.5 Ghz (Python + C/C++)
B. Yang, J. Yan, Z. Lei and S. Li: Craft Objects from Images. CVPR 2016.
38 Mono3D code 67.29 % 80.30 % 62.23 % 4.2 s GPU @ 2.5 Ghz (Matlab + C/C++)
X. Chen, K. Kundu, Z. Zhang, H. Ma, S. Fidler and R. Urtasun: Monocular 3D Object Detection for Autonomous Driving. CVPR 2016.
39 dgist_multiDetNet 67.03 % 81.96 % 61.39 % 0.05 s 1 core @ 2.5 Ghz (C/C++)
40 YOLOv3.5 66.54 % 83.87 % 61.45 % 0.05 s GPU @ 2.5 Ghz (Python)
41 Faster R-CNN code 66.24 % 79.97 % 61.09 % 2 s GPU @ 3.5 Ghz (Python + C/C++)
S. Ren, K. He, R. Girshick and J. Sun: Faster R-CNN: Towards Real- Time Object Detection with Region Proposal Networks. NIPS 2015.
42 AtrousDet 64.97 % 80.79 % 58.36 % 0.05 s TITAN X
43 SDP+CRC (ft) 64.36 % 79.22 % 59.16 % 0.6 s GPU @ 2.5 Ghz (C/C++)
F. Yang, W. Choi and Y. Lin: Exploit All the Layers: Fast and Accurate CNN Object Detector with Scale Dependent Pooling and Cascaded Rejection Classifiers. Proceedings of the IEEE International Conference on Computer Vision and Pattern Recognition 2016.
44 CRCNNA 63.69 % 78.10 % 58.41 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
45 Pose-RCNN 63.54 % 80.07 % 57.02 % 2 s >8 cores @ 2.5 Ghz (Python)
M. Braun, Q. Rao, Y. Wang and F. Flohr: Pose-RCNN: Joint object detection and pose estimation using 3D object proposals. Intelligent Transportation Systems (ITSC), 2016 IEEE 19th International Conference on 2016.
46 PCN 63.41 % 80.08 % 58.55 % 0.6 s
47 CFM 62.84 % 74.76 % 56.06 % <2 s GPU @ 2.5 Ghz (Matlab + C/C++)
Q. Hu, P. Wang, C. Shen, A. Hengel and F. Porikli: Pushing the Limits of Deep CNNs for Pedestrian Detection. IEEE Transactions on Circuits and Systems for Video Technology 2017.
48 merge12-12 62.84 % 80.27 % 56.08 % 0.2 s 4 cores @ 2.5 Ghz (Python)
49 cas+res+soft 62.71 % 80.11 % 55.99 % 0.2 s 4 cores @ 2.5 Ghz (Python)
50 cas_retina 62.37 % 79.82 % 57.15 % 0.2 s 4 cores @ 2.5 Ghz (Python)
51 cas_retina_1_13 61.87 % 79.09 % 56.70 % 0.03 s 4 cores @ 2.5 Ghz (Python)
52 MonoPair 61.57 % 78.81 % 56.51 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
53 ReSqueeze 61.33 % 73.69 % 56.65 % 0.03 s GPU @ >3.5 Ghz (Python)
54 RPN+BF code 61.22 % 77.06 % 55.22 % 0.6 s GPU @ 2.5 Ghz (Matlab + C/C++)
L. Zhang, L. Lin, X. Liang and K. He: Is Faster R-CNN Doing Well for Pedestrian Detection?. ECCV 2016.
55 JSU-NET 61.19 % 83.17 % 56.20 % 0.1 s 1 core @ 2.5 Ghz (Python)
56 Regionlets 60.83 % 73.79 % 54.72 % 1 s >8 cores @ 2.5 Ghz (C/C++)
X. Wang, M. Yang, S. Zhu and Y. Lin: Regionlets for Generic Object Detection. T-PAMI 2015.
W. Zou, X. Wang, M. Sun and Y. Lin: Generic Object Detection with Dense Neural Patterns and Regionlets. British Machine Vision Conference 2014.
C. Long, X. Wang, G. Hua, M. Yang and Y. Lin: Accurate Object Detection with Location Relaxation and Regionlets Relocalization. Asian Conference on Computer Vision 2014.
57 bin 60.73 % 71.43 % 55.78 % 15ms s GPU @ >3.5 Ghz (Python)
58 OHS-Direct 60.63 % 69.37 % 57.64 % 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.
59 anm 60.35 % 76.02 % 55.46 % 3 s 1 core @ 2.5 Ghz (C/C++)
60 PiP 59.94 % 70.52 % 56.51 % 0.05 s 1 core @ 2.5 Ghz (Python)
61 cascadercnn 59.50 % 78.79 % 54.44 % 0.36 s 4 cores @ 2.5 Ghz (Python)
62 A-VoxelNet 59.07 % 69.90 % 56.49 % 0.029 s GPU @ 2.5 Ghz (Python)
63 TANet code 59.07 % 69.90 % 56.44 % 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.
64 mymask-rcnn 58.89 % 72.55 % 52.58 % 0.3 s 1 core @ 2.5 Ghz (Python)
65 OHS-Dense 58.70 % 68.18 % 54.68 % 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.
66 DDB
This method makes use of Velodyne laser scans.
 58.53 % 69.03 % 55.90 % 0.05 s GPU @ 2.5 Ghz (Python)
67 Point-GNN
This method makes use of Velodyne laser scans.
 58.20 % 71.59 % 54.06 % 0.6 s GPU @ 2.5 Ghz (Python)
68 DeepParts 58.15 % 71.47 % 51.92 % ~1 s GPU @ 2.5 Ghz (Matlab)
Y. Tian, P. Luo, X. Wang and X. Tang: Deep Learning Strong Parts for Pedestrian Detection. ICCV 2015.
69 CompACT-Deep 58.14 % 70.93 % 52.29 % 1 s 1 core @ 2.5 Ghz (Matlab + C/C++)
Z. Cai, M. Saberian and N. Vasconcelos: Learning Complexity-Aware Cascades for Deep Pedestrian Detection. ICCV 2015.
70 MMLab-PartA^2
This method makes use of Velodyne laser scans.
 57.96 % 68.78 % 54.01 % 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.
71 AVOD-FPN
This method makes use of Velodyne laser scans.
 code 57.87 % 67.95 % 55.23 % 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.
72 LPN 57.69 % 71.87 % 53.21 % 0.2 s GPU @ 2.5 Ghz (Python + C/C++)
73 LDAM 56.68 % 64.73 % 54.21 % 0.05 s GPU @ 2.5 Ghz (C/C++)
74 FRCNN+Or code 56.68 % 71.64 % 51.53 % 0.09 s Titan Xp GPU
C. Guindel, D. Martin and J. Armingol: Fast Joint Object Detection and Viewpoint Estimation for Traffic Scene Understanding. IEEE Intelligent Transportation Systems Magazine 2018.
C. Guindel, D. Martin and J. Armingol: Joint Object Detection and Viewpoint Estimation using CNN features. IEEE International Conference on Vehicular Electronics and Safety (ICVES) 2017.
75 yolo800 56.67 % 71.26 % 50.91 % 0.13 s 4 cores @ 2.5 Ghz (Python)
76 ZKNet 56.58 % 71.15 % 51.87 % 0.01 s GPU @ 2.0 Ghz (Python)
77 CentrNet-v1
This method makes use of Velodyne laser scans.
 56.57 % 66.27 % 54.19 % 0.03 s GPU @ 2.5 Ghz (Python)
78 FilteredICF 56.53 % 69.79 % 50.32 % ~ 2 s >8 cores @ 2.5 Ghz (Matlab + C/C++)
S. Zhang, R. Benenson and B. Schiele: Filtered Channel Features for Pedestrian Detection. CVPR 2015.
79 Tencent_ADlab_Lidar
This method makes use of Velodyne laser scans.
 56.49 % 66.91 % 54.06 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
80 ARPNET 56.42 % 69.08 % 52.69 % 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.
81 FD2 56.35 % 71.37 % 51.08 % 0.01 s GPU @ >3.5 Ghz (Python + C/C++)
82 MV-RGBD-RF
This method makes use of Velodyne laser scans.
 56.18 % 72.99 % 49.72 % 4 s 4 cores @ 2.5 Ghz (C/C++)
A. Gonzalez, D. Vazquez, A. Lopez and J. Amores: On-Board Object Detection: Multicue, Multimodal, and Multiview Random Forest of Local Experts.. IEEE Trans. on Cybernetics 2016.
A. Gonzalez, G. Villalonga, J. Xu, D. Vazquez, J. Amores and A. Lopez: Multiview Random Forest of Local Experts Combining RGB and LIDAR data for Pedestrian Detection. IEEE Intelligent Vehicles Symposium (IV) 2015.
83 RFCN 55.96 % 72.32 % 49.66 % 0.2 s 4 cores @ 2.5 Ghz (Python)
84 MLOD
This method makes use of Velodyne laser scans.
 code 55.62 % 68.42 % 51.45 % 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.
85 PointPillars
This method makes use of Velodyne laser scans.
 code 55.10 % 65.29 % 52.39 % 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.
86 STD 55.04 % 68.33 % 50.85 % 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.
87 RFCN_RFB 54.98 % 70.61 % 48.75 % 0.2 s 4 cores @ 2.5 Ghz (Python)
88 Vote3Deep
This method makes use of Velodyne laser scans.
 54.80 % 67.99 % 51.17 % 1.5 s 4 cores @ 2.5 Ghz (C/C++)
M. Engelcke, D. Rao, D. Zeng Wang, C. Hay Tong and I. Posner: Vote3Deep: Fast Object Detection in 3D Point Clouds Using Efficient Convolutional Neural Networks. ArXiv e-prints 2016.
89 CHTTL MMF 54.28 % 72.79 % 49.31 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
90 epBRM
This method makes use of Velodyne laser scans.
 code 54.13 % 62.90 % 51.95 % 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.
91 NM code 54.05 % 69.81 % 49.32 % 0.01 s GPU @ 2.5 Ghz (Python)
92 PointPainting
This method makes use of Velodyne laser scans.
 53.76 % 61.86 % 50.61 % 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.
93 PDV2 53.54 % 65.59 % 47.65 % 3.7 s 1 core @ 3.0 Ghz Matlab (C/C++)
J. Shen, X. Zuo, J. Li, W. Yang and H. Ling: A novel pixel neighborhood differential statistic feature for pedestrian and face detection . Pattern Recognition 2017.
94 fasterrcnn 53.42 % 69.29 % 48.76 % 0.2 s 4 cores @ 2.5 Ghz (Python)
95 TAFT 53.15 % 67.62 % 47.08 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
J. Shen, X. Zuo, W. Yang, D. Prokhorov, X. Mei and H. Ling: Differential Features for Pedestrian Detection: A Taylor Series Perspective. IEEE Transactions on Intelligent Transportation Systems 2018.
96 pAUCEnsT 52.88 % 65.84 % 46.97 % 60 s 1 core @ 2.5 Ghz (Matlab + C/C++)
S. Paisitkriangkrai, C. Shen and A. Hengel: Pedestrian Detection with Spatially Pooled Features and Structured Ensemble Learning. arXiv 2014.
97 Multi-task DG 52.87 % 71.27 % 48.10 % 0.06 s GPU @ 2.5 Ghz (Python)
98 NEUAV 52.53 % 68.50 % 47.99 % 0.06 s GPU @ 2.5 Ghz (Python)
99 deprecated 52.32 % 67.93 % 47.77 % 0.05 s GPU @ 2.0 Ghz (Python)
100 PP-3D 52.11 % 63.07 % 49.79 % 0.1 s 1 core @ 2.5 Ghz (Python)
101 MTDP 51.81 % 68.12 % 46.95 % 0.15 s GPU @ 2.0 Ghz (Python)
102 Shift R-CNN (mono) code 51.30 % 70.86 % 46.37 % 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.
103 SCANet 51.23 % 65.06 % 47.06 % 0.17 s >8 cores @ 2.5 Ghz (Python)
104 centernet 51.09 % 69.27 % 45.40 % 0.01 s GPU @ 2.5 Ghz (Python)
105 myfaster-rcnn-v1.5 50.95 % 67.68 % 46.29 % 0.1 s 1 core @ 2.5 Ghz (Python)
106 PPFNet code 50.52 % 57.82 % 47.44 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
107 FOFNet
This method makes use of Velodyne laser scans.
 50.08 % 62.64 % 46.27 % 0.04 s GPU @ 2.5 Ghz (Python)
108 SCNet
This method makes use of Velodyne laser scans.
 49.61 % 60.95 % 46.91 % 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.
109 MMLab-PointRCNN
This method makes use of Velodyne laser scans.
 code 49.41 % 58.93 % 46.44 % 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.
110 Resnet101Faster rcnn 49.12 % 64.72 % 44.60 % 1 s 1 core @ 2.5 Ghz (Python)
111 VOXEL_FPN_HR 49.09 % 60.28 % 45.47 % 0.12 s 8 cores @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
112 SS3D_HW 49.01 % 64.67 % 42.86 % 0.4 s GPU @ 2.5 Ghz (Python)
113 cascade_gw 48.99 % 67.35 % 44.25 % 0.2 s 4 cores @ 2.5 Ghz (Python)
114 Int-YOLO code 48.76 % 64.09 % 44.31 % 0.03 s 1 core @ 2.5 Ghz (C/C++)
ERROR: Wrong syntax in BIBTEX file.
115 MP 48.73 % 60.26 % 45.05 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
116 ACFD
This method makes use of Velodyne laser scans.
 code 48.63 % 61.62 % 44.15 % 0.2 s 4 cores @ >3.5 Ghz (C/C++)
M. Dimitrievski, P. Veelaert and W. Philips: Semantically aware multilateral filter for depth upsampling in automotive LiDAR point clouds. IEEE Intelligent Vehicles Symposium, IV 2017, Los Angeles, CA, USA, June 11-14, 2017 2017.
117 R-CNN 48.57 % 62.88 % 43.05 % 4 s GPU @ 3.3 Ghz (C/C++)
J. Hosang, M. Omran, R. Benenson and B. Schiele: Taking a Deeper Look at Pedestrians. arXiv 2015.
118 mylsi-faster-rcnn 47.99 % 65.03 % 43.43 % 0.3 s 1 core @ 2.5 Ghz (Python)
119 HR-SECOND code 46.69 % 58.68 % 42.93 % 0.11 s 1 core @ 2.5 Ghz (Python + C/C++)
120 Cmerge 46.51 % 63.68 % 41.60 % 0.2 s 4 cores @ 2.5 Ghz (Python)
121 SS3D 45.79 % 61.58 % 41.14 % 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.
122 ACF 45.67 % 59.81 % 40.88 % 1 s 1 core @ 3.5 Ghz (Matlab + C/C++)
P. Doll\'ar, R. Appel, S. Belongie and P. Perona: Fast Feature Pyramids for Object Detection. PAMI 2014.
123 Fusion-DPM
This method makes use of Velodyne laser scans.
 code 44.99 % 58.93 % 40.19 % ~ 30 s 1 core @ 3.5 Ghz (Matlab + C/C++)
C. Premebida, J. Carreira, J. Batista and U. Nunes: Pedestrian Detection Combining RGB and Dense LIDAR Data. IROS 2014.
124 ACF-MR 44.79 % 58.29 % 39.94 % 0.6 s 1 core @ 3.5 Ghz (C/C++)
R. Rajaram, E. Ohn-Bar and M. Trivedi: Looking at Pedestrians at Different Scales: A Multi-resolution Approach and Evaluations. T-ITS 2016.
125 HA-SSVM 43.87 % 58.76 % 38.81 % 21 s 1 core @ >3.5 Ghz (Matlab + C/C++)
J. Xu, S. Ramos, D. Vázquez and A. López: Hierarchical Adaptive Structural SVM for Domain Adaptation. IJCV 2016.
126 AB3DMOT
This method makes use of Velodyne laser scans.
This is an online method (no batch processing).
 code 43.86 % 54.55 % 40.99 % 0.0047s 1 core @ 2.5 Ghz (python)
X. Weng and K. Kitani: A Baseline for 3D Multi-Object Tracking. arXiv:1907.03961 2019.
127 D4LCN code 43.50 % 59.55 % 37.12 % 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.
128 DPM-VOC+VP 43.26 % 59.21 % 38.12 % 8 s 1 core @ 2.5 Ghz (C/C++)
B. Pepik, M. Stark, P. Gehler and B. Schiele: Multi-view and 3D Deformable Part Models. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015.
129 PG-MonoNet 43.12 % 58.51 % 38.92 % 0.19 s GPU @ 2.5 Ghz (Python)
130 ACF-SC 42.97 % 53.30 % 38.12 % <0.3 s 1 core @ >3.5 Ghz (Matlab + C/C++)
C. Cadena, A. Dick and I. Reid: A Fast, Modular Scene Understanding System using Context-Aware Object Detection. Robotics and Automation (ICRA), 2015 IEEE International Conference on 2015.
131 SquaresICF code 42.61 % 57.08 % 37.85 % 1 s GPU @ >3.5 Ghz (C/C++)
R. Benenson, M. Mathias, T. Tuytelaars and L. Gool: Seeking the strongest rigid detector. CVPR 2013.
132 GNN 42.28 % 58.09 % 37.81 % 0.2 s 1 core @ 2.5 Ghz (Python)
133 MMRetina
This method uses stereo information.
This method uses optical flow information.
This method makes use of Velodyne laser scans.
 41.63 % 59.63 % 36.97 % 0.38 s GPU @ 2.5 Ghz (Python)
134 CSW3D
This method makes use of Velodyne laser scans.
 41.50 % 53.76 % 37.25 % 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.
135 M3D-RPN code 41.46 % 56.64 % 37.31 % 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 .
136 RTM3D code 41.43 % 58.14 % 37.09 % 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.
137 myfaster-rcnn 41.22 % 56.87 % 36.99 % 0.01 s 1 core @ 2.5 Ghz (Python)
138 FCY
This method makes use of Velodyne laser scans.
 40.85 % 56.09 % 36.40 % 0.02 s GPU @ 2.5 Ghz (Python)
139 yolov3_warp 40.64 % 55.04 % 36.33 % 0.5 s 1 core @ 2.5 Ghz (C/C++)
ERROR: Wrong syntax in BIBTEX file.
140 SubCat 40.50 % 53.75 % 35.66 % 1.2 s 6 cores @ 2.5 Ghz (Matlab + C/C++)
E. Ohn-Bar and M. Trivedi: Fast and Robust Object Detection Using Visual Subcategories. Computer Vision and Pattern Recognition Workshops Mobile Vision 2014.
141 SAANet 40.43 % 51.16 % 38.38 % 0.10 s 1 core @ 2.5 Ghz (Python)
142 Retinanet100 40.03 % 54.30 % 35.33 % 0.2 s 4 cores @ 2.5 Ghz (Python)
143 DSGN
This method uses stereo information.
 code 39.93 % 49.28 % 38.13 % 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.
144 SparsePool code 39.59 % 50.81 % 35.91 % 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.
145 SparsePool code 39.43 % 50.94 % 35.78 % 0.13 s 8 cores @ 2.5 Ghz (Python)
Z. Wang, W. Zhan and M. Tomizuka: Fusing bird view lidar point cloud and front view camera image for deep object detection. arXiv preprint arXiv:1711.06703 2017.
146 AVOD
This method makes use of Velodyne laser scans.
 code 39.43 % 50.90 % 35.75 % 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.
147 softyolo 39.30 % 54.49 % 36.66 % 0.16 s 4 cores @ 2.5 Ghz (Python)
148 ACF 39.12 % 48.42 % 35.03 % 0.2 s 1 core @ >3.5 Ghz (Matlab + C/C++)
P. Doll\'ar, R. Appel, S. Belongie and P. Perona: Fast Feature Pyramids for Object Detection. PAMI 2014.
P. Doll\'ar: Piotr's Image and Video Matlab Toolbox (PMT). .
149 pedestrian_cnn 37.90 % 52.07 % 33.58 % 1 s 1 core @ 2.5 Ghz (C/C++)
150 LSVM-MDPM-sv 37.26 % 50.74 % 33.13 % 10 s 4 cores @ 3.0 Ghz (C/C++)
P. Felzenszwalb, R. Girshick, D. McAllester and D. Ramanan: Object Detection with Discriminatively Trained Part-Based Models. PAMI 2010.
A. Geiger, C. Wojek and R. Urtasun: Joint 3D Estimation of Objects and Scene Layout. NIPS 2011.
151 multi-task CNN 37.00 % 49.38 % 33.46 % 25.1 ms GPU @ 2.0 Ghz (Python)
M. Oeljeklaus, F. Hoffmann and T. Bertram: A Fast Multi-Task CNN for Spatial Understanding of Traffic Scenes. IEEE Intelligent Transportation Systems Conference 2018.
152 Complexer-YOLO
This method makes use of Velodyne laser scans.
 36.45 % 42.16 % 32.91 % 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.
153 KD53-20 36.03 % 45.78 % 32.79 % 0.19 s 4 cores @ 2.5 Ghz (Python)
154 LSVM-MDPM-us code 35.92 % 48.73 % 31.70 % 10 s 4 cores @ 3.0 Ghz (C/C++)
P. Felzenszwalb, R. Girshick, D. McAllester and D. Ramanan: Object Detection with Discriminatively Trained Part-Based Models. PAMI 2010.
155 Lidar_ROI+Yolo(UJS) 35.58 % 47.74 % 31.48 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
156 34.81 % 44.38 % 32.10 %
157 Vote3D
This method makes use of Velodyne laser scans.
 33.04 % 42.66 % 30.59 % 0.5 s 4 cores @ 2.8 Ghz (C/C++)
D. Wang and I. Posner: Voting for Voting in Online Point Cloud Object Detection. Proceedings of Robotics: Science and Systems 2015.
158 OC Stereo
This method uses stereo information.
 30.79 % 43.50 % 28.40 % 0.35 s 1 core @ 2.5 Ghz (Python + C/C++)
159 mBoW
This method makes use of Velodyne laser scans.
 30.26 % 41.52 % 26.34 % 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.
160 BirdNet
This method makes use of Velodyne laser scans.
 29.58 % 36.62 % 28.25 % 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.
161 RT3DStereo
This method uses stereo information.
 29.30 % 41.12 % 25.25 % 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.
162 DPM-C8B1
This method uses stereo information.
 25.34 % 36.40 % 22.00 % 15 s 4 cores @ 2.5 Ghz (Matlab + C/C++)
J. Yebes, L. Bergasa and M. García-Garrido: Visual Object Recognition with 3D-Aware Features in KITTI Urban Scenes. Sensors 2015.
J. Yebes, L. Bergasa, R. Arroyo and A. Lázaro: Supervised learning and evaluation of KITTI's cars detector with DPM. IV 2014.
163 100Frcnn 21.92 % 34.07 % 19.48 % 2 s 4 cores @ 2.5 Ghz (Python + C/C++)
164 RefinedMPL 20.81 % 30.41 % 18.72 % 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.
165 R-CNN_VGG 19.97 % 26.62 % 17.96 % 10 s GPU @ 2.5 Ghz (Matlab + C/C++)
166 TopNet-Retina
This method makes use of Velodyne laser scans.
 16.45 % 22.37 % 15.43 % 52ms GeForce 1080Ti (tensorflow-gpu, v1.12)
S. Wirges, T. Fischer, C. Stiller and J. Frias: Object Detection and Classification in Occupancy Grid Maps Using Deep Convolutional Networks. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
167 TopNet-HighRes
This method makes use of Velodyne laser scans.
 15.28 % 21.22 % 13.89 % 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.
168 YOLOv2 code 11.46 % 15.37 % 9.67 % 0.02 s GPU @ 3.5 Ghz (C/C++)
J. Redmon, S. Divvala, R. Girshick and A. Farhadi: You only look once: Unified, real-time object detection. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition 2016.
J. Redmon and A. Farhadi: YOLO9000: Better, Faster, Stronger. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition 2017.
169 TopNet-UncEst
This method makes use of Velodyne laser scans.
 8.58 % 13.00 % 7.38 % 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.
170 BIP-HETERO 7.05 % 8.51 % 6.30 % ~2 s 1 core @ 2.5 Ghz (C/C++)
A. Mekonnen, F. Lerasle, A. Herbulot and C. Briand: People Detection with Heterogeneous Features and Explicit Optimization on Computation Time. Pattern Recognition (ICPR), 2014 22nd International Conference on 2014.
171 CBNet 1.33 % 1.03 % 1.41 % 1 s 4 cores @ 2.5 Ghz (Python)
172 softretina 0.26 % 0.19 % 0.26 % 0.16 s 4 cores @ 2.5 Ghz (Python)
173 JSyolo 0.12 % 0.19 % 0.12 % 0.16 s 4 cores @ 2.5 Ghz (Python)
174 TopNet-DecayRate
This method makes use of Velodyne laser scans.
 0.01 % 0.01 % 0.01 % 92 ms NVIDIA GeForce 1080 Ti (tensorflow-gpu)
S. Wirges, T. Fischer, C. Stiller and J. Frias: Object Detection and Classification in Occupancy Grid Maps Using Deep Convolutional Networks. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
Table as LaTeX | Only published Methods

Cyclist


Method Setting Code Moderate Easy Hard Runtime Environment
1 MMLab PV-RCNN
This method makes use of Velodyne laser scans.
 80.42 % 86.62 % 73.64 % 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 FichaDL 80.38 % 88.41 % 69.72 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
3 Noah CV Lab - SSL 79.10 % 86.71 % 69.66 % 0.1 s GPU @ 2.5 Ghz (Python)
4 OHS-Dense 78.42 % 85.79 % 71.80 % 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.
5 MMLab-PartA^2
This method makes use of Velodyne laser scans.
 78.29 % 88.90 % 71.19 % 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 F-ConvNet
This method makes use of Velodyne laser scans.
 code 78.05 % 86.75 % 68.12 % 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 PointPainting
This method makes use of Velodyne laser scans.
 78.04 % 87.70 % 69.27 % 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.
8 RRC code 76.81 % 86.81 % 66.59 % 3.6 s GPU @ 2.5 Ghz (C/C++)
J. Ren, X. Chen, J. Liu, W. Sun, J. Pang, Q. Yan, Y. Tai and L. Xu: Accurate Single Stage Detector Using Recurrent Rolling Convolution. CVPR 2017.
9 HWFD 76.17 % 87.02 % 66.45 % 0.21 s one 1080Ti
10 OHS-Direct 75.98 % 83.71 % 68.80 % 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.
11 Multi-3D
This method makes use of Velodyne laser scans.
 75.77 % 84.71 % 65.95 % 0.15 s 1 core @ 2.5 Ghz (C/C++)
12 MS-CNN code 75.30 % 84.88 % 65.27 % 0.4 s GPU @ 2.5 Ghz (C/C++)
Z. Cai, Q. Fan, R. Feris and N. Vasconcelos: A Unified Multi-scale Deep Convolutional Neural Network for Fast Object Detection. ECCV 2016.
13 VOXEL_FPN_HR 75.24 % 87.73 % 68.60 % 0.12 s 8 cores @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
14 TuSimple code 75.22 % 83.68 % 65.22 % 1.6 s GPU @ 2.5 Ghz (Python + C/C++)
F. Yang, W. Choi and Y. Lin: Exploit all the layers: Fast and accurate cnn object detector with scale dependent pooling and cascaded rejection classifiers. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition 2016.
K. He, X. Zhang, S. Ren and J. Sun: Deep residual learning for image recognition. Proceedings of the IEEE conference on computer vision and pattern recognition 2016.
15 Point-GNN
This method makes use of Velodyne laser scans.
 75.08 % 85.75 % 68.69 % 0.6 s GPU @ 2.5 Ghz (Python)
16 ExtAtt 75.08 % 86.09 % 65.30 % 1.2 s GPU @ 2.5 Ghz (Python + C/C++)
17 Deep3DBox 74.78 % 84.36 % 64.05 % 1.5 s GPU @ 2.5 Ghz (C/C++)
A. Mousavian, D. Anguelov, J. Flynn and J. Kosecka: 3D Bounding Box Estimation Using Deep Learning and Geometry. CVPR 2017.
18 SDP+RPN 73.85 % 82.59 % 64.87 % 0.4 s GPU @ 2.5 Ghz (Python + C/C++)
F. Yang, W. Choi and Y. Lin: Exploit All the Layers: Fast and Accurate CNN Object Detector with Scale Dependent Pooling and Cascaded Rejection Classifiers. Proceedings of the IEEE International Conference on Computer Vision and Pattern Recognition 2016.
S. Ren, K. He, R. Girshick and J. Sun: Faster R-CNN: Towards real-time object detection with region proposal networks. Advances in Neural Information Processing Systems 2015.
19 sensekitti code 73.48 % 82.90 % 64.03 % 4.5 s GPU @ 2.5 Ghz (Python + C/C++)
B. Yang, J. Yan, Z. Lei and S. Li: Craft Objects from Images. CVPR 2016.
20 MMLab-PointRCNN
This method makes use of Velodyne laser scans.
 code 73.42 % 86.21 % 66.45 % 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.
21 F-PointNet
This method makes use of Velodyne laser scans.
 code 73.16 % 86.86 % 65.21 % 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.
22 FOFNet
This method makes use of Velodyne laser scans.
 72.96 % 87.12 % 66.37 % 0.04 s GPU @ 2.5 Ghz (Python)
23 HR-SECOND code 72.77 % 84.21 % 66.25 % 0.11 s 1 core @ 2.5 Ghz (Python + C/C++)
24 MonoPSR code 72.08 % 82.06 % 62.43 % 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.
25 ARPNET 71.95 % 84.96 % 65.21 % 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 SubCNN 71.72 % 79.36 % 62.74 % 2 s GPU @ 3.5 Ghz (Python + C/C++)
Y. Xiang, W. Choi, Y. Lin and S. Savarese: Subcategory-aware Convolutional Neural Networks for Object Proposals and Detection. IEEE Winter Conference on Applications of Computer Vision (WACV) 2017.
27 STD 71.63 % 83.99 % 64.92 % 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 Sogo_MM 71.57 % 79.35 % 62.22 % 1.5 s GPU @ 2.5 Ghz (C/C++)
29 PiP 71.52 % 82.97 % 65.52 % 0.05 s 1 core @ 2.5 Ghz (Python)
30 Faster RCNN + Gr + A 70.78 % 83.99 % 63.36 % 1.29 s GPU @ 2.5 Ghz (Python)
31 AB3DMOT
This method makes use of Velodyne laser scans.
This is an online method (no batch processing).
 code 70.18 % 82.86 % 63.55 % 0.0047s 1 core @ 2.5 Ghz (Python)
X. Weng and K. Kitani: A Baseline for 3D Multi-Object Tracking. arXiv:1907.03961 2019.
32 TridentNet 69.63 % 81.97 % 59.52 % 0.2 s GPU @ 2.5 Ghz (Python)
33 MP 69.52 % 85.05 % 63.17 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
34 LDAM 69.31 % 80.20 % 63.85 % 0.05 s GPU @ 2.5 Ghz (C/C++)
35 PointPillars
This method makes use of Velodyne laser scans.
 code 68.98 % 83.97 % 62.17 % 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.
36 DGIST-CellBox 68.92 % 83.72 % 61.32 % 0.1 s GPU @ 2.5 Ghz (Java + C/C++)
37 Vote3Deep
This method makes use of Velodyne laser scans.
 68.82 % 78.41 % 62.50 % 1.5 s 4 cores @ 2.5 Ghz (C/C++)
M. Engelcke, D. Rao, D. Zeng Wang, C. Hay Tong and I. Posner: Vote3Deep: Fast Object Detection in 3D Point Clouds Using Efficient Convolutional Neural Networks. ArXiv e-prints 2016.
38 3DOP
This method uses stereo information.
 code 68.71 % 80.52 % 61.07 % 3s GPU @ 2.5 Ghz (Matlab + C/C++)
X. Chen, K. Kundu, Y. Zhu, A. Berneshawi, H. Ma, S. Fidler and R. Urtasun: 3D Object Proposals for Accurate Object Class Detection. NIPS 2015.
39 Pose-RCNN 68.40 % 81.53 % 59.43 % 2 s >8 cores @ 2.5 Ghz (Python)
M. Braun, Q. Rao, Y. Wang and F. Flohr: Pose-RCNN: Joint object detection and pose estimation using 3D object proposals. Intelligent Transportation Systems (ITSC), 2016 IEEE 19th International Conference on 2016.
40 TANet code 68.20 % 82.24 % 62.13 % 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.
41 Faster RCNN + G 68.09 % 83.51 % 60.60 % 1.1 s GPU @ 2.5 Ghz (Python)
42 Faster RCNN + A 67.84 % 82.06 % 60.52 % 0.19 s GPU @ 2.5 Ghz (Python)
43 Tencent_ADlab_Lidar
This method makes use of Velodyne laser scans.
 67.82 % 82.74 % 61.06 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
44 IVA code 67.57 % 78.48 % 58.83 % 0.4 s GPU @ 2.5 Ghz (C/C++)
Y. Zhu, J. Wang, C. Zhao, H. Guo and H. Lu: Scale-adaptive Deconvolutional Regression Network for Pedestrian Detection. ACCV 2016.
S. Ren, K. He, R. Girshick and J. Sun: Faster R-CNN: Towards real-time object detection with region proposal networks. Advances in neural information processing systems 2015.
45 A-VoxelNet 67.37 % 81.32 % 60.27 % 0.029 s GPU @ 2.5 Ghz (Python)
46 DeepStereoOP 67.22 % 79.35 % 58.60 % 3.4 s GPU @ 3.5 Ghz (Matlab + C/C++)
C. Pham and J. Jeon: Robust Object Proposals Re-ranking for Object Detection in Autonomous Driving Using Convolutional Neural Networks. Signal Processing: Image Communiation 2017.
47 Faster RCNN + A 67.15 % 83.77 % 59.85 % 0.19 s GPU @ 2.5 Ghz (Python + C/C++)
48 SAANet 66.58 % 83.07 % 59.88 % 0.10 s 1 core @ 2.5 Ghz (Python)
49 epBRM
This method makes use of Velodyne laser scans.
 code 66.51 % 79.65 % 60.31 % 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.
50 Mono3D code 65.15 % 77.19 % 57.88 % 4.2 s GPU @ 2.5 Ghz (Matlab + C/C++)
X. Chen, K. Kundu, Z. Zhang, H. Ma, S. Fidler and R. Urtasun: Monocular 3D Object Detection for Autonomous Driving. CVPR 2016.
51 deprecated 63.34 % 83.91 % 53.78 % 0.05 s GPU @ 2.0 Ghz (Python)
52 CentrNet-v1
This method makes use of Velodyne laser scans.
 62.99 % 78.90 % 56.46 % 0.03 s GPU @ 2.5 Ghz (Python)
53 Faster R-CNN code 62.86 % 72.40 % 54.97 % 2 s GPU @ 3.5 Ghz (Python + C/C++)
S. Ren, K. He, R. Girshick and J. Sun: Faster R-CNN: Towards Real- Time Object Detection with Region Proposal Networks. NIPS 2015.
54 AtrousDet 62.50 % 79.02 % 53.87 % 0.05 s TITAN X
55 SCNet
This method makes use of Velodyne laser scans.
 62.50 % 78.48 % 56.34 % 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.
56 SCANet 62.31 % 76.50 % 56.06 % 0.17 s >8 cores @ 2.5 Ghz (Python)
57 DDB
This method makes use of Velodyne laser scans.
 61.41 % 78.04 % 55.37 % 0.05 s GPU @ 2.5 Ghz (Python)
58 PP-3D 61.29 % 77.75 % 54.59 % 0.1 s 1 core @ 2.5 Ghz (Python)
59 AVOD-FPN
This method makes use of Velodyne laser scans.
 code 60.79 % 70.38 % 55.37 % 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.
60 SDP+CRC (ft) 60.72 % 75.63 % 53.00 % 0.6 s GPU @ 2.5 Ghz (C/C++)
F. Yang, W. Choi and Y. Lin: Exploit All the Layers: Fast and Accurate CNN Object Detector with Scale Dependent Pooling and Cascaded Rejection Classifiers. Proceedings of the IEEE International Conference on Computer Vision and Pattern Recognition 2016.
61 Complexer-YOLO
This method makes use of Velodyne laser scans.
 59.78 % 66.94 % 55.63 % 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.
62 merge12-12 59.48 % 77.66 % 51.41 % 0.2 s 4 cores @ 2.5 Ghz (Python)
63 cas+res+soft 59.43 % 77.85 % 51.34 % 0.2 s 4 cores @ 2.5 Ghz (Python)
64 YOLOv3.5 58.57 % 79.16 % 51.74 % 0.05 s GPU @ 2.5 Ghz (Python)
65 Regionlets 58.52 % 71.12 % 50.83 % 1 s >8 cores @ 2.5 Ghz (C/C++)
X. Wang, M. Yang, S. Zhu and Y. Lin: Regionlets for Generic Object Detection. T-PAMI 2015.
W. Zou, X. Wang, M. Sun and Y. Lin: Generic Object Detection with Dense Neural Patterns and Regionlets. British Machine Vision Conference 2014.
C. Long, X. Wang, G. Hua, M. Yang and Y. Lin: Accurate Object Detection with Location Relaxation and Regionlets Relocalization. Asian Conference on Computer Vision 2014.
66 cascadercnn 58.08 % 77.24 % 51.13 % 0.36 s 4 cores @ 2.5 Ghz (Python)
67 GA_rpn500 57.82 % 76.06 % 49.00 % 1 s 1 core @ 2.5 Ghz (Python)
68 GA2500 57.82 % 76.06 % 48.99 % 0.2 s 1 core @ 2.5 Ghz (Python)
69 bin 57.62 % 64.36 % 50.70 % 15ms s GPU @ >3.5 Ghz (Python)
70 dgist_multiDetNet 57.44 % 78.42 % 49.84 % 0.05 s 1 core @ 2.5 Ghz (C/C++)
71 GA_FULLDATA 57.20 % 75.50 % 50.26 % 1 s 4 cores @ 2.5 Ghz (Python)
72 cas_retina 57.14 % 73.97 % 50.32 % 0.2 s 4 cores @ 2.5 Ghz (Python)
73 FRCNN+Or code 57.01 % 70.99 % 50.14 % 0.09 s Titan Xp GPU
C. Guindel, D. Martin and J. Armingol: Fast Joint Object Detection and Viewpoint Estimation for Traffic Scene Understanding. IEEE Intelligent Transportation Systems Magazine 2018.
C. Guindel, D. Martin and J. Armingol: Joint Object Detection and Viewpoint Estimation using CNN features. IEEE International Conference on Vehicular Electronics and Safety (ICVES) 2017.
74 CSFADet 56.88 % 73.82 % 50.22 % 0.05 s GPU @ 2.5 Ghz (Python)
75 cas_retina_1_13 56.39 % 72.80 % 49.71 % 0.03 s 4 cores @ 2.5 Ghz (Python)
76 MonoPair 56.37 % 74.77 % 48.37 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
77 GA_BALANCE 56.07 % 78.33 % 49.02 % 1 s 1 core @ 2.5 Ghz (Python)
78 MLOD
This method makes use of Velodyne laser scans.
 code 56.04 % 75.35 % 49.11 % 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.
79 bigger_ga 55.66 % 73.05 % 47.31 % 1 s 1 core @ 2.5 Ghz (Python)
80 Multi-task DG 55.30 % 75.48 % 48.22 % 0.06 s GPU @ 2.5 Ghz (Python)
81 ReSqueeze 54.50 % 69.64 % 48.24 % 0.03 s GPU @ >3.5 Ghz (Python)
82 CRCNNA 53.41 % 69.81 % 46.29 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
83 AVOD
This method makes use of Velodyne laser scans.
 code 52.60 % 66.45 % 46.39 % 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.
84 GAFM 51.40 % 73.43 % 44.61 % 0.5 s 1 core @ 2.5 Ghz (Python)
85 JSU-NET 51.10 % 72.92 % 44.26 % 0.1 s 1 core @ 2.5 Ghz (Python)
86 ZKNet 49.48 % 66.29 % 42.81 % 0.01 s GPU @ 2.0 Ghz (Python)
87 anm 49.05 % 66.96 % 43.44 % 3 s 1 core @ 2.5 Ghz (C/C++)
88 ga50 49.02 % 70.25 % 42.52 % 1 s 1 core @ 2.5 Ghz (Python)
89 NEUAV 48.65 % 69.50 % 42.64 % 0.06 s GPU @ 2.5 Ghz (Python)
90 LPN 48.57 % 65.77 % 42.66 % 0.2 s GPU @ 2.5 Ghz (Python + C/C++)
91 mylsi-faster-rcnn 47.90 % 69.04 % 41.72 % 0.3 s 1 core @ 2.5 Ghz (Python)
92 fasterrcnn 47.87 % 64.39 % 42.03 % 0.2 s 4 cores @ 2.5 Ghz (Python)
93 BirdNet
This method makes use of Velodyne laser scans.
 47.64 % 64.97 % 44.66 % 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.
94 yolo800 47.31 % 63.22 % 42.28 % 0.13 s 4 cores @ 2.5 Ghz (Python)
95 RTM3D code 46.79 % 66.89 % 40.09 % 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.
96 RFCN 46.70 % 62.09 % 40.71 % 0.2 s 4 cores @ 2.5 Ghz (Python)
97 NM code 45.82 % 60.69 % 40.83 % 0.01 s GPU @ 2.5 Ghz (Python)
98 SS3D_HW 45.53 % 61.79 % 39.03 % 0.4 s GPU @ 2.5 Ghz (Python)
99 FCY
This method makes use of Velodyne laser scans.
 45.41 % 63.53 % 41.09 % 0.02 s GPU @ 2.5 Ghz (Python)
100 PG-MonoNet 45.40 % 63.75 % 37.14 % 0.19 s GPU @ 2.5 Ghz (Python)
101 RFCN_RFB 45.28 % 60.06 % 39.66 % 0.2 s 4 cores @ 2.5 Ghz (Python)
102 Cmerge 44.87 % 64.38 % 37.80 % 0.2 s 4 cores @ 2.5 Ghz (Python)
103 SparsePool code 44.57 % 60.53 % 40.37 % 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.
104 Shift R-CNN (mono) code 42.96 % 63.24 % 38.22 % 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.
105 myfaster-rcnn-v1.5 42.89 % 59.60 % 38.07 % 0.1 s 1 core @ 2.5 Ghz (Python)
106 D4LCN code 42.86 % 65.29 % 36.29 % 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.
107 cascade_gw 42.84 % 63.58 % 36.94 % 0.2 s 4 cores @ 2.5 Ghz (Python)
108 FD2 42.67 % 62.54 % 38.41 % 0.01 s GPU @ >3.5 Ghz (Python + C/C++)
109 centernet 42.45 % 58.95 % 37.56 % 0.01 s GPU @ 2.5 Ghz (Python)
110 M3D-RPN code 41.54 % 61.54 % 35.23 % 0.16 s GPU @ 1.5 Ghz (Python)
G. Brazil and X. Liu: M3D-RPN: Monocular 3D Region Proposal Network for Object Detection . ICCV 2019 .
111 MV-RGBD-RF
This method makes use of Velodyne laser scans.
 40.94 % 51.10 % 34.83 % 4 s 4 cores @ 2.5 Ghz (C/C++)
A. Gonzalez, D. Vazquez, A. Lopez and J. Amores: On-Board Object Detection: Multicue, Multimodal, and Multiview Random Forest of Local Experts.. IEEE Trans. on Cybernetics 2016.
A. Gonzalez, G. Villalonga, J. Xu, D. Vazquez, J. Amores and A. Lopez: Multiview Random Forest of Local Experts Combining RGB and LIDAR data for Pedestrian Detection. IEEE Intelligent Vehicles Symposium (IV) 2015.
112 MTDP 40.46 % 53.83 % 35.74 % 0.15 s GPU @ 2.0 Ghz (Python)
113 Int-YOLO code 39.83 % 53.34 % 34.16 % 0.03 s 1 core @ 2.5 Ghz (C/C++)
ERROR: Wrong syntax in BIBTEX file.
114 GNN 39.80 % 58.30 % 34.56 % 0.2 s 1 core @ 2.5 Ghz (Python)
115 SparsePool code 36.26 % 44.21 % 32.57 % 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.
116 myfaster-rcnn 35.81 % 54.28 % 31.82 % 0.01 s 1 core @ 2.5 Ghz (Python)
117 SS3D 35.48 % 52.97 % 31.07 % 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.
118 DSGN
This method uses stereo information.
 code 35.15 % 49.10 % 31.41 % 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.
119 pAUCEnsT 34.90 % 50.51 % 30.35 % 60 s 1 core @ 2.5 Ghz (Matlab + C/C++)
S. Paisitkriangkrai, C. Shen and A. Hengel: Pedestrian Detection with Spatially Pooled Features and Structured Ensemble Learning. arXiv 2014.
120 Retinanet100 32.30 % 46.60 % 28.29 % 0.2 s 4 cores @ 2.5 Ghz (Python)
121 TopNet-Retina
This method makes use of Velodyne laser scans.
 31.98 % 47.51 % 29.84 % 52ms GeForce 1080Ti (tensorflow-gpu, v1.12)
S. Wirges, T. Fischer, C. Stiller and J. Frias: Object Detection and Classification in Occupancy Grid Maps Using Deep Convolutional Networks. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
122 yolov3_warp 29.48 % 44.46 % 25.84 % 0.5 s 1 core @ 2.5 Ghz (C/C++)
ERROR: Wrong syntax in BIBTEX file.
123 OC Stereo
This method uses stereo information.
 28.76 % 43.18 % 24.80 % 0.35 s 1 core @ 2.5 Ghz (Python + C/C++)
124 MMRetina
This method uses stereo information.
This method uses optical flow information.
This method makes use of Velodyne laser scans.
 28.00 % 43.71 % 24.62 % 0.38 s GPU @ 2.5 Ghz (Python)
125 Vote3D
This method makes use of Velodyne laser scans.
 27.99 % 39.81 % 25.19 % 0.5 s 4 cores @ 2.8 Ghz (C/C++)
D. Wang and I. Posner: Voting for Voting in Online Point Cloud Object Detection. Proceedings of Robotics: Science and Systems 2015.
126 softyolo 27.90 % 41.90 % 24.74 % 0.16 s 4 cores @ 2.5 Ghz (Python)
127 LSVM-MDPM-us code 27.81 % 37.66 % 24.83 % 10 s 4 cores @ 3.0 Ghz (C/C++)
P. Felzenszwalb, R. Girshick, D. McAllester and D. Ramanan: Object Detection with Discriminatively Trained Part-Based Models. PAMI 2010.
128 DPM-VOC+VP 27.73 % 41.58 % 24.61 % 8 s 1 core @ 2.5 Ghz (C/C++)
B. Pepik, M. Stark, P. Gehler and B. Schiele: Multi-view and 3D Deformable Part Models. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015.
129 100Frcnn 27.69 % 43.23 % 23.91 % 2 s 4 cores @ 2.5 Ghz (Python + C/C++)
130 RefinedMPL 27.17 % 44.47 % 22.84 % 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.
131 LSVM-MDPM-sv 26.05 % 35.70 % 23.56 % 10 s 4 cores @ 3.0 Ghz (C/C++)
P. Felzenszwalb, R. Girshick, D. McAllester and D. Ramanan: Object Detection with Discriminatively Trained Part-Based Models. PAMI 2010.
A. Geiger, C. Wojek and R. Urtasun: Joint 3D Estimation of Objects and Scene Layout. NIPS 2011.
132 DPM-C8B1
This method uses stereo information.
 25.57 % 41.47 % 21.93 % 15 s 4 cores @ 2.5 Ghz (Matlab + C/C++)
J. Yebes, L. Bergasa and M. García-Garrido: Visual Object Recognition with 3D-Aware Features in KITTI Urban Scenes. Sensors 2015.
J. Yebes, L. Bergasa, R. Arroyo and A. Lázaro: Supervised learning and evaluation of KITTI's cars detector with DPM. IV 2014.
133 BdCost+DA+BB+MS 25.52 % 33.92 % 21.14 % TBD s 4 cores @ 2.5 Ghz (C/C++)
134 R-CNN_VGG 25.14 % 34.28 % 22.17 % 10 s GPU @ 2.5 Ghz (Matlab + C/C++)
135 Lidar_ROI+Yolo(UJS) 24.42 % 36.43 % 21.78 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
136 BdCost+DA+BB 20.00 % 26.87 % 16.76 % TBD s 4 cores @ 2.5 Ghz (C/C++)
137 mBoW
This method makes use of Velodyne laser scans.
 17.63 % 26.66 % 16.02 % 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.
138 TopNet-HighRes
This method makes use of Velodyne laser scans.
 13.98 % 22.86 % 14.52 % 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.
139 mymask-rcnn 13.58 % 18.03 % 12.42 % 0.3 s 1 core @ 2.5 Ghz (Python)
140 RT3DStereo
This method uses stereo information.
 12.96 % 19.58 % 11.47 % 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.
141 KD53-20 12.81 % 20.05 % 11.99 % 0.19 s 4 cores @ 2.5 Ghz (Python)
142 TopNet-UncEst
This method makes use of Velodyne laser scans.
 12.00 % 18.14 % 11.85 % 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.
143 CBNet 0.39 % 0.24 % 0.44 % 1 s 4 cores @ 2.5 Ghz (Python)
144 softretina 0.25 % 0.16 % 0.18 % 0.16 s 4 cores @ 2.5 Ghz (Python)
145 YOLOv2 code 0.06 % 0.15 % 0.07 % 0.02 s GPU @ 3.5 Ghz (C/C++)
J. Redmon, S. Divvala, R. Girshick and A. Farhadi: You only look once: Unified, real-time object detection. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition 2016.
J. Redmon and A. Farhadi: YOLO9000: Better, Faster, Stronger. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition 2017.
146 TopNet-DecayRate
This method makes use of Velodyne laser scans.
 0.04 % 0.00 % 0.04 % 92 ms NVIDIA GeForce 1080 Ti (tensorflow-gpu)
S. Wirges, T. Fischer, C. Stiller and J. Frias: Object Detection and Classification in Occupancy Grid Maps Using Deep Convolutional Networks. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
147 JSyolo 0.03 % 0.02 % 0.04 % 0.16 s 4 cores @ 2.5 Ghz (Python)
Table as LaTeX | Only published Methods

Object Detection and Orientation Estimation Evaluation

Cars


Method Setting Code Moderate Easy Hard Runtime Environment
1 MMLab PV-RCNN
This method makes use of Velodyne laser scans.
 94.57 % 98.15 % 91.85 % 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 MVRA + I-FRCNN+ 94.46 % 95.66 % 81.74 % 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.
3 CPRCCNN 94.24 % 96.31 % 89.71 % 0.1 s 1 core @ 2.5 Ghz (Python)
4 EPNet 94.22 % 96.13 % 89.68 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
5 Patches - EMP
This method makes use of Velodyne laser scans.
 93.58 % 97.88 % 90.31 % 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.
6 MLF_PointCas 93.34 % 96.66 % 85.87 % 0.1 s GPU @ 2.5 Ghz (Python)
7 Deep MANTA 93.31 % 98.83 % 82.95 % 0.7 s GPU @ 2.5 Ghz (Python + C/C++)
F. Chabot, M. Chaouch, J. Rabarisoa, C. Teulière and T. Chateau: Deep MANTA: A Coarse-to-fine Many-Task Network for joint 2D and 3D vehicle analysis from monocular image. CVPR 2017.
8 THICV-YDM 93.11 % 96.07 % 80.52 % 0.06 s GPU @ 2.5 Ghz (Python)
9 ELE 93.07 % 98.42 % 90.17 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
10 RGB3D
This method makes use of Velodyne laser scans.
 92.94 % 96.52 % 87.83 % 0.39 s GPU @ 2.5 Ghz (Python)
11 PointRCNN-deprecated
This method makes use of Velodyne laser scans.
 92.74 % 96.70 % 85.51 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
12 OHS-Direct 92.58 % 96.08 % 89.60 % 0.03 s 1 core @ 2.5 Ghz (Python + C/C++)
Q. Chen, L. Sun, Z. Wang, K. Jia and A. Yuille: Object as Hotspots: An Anchor-Free 3D Object Detection Approach via Firing of Hotspots. 2019.
13 SARPNET 92.58 % 95.82 % 87.33 % 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.
14 Patches
This method makes use of Velodyne laser scans.
 92.57 % 96.31 % 87.41 % 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.
15 ORP 92.57 % 96.22 % 85.06 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
16 R-GCN 92.53 % 96.16 % 87.45 % 0.16 s GPU @ 2.5 Ghz (Python)
17 PPFNet code 92.52 % 96.30 % 87.44 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
18 PI-RCNN 92.52 % 96.15 % 87.47 % 0.1 s 1 core @ 2.5 Ghz (Python)
19 PointPainting
This method makes use of Velodyne laser scans.
 92.43 % 98.36 % 89.49 % 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.
20 OHS-Dense 92.32 % 95.83 % 89.39 % 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.
21 SegVoxelNet 92.16 % 95.86 % 86.90 % 0.04 s 1 core @ 2.5 Ghz (Python)
22 CP
This method makes use of Velodyne laser scans.
 92.16 % 96.05 % 87.22 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
23 PointRGCN 92.15 % 97.48 % 86.83 % 0.26 s GPU @ V100 (Python)
24 F-ConvNet
This method makes use of Velodyne laser scans.
 code 91.98 % 95.81 % 79.83 % 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.
25 PointCSE 91.95 % 95.52 % 86.75 % 0.02 s 1 core @ 2.5 Ghz (C/C++)
26 IE-PointRCNN 91.94 % 96.00 % 86.84 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
27 AB3DMOT
This method makes use of Velodyne laser scans.
This is an online method (no batch processing).
 code 91.87 % 95.86 % 86.78 % 0.0047s 1 core @ 2.5 Ghz (Python)
X. Weng and K. Kitani: A Baseline for 3D Multi-Object Tracking. arXiv:1907.03961 2019.
28 NU-optim 91.87 % 95.17 % 86.54 % 0.04 s GPU @ >3.5 Ghz (Python)
29 MMLab-PointRCNN
This method makes use of Velodyne laser scans.
 code 91.77 % 95.90 % 86.92 % 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.
30 MMLab-PartA^2
This method makes use of Velodyne laser scans.
 91.73 % 95.00 % 88.86 % 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.
31 RUC 91.25 % 95.01 % 88.14 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
32 MLF_SecCas 91.18 % 96.19 % 83.25 % 0.05 s 1 core @ 2.5 Ghz (Python)
33 HRI-FusionRCNN 91.03 % 94.30 % 83.31 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
34 deprecated 91.02 % 94.06 % 78.56 % 0.05 s GPU @ 2.0 Ghz (Python)
35 HRI-VoxelFPN 90.76 % 96.35 % 85.37 % 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.
36 Alibaba-AILabsX
This method makes use of Velodyne laser scans.
 90.75 % 96.06 % 83.22 % 0.05 s 1 core @ >3.5 Ghz (C/C++)
37 PointPillars
This method makes use of Velodyne laser scans.
 code 90.70 % 93.84 % 87.47 % 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.
38 TBA 90.69 % 93.55 % 87.56 % 0.07 s 1 core @ 2.5 Ghz (Python)
39 SRF 90.54 % 95.86 % 85.30 % 0.05 s GPU @ 2.5 Ghz (Python + C/C++)
40 CentrNet-v1
This method makes use of Velodyne laser scans.
 90.48 % 93.79 % 87.43 % 0.03 s GPU @ 2.5 Ghz (Python)
41 MMV 90.41 % 93.93 % 82.79 % 0.4 s GPU @ 2.5 Ghz (C/C++)
42 DDB
This method makes use of Velodyne laser scans.
 90.38 % 93.21 % 86.42 % 0.05 s GPU @ 2.5 Ghz (Python)
43 OACV 90.35 % 93.95 % 81.90 % 0.23 s GPU @ 2.5 Ghz (Python)
44 MVSLN 90.26 % 95.95 % 82.75 % 0.1s s 1 core @ 2.5 Ghz (C/C++)
45 3D IoU Loss
This method makes use of Velodyne laser scans.
 90.21 % 95.60 % 84.96 % 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.
46 ARPNET 90.11 % 93.42 % 82.56 % 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.
47 TANet code 90.11 % 93.52 % 84.61 % 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.
48 FOFNet
This method makes use of Velodyne laser scans.
 90.05 % 93.87 % 84.52 % 0.04 s GPU @ 2.5 Ghz (Python)
49 SFB-SECOND 90.04 % 95.99 % 84.70 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
50 PTS
This method makes use of Velodyne laser scans.
 code 90.03 % 95.41 % 84.73 % 0.01 s 1 core @ 2.5 Ghz (C/C++)
51 A-VoxelNet 90.00 % 93.24 % 82.31 % 0.029 s GPU @ 2.5 Ghz (Python)
52 Sogo_MM 89.97 % 94.15 % 79.94 % 1.5 s GPU @ 2.5 Ghz (C/C++)
53 PCSC-Net 89.90 % 93.54 % 86.10 % 0.04 s 1 core @ 2.5 Ghz (C/C++)
54 Deep3DBox 89.88 % 94.62 % 76.40 % 1.5 s GPU @ 2.5 Ghz (C/C++)
A. Mousavian, D. Anguelov, J. Flynn and J. Kosecka: 3D Bounding Box Estimation Using Deep Learning and Geometry. CVPR 2017.
55 SECOND-V1.5
This method makes use of Velodyne laser scans.
 code 89.88 % 95.53 % 84.46 % 0.04 s GPU @ 2.0 Ghz (Python + C/C++)
56 PointPiallars_SECA 89.86 % 92.96 % 86.46 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
57 Tencent_ADlab_Lidar
This method makes use of Velodyne laser scans.
 89.82 % 93.37 % 85.67 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
58 VOXEL_FPN_HR 89.81 % 93.52 % 84.59 % 0.12 s 8 cores @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
59 BVVF 89.77 % 95.55 % 84.48 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
60 MPNet
This method makes use of Velodyne laser scans.
 89.75 % 94.31 % 86.07 % 0.02 s GPU @ 2.5 Ghz (Python + C/C++)
61 GPP code 89.68 % 93.94 % 80.60 % 0.23 s GPU @ 1.5 Ghz (Python + C/C++)
A. Rangesh and M. Trivedi: Ground plane polling for 6dof pose estimation of objects on the road. arXiv preprint arXiv:1811.06666 2018.
62 SubCNN 89.53 % 94.11 % 79.14 % 2 s GPU @ 3.5 Ghz (Python + C/C++)
Y. Xiang, W. Choi, Y. Lin and S. Savarese: Subcategory-aware Convolutional Neural Networks for Object Proposals and Detection. IEEE Winter Conference on Applications of Computer Vision (WACV) 2017.
63 SAANet 89.46 % 95.64 % 82.12 % 0.10 s 1 core @ 2.5 Ghz (Python)
64 SCNet
This method makes use of Velodyne laser scans.
 89.36 % 95.23 % 84.03 % 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.
65 RUC code 89.26 % 92.28 % 85.38 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
66 AVOD
This method makes use of Velodyne laser scans.
 code 89.22 % 94.98 % 82.14 % 0.08 s Titan X (pascal)
J. Ku, M. Mozifian, J. Lee, A. Harakeh and S. Waslander: Joint 3D Proposal Generation and Object Detection from View Aggregation. IROS 2018.
67 RUC code 88.90 % 92.68 % 84.04 % 0.12 s 1 core @ 2.5 Ghz (C/C++)
68 PAD 88.71 % 93.09 % 84.86 % 0.15 s 1 core @ 2.5 Ghz (Python)
69 AVOD-FPN
This method makes use of Velodyne laser scans.
 code 88.61 % 94.65 % 83.71 % 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.
70 SS3D_HW 88.50 % 94.45 % 68.61 % 0.4 s GPU @ 2.5 Ghz (Python)
71 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).
 87.83 % 95.24 % 83.39 % 0.013 s 1 core @ 2.5 Ghz (Python + C/C++)
72 DeepStereoOP 87.81 % 93.68 % 77.60 % 3.4 s GPU @ 3.5 Ghz (Matlab + C/C++)
C. Pham and J. Jeon: Robust Object Proposals Re-ranking for Object Detection in Autonomous Driving Using Convolutional Neural Networks. Signal Processing: Image Communiation 2017.
73 FCY
This method makes use of Velodyne laser scans.
 87.78 % 93.43 % 82.50 % 0.02 s GPU @ 2.5 Ghz (Python)
74 3DBN
This method makes use of Velodyne laser scans.
 87.59 % 93.34 % 79.91 % 0.13s 1080Ti (Python+C/C++)
X. Li, J. Guivant, N. Kwok and Y. Xu: 3D Backbone Network for 3D Object Detection. CoRR 2019.
75 FQNet 87.49 % 93.66 % 73.61 % 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.
76 Shift R-CNN (mono) code 87.47 % 93.75 % 77.19 % 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.
77 PP-3D 87.46 % 93.09 % 79.88 % 0.1 s 1 core @ 2.5 Ghz (Python)
78 MonoPSR code 87.45 % 93.29 % 72.26 % 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.
79 Mono3D code 87.28 % 93.13 % 77.00 % 4.2 s GPU @ 2.5 Ghz (Matlab + C/C++)
X. Chen, K. Kundu, Z. Zhang, H. Ma, S. Fidler and R. Urtasun: Monocular 3D Object Detection for Autonomous Driving. CVPR 2016.
80 3DNN 87.08 % 93.78 % 79.72 % 0.09 s GPU @ 2.5 Ghz (Python)
81 SMOKE 87.02 % 92.94 % 77.12 % 0.03 s GPU @ 2.5 Ghz (Python)
82 MonoSS 86.95 % 92.88 % 77.04 % 0.03 s GPU @ 2.5 Ghz (Python + C/C++)
83 3DOP
This method uses stereo information.
 code 86.93 % 91.31 % 76.72 % 3s GPU @ 2.5 Ghz (Matlab + C/C++)
X. Chen, K. Kundu, Y. Zhu, A. Berneshawi, H. Ma, S. Fidler and R. Urtasun: 3D Object Proposals for Accurate Object Class Detection. NIPS 2015.
84 RTM3D code 86.76 % 91.80 % 77.20 % 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.
85 MBR-SSD 86.57 % 90.97 % 78.03 % 4.0 s GPU @ 2.5 Ghz (Python + C/C++)
86 SCANet 86.51 % 92.47 % 81.09 % 0.17 s >8 cores @ 2.5 Ghz (Python)
87 MonoPair 86.11 % 91.65 % 76.45 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
88 DSGN
This method uses stereo information.
 code 86.03 % 95.42 % 78.27 % 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.
89 StereoFENet
This method uses stereo information.
 85.14 % 91.28 % 76.80 % 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.
90 PL++ (SDN+GDC)
This method uses stereo information.
This method makes use of Velodyne laser scans.
 84.42 % 94.83 % 76.95 % 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.
91 SS3D 84.38 % 92.57 % 69.82 % 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.
92 MonoFENet 84.09 % 91.42 % 75.93 % 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.
93 SECA 83.99 % 92.34 % 78.85 % 1 s GPU @ 2.5 Ghz (Python)
94 Complexer-YOLO
This method makes use of Velodyne laser scans.
 83.89 % 91.77 % 79.24 % 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.
95 ZoomNet
This method uses stereo information.
 code 83.79 % 94.14 % 68.78 % 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.
96 M3D-RPN code 82.81 % 88.38 % 67.08 % 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 .
97 RAR-Net 82.63 % 88.40 % 66.90 % 0.5 s 1 core @ 2.5 Ghz (C/C++)
98 ASOD 82.13 % 93.56 % 67.32 % 0.28 s GPU @ 2.5 Ghz (Python)
99 D4LCN code 82.08 % 90.01 % 63.98 % 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.
100 Pseudo-LiDAR++
This method uses stereo information.
 code 81.87 % 94.14 % 74.29 % 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.
101 PG-MonoNet 81.77 % 87.61 % 66.06 % 0.19 s GPU @ 2.5 Ghz (Python)
102 Disp R-CNN
This method uses stereo information.
 81.61 % 92.91 % 69.20 % 0.42 s GPU @ 2.5 Ghz (Python + C/C++)
103 Pseudo-LiDAR E2E
This method uses stereo information.
 81.56 % 93.74 % 74.23 % 0.4 s GPU @ 2.5 Ghz (Python)
104 HR-SECOND code 81.23 % 88.32 % 74.89 % 0.11 s 1 core @ 2.5 Ghz (Python + C/C++)
105 BS3D 81.22 % 94.66 % 68.39 % 22 ms Titan Xp
N. Gählert, J. Wan, M. Weber, J. Zöllner, U. Franke and J. Denzler: Beyond Bounding Boxes: Using Bounding Shapes for Real-Time 3D Vehicle Detection from Monocular RGB Images. 2019 IEEE Intelligent Vehicles Symposium (IV) 2019.
106 FRCNN+Or code 80.57 % 91.50 % 67.49 % 0.09 s Titan Xp GPU
C. Guindel, D. Martin and J. Armingol: Fast Joint Object Detection and Viewpoint Estimation for Traffic Scene Understanding. IEEE Intelligent Transportation Systems Magazine 2018.
C. Guindel, D. Martin and J. Armingol: Joint Object Detection and Viewpoint Estimation using CNN features. IEEE International Conference on Vehicular Electronics and Safety (ICVES) 2017.
107 Manhnet 79.97 % 88.71 % 63.47 % 26 ms 1 core @ 2.5 Ghz (C/C++)
108 3D GCK 78.44 % 88.59 % 66.28 % 24 ms Tesla V100
109 3D-SSMFCNN code 77.82 % 77.84 % 68.67 % 0.1 s GPU @ 1.5 Ghz (C/C++)
L. Novak: Vehicle Detection and Pose Estimation for Autonomous Driving. 2017.
110 3DVP code 75.71 % 84.44 % 64.41 % 40 s 8 cores @ 3.5 Ghz (Matlab + C/C++)
Y. Xiang, W. Choi, Y. Lin and S. Savarese: Data-Driven 3D Voxel Patterns for Object Category Recognition. IEEE Conference on Computer Vision and Pattern Recognition 2015.
111 GS3D 75.63 % 85.79 % 61.85 % 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.
112 Pose-RCNN 75.41 % 89.49 % 63.57 % 2 s >8 cores @ 2.5 Ghz (Python)
M. Braun, Q. Rao, Y. Wang and F. Flohr: Pose-RCNN: Joint object detection and pose estimation using 3D object proposals. Intelligent Transportation Systems (ITSC), 2016 IEEE 19th International Conference on 2016.
113 avodC 75.35 % 86.76 % 70.17 % 0.1 s GPU @ 2.5 Ghz (Python)
114 SubCat code 75.26 % 83.31 % 59.55 % 0.7 s 6 cores @ 3.5 Ghz (Matlab + C/C++)
E. Ohn-Bar and M. Trivedi: Learning to Detect Vehicles by Clustering Appearance Patterns. T-ITS 2015.
115 3D FCN
This method makes use of Velodyne laser scans.
 74.54 % 86.65 % 67.73 % >5 s 1 core @ 2.5 Ghz (C/C++)
B. Li: 3D Fully Convolutional Network for Vehicle Detection in Point Cloud. IROS 2017.
116 OC Stereo
This method uses stereo information.
 73.34 % 86.86 % 61.37 % 0.35 s 1 core @ 2.5 Ghz (Python + C/C++)
117 BdCost+DA+BB+MS 72.87 % 84.39 % 57.07 % TBD s 4 cores @ 2.5 Ghz (Matlab + C/C++)
118 BdCost+DA+MS 72.65 % 84.06 % 58.08 % TBD s 4 cores @ 2.5 Ghz (Matlab/C++)
119 BdCost+DA+BB 70.07 % 84.66 % 55.50 % TBD s 4 cores @ 2.5 Ghz (Matlab + C/C++)
120 ROI-10D 68.14 % 75.32 % 58.98 % 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.
121 multi-task CNN 67.51 % 79.00 % 58.80 % 25.1 ms GPU @ 2.0 Ghz (Python)
M. Oeljeklaus, F. Hoffmann and T. Bertram: A Fast Multi-Task CNN for Spatial Understanding of Traffic Scenes. IEEE Intelligent Transportation Systems Conference 2018.
122 Decoupled-3D v2 67.47 % 88.23 % 54.04 % 0.08 s GPU @ 2.5 Ghz (C/C++)
123 Decoupled-3D 67.23 % 87.34 % 53.84 % 0.08 s GPU @ 2.5 Ghz (C/C++)
124 BdCost48LDCF code 65.50 % 80.44 % 51.24 % 0.5 s 8 cores @ 3.5 Ghz (Matlab + C/C++)
A. Fernández-Baldera, J. Buenaposada and L. Baumela: BAdaCost: Multi-class Boosting with Costs . Pattern Recognition 2018.
125 OC-DPM 65.32 % 77.35 % 51.00 % 10 s 8 cores @ 2.5 Ghz (Matlab)
B. Pepik, M. Stark, P. Gehler and B. Schiele: Occlusion Patterns for Object Class Detection. IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2013.
126 deprecated 65.30 % 69.02 % 63.66 % 0.05 s GPU @ >3.5 Ghz (Python)
127 3DVSSD 65.28 % 79.56 % 55.73 % 0.06 s 1 core @ 2.5 Ghz (C/C++)
128 RefinedMPL 64.02 % 87.95 % 52.06 % 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.
129 BdCost48-25C 63.90 % 80.69 % 51.54 % 4 s 1 core @ 2.5 Ghz (C/C++)
130 DPM-VOC+VP 63.58 % 79.09 % 46.59 % 8 s 1 core @ 2.5 Ghz (C/C++)
B. Pepik, M. Stark, P. Gehler and B. Schiele: Multi-view and 3D Deformable Part Models. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015.
131 AOG-View 62.62 % 77.62 % 48.27 % 3 s 1 core @ 2.5 Ghz (Matlab, C/C++)
B. Li, T. Wu and S. Zhu: Integrating Context and Occlusion for Car Detection by Hierarchical And-Or Model. ECCV 2014.
132 monoref3d 58.30 % 77.65 % 46.90 % 0.1 s 1 core @ 2.5 Ghz (Python)
133 ref3D 58.30 % 77.65 % 46.90 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
134 LSVM-MDPM-sv 57.48 % 70.23 % 42.54 % 10 s 4 cores @ 3.0 Ghz (C/C++)
P. Felzenszwalb, R. Girshick, D. McAllester and D. Ramanan: Object Detection with Discriminatively Trained Part-Based Models. PAMI 2010.
A. Geiger, C. Wojek and R. Urtasun: Joint 3D Estimation of Objects and Scene Layout. NIPS 2011.
135 SAMME48LDCF code 57.26 % 76.28 % 43.55 % 0.5 s 8 cores @ 3.5 Ghz (Matlab + C/C++)
A. Fernández-Baldera, J. Buenaposada and L. Baumela: BAdaCost: Multi-class Boosting with Costs . Pattern Recognition 2018.
136 3D-CVF
This method makes use of Velodyne laser scans.
 57.01 % 62.54 % 54.94 % 0.05 s GPU @ >3.5 Ghz (Python)
137 ref3D 56.49 % 77.52 % 45.17 % 0.1 s 1 core @ 2.5 Ghz (Python)
138 DEFT 51.66 % 57.41 % 50.02 % 1 s GPU @ 2.5 Ghz (Python)
139 VeloFCN
This method makes use of Velodyne laser scans.
 51.05 % 70.03 % 44.82 % 1 s GPU @ 2.5 Ghz (Python + C/C++)
B. Li, T. Zhang and T. Xia: Vehicle Detection from 3D Lidar Using Fully Convolutional Network. RSS 2016 .
140 Mono3D_PLiDAR code 49.39 % 76.90 % 41.13 % 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.
141 DPM-C8B1
This method uses stereo information.
 48.00 % 57.76 % 35.52 % 15 s 4 cores @ 2.5 Ghz (Matlab + C/C++)
J. Yebes, L. Bergasa and M. García-Garrido: Visual Object Recognition with 3D-Aware Features in KITTI Urban Scenes. Sensors 2015.
J. Yebes, L. Bergasa, R. Arroyo and A. Lázaro: Supervised learning and evaluation of KITTI's cars detector with DPM. IV 2014.
142 LTN 46.54 % 48.96 % 41.58 % 0.4 s GPU @ >3.5 Ghz (Python)
T. Wang, X. He, Y. Cai and G. Xiao: Learning a Layout Transfer Network for Context Aware Object Detection. IEEE Transactions on Intelligent Transportation Systems 2019.
143 sensekitti code 46.12 % 49.16 % 42.79 % 4.5 s GPU @ 2.5 Ghz (Python + C/C++)
B. Yang, J. Yan, Z. Lei and S. Li: Craft Objects from Images. CVPR 2016.
144 ReSqueeze 45.58 % 49.08 % 41.33 % 0.03 s GPU @ >3.5 Ghz (Python)
145 Resnet101Faster rcnn 44.01 % 51.21 % 39.19 % 1 s 1 core @ 2.5 Ghz (Python)
146 FD2 38.89 % 48.29 % 34.35 % 0.01 s GPU @ >3.5 Ghz (Python + C/C++)
147 bin 38.58 % 43.36 % 32.42 % 15ms s GPU @ >3.5 Ghz (Python)
148 DGIST-CellBox 38.36 % 39.11 % 36.15 % 0.1 s GPU @ 2.5 Ghz (Java + C/C++)
149 SA-SSD 38.30 % 39.40 % 37.07 % 0.04 s 1 core @ 2.5 Ghz (Python)
150 Faster RCNN + A 37.92 % 39.50 % 33.85 % 0.19 s GPU @ 2.5 Ghz (Python + C/C++)
151 JSU-NET 37.60 % 41.33 % 33.41 % 0.1 s 1 core @ 2.5 Ghz (Python)
152 Faster RCNN + G 37.49 % 39.05 % 33.40 % 1.1 s GPU @ 2.5 Ghz (Python)
153 Faster RCNN + A 37.35 % 38.75 % 33.38 % 0.19 s GPU @ 2.5 Ghz (Python)
154 Point-GNN
This method makes use of Velodyne laser scans.
 37.20 % 38.66 % 36.29 % 0.6 s GPU @ 2.5 Ghz (Python)
155 GAFM 37.08 % 40.28 % 33.08 % 0.5 s 1 core @ 2.5 Ghz (Python)
156 CRCNNA 37.04 % 40.19 % 32.03 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
157 Faster RCNN + Gr + A 36.95 % 38.22 % 33.16 % 1.29 s GPU @ 2.5 Ghz (Python)
158 CSFADet 36.83 % 39.76 % 32.73 % 0.05 s GPU @ 2.5 Ghz (Python)
159 cas_retina 36.63 % 39.70 % 31.52 % 0.2 s 4 cores @ 2.5 Ghz (Python)
160 GA_BALANCE 36.62 % 38.44 % 31.94 % 1 s 1 core @ 2.5 Ghz (Python)
161 GA_rpn500 36.54 % 38.33 % 32.67 % 1 s 1 core @ 2.5 Ghz (Python)
162 GA2500 36.54 % 38.33 % 32.67 % 0.2 s 1 core @ 2.5 Ghz (Python)
163 cas+res+soft 36.53 % 38.82 % 32.26 % 0.2 s 4 cores @ 2.5 Ghz (Python)
164 merge12-12 36.47 % 38.83 % 32.20 % 0.2 s 4 cores @ 2.5 Ghz (Python)
165 GA_FULLDATA 36.43 % 38.90 % 31.61 % 1 s 4 cores @ 2.5 Ghz (Python)
166 AtrousDet 36.36 % 38.86 % 31.79 % 0.05 s TITAN X
167 bigger_ga 36.21 % 38.41 % 31.58 % 1 s 1 core @ 2.5 Ghz (Python)
168 cas_retina_1_13 35.89 % 39.02 % 31.33 % 0.03 s 4 cores @ 2.5 Ghz (Python)
169 cascadercnn 35.61 % 36.22 % 30.16 % 0.36 s 4 cores @ 2.5 Ghz (Python)
170 Cmerge 35.02 % 38.33 % 29.06 % 0.2 s 4 cores @ 2.5 Ghz (Python)
171 BirdNet
This method makes use of Velodyne laser scans.
 35.00 % 50.34 % 33.40 % 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.
172 ga50 34.95 % 38.21 % 30.29 % 1 s 1 core @ 2.5 Ghz (Python)
173 softretina 34.57 % 39.31 % 29.27 % 0.16 s 4 cores @ 2.5 Ghz (Python)
174 Retinanet100 34.37 % 39.15 % 28.43 % 0.2 s 4 cores @ 2.5 Ghz (Python)
175 ZKNet 34.27 % 38.09 % 29.93 % 0.01 s GPU @ 2.0 Ghz (Python)
176 LPN 33.61 % 34.57 % 29.72 % 0.2 s GPU @ 2.5 Ghz (Python + C/C++)
177 cascade_gw 33.53 % 34.76 % 29.71 % 0.2 s 4 cores @ 2.5 Ghz (Python)
178 RADNet-Fusion
This method makes use of Velodyne laser scans.
 33.31 % 31.96 % 32.72 % 0.1 s 1 core @ 2.5 Ghz (Python)
179 RADNet-LIDAR
This method makes use of Velodyne laser scans.
 33.08 % 31.30 % 32.31 % 0.1 s 1 core @ 2.5 Ghz (Python)
180 NM code 32.78 % 37.21 % 28.36 % 0.01 s GPU @ 2.5 Ghz (Python)
181 SceneNet 32.78 % 37.79 % 28.30 % 0.03 s GPU @ 2.5 Ghz (C/C++)
182 MTDP 32.68 % 36.06 % 27.12 % 0.15 s GPU @ 2.0 Ghz (Python)
183 CBNet 32.63 % 36.51 % 29.26 % 1 s 4 cores @ 2.5 Ghz (Python)
184 Fast-SSD 32.51 % 41.41 % 28.45 % 0.06 s GTX650Ti
185 centernet 32.22 % 35.79 % 28.50 % 0.01 s GPU @ 2.5 Ghz (Python)
186 RTL3D 32.16 % 33.73 % 30.58 % 0.02 s GPU @ 2.5 Ghz (Python)
187 RFCN_RFB 32.06 % 35.39 % 27.94 % 0.2 s 4 cores @ 2.5 Ghz (Python)
188 dgist_multiDetNet 32.01 % 38.16 % 28.72 % 0.05 s 1 core @ 2.5 Ghz (C/C++)
189 FailNet-Fusion
This method makes use of Velodyne laser scans.
 31.68 % 30.84 % 30.56 % 0.1 s 1 core @ 2.5 Ghz (Python)
190 yolo800 31.13 % 32.49 % 26.76 % 0.13 s 4 cores @ 2.5 Ghz (Python)
191 FailNet-LIDAR
This method makes use of Velodyne laser scans.
 31.10 % 30.32 % 29.89 % 0.1 s 1 core @ 2.5 Ghz (Python)
192 VoxelNet(Unofficial) 31.08 % 34.54 % 28.79 % 0.5 s GPU @ 2.0 Ghz (Python)
193 SAIC-SA-3D
This method makes use of Velodyne laser scans.
 31.02 % 41.38 % 29.60 % 0.05 s GPU @ 2.5 Ghz (Python)
194 RFCN 30.93 % 34.24 % 25.27 % 0.2 s 4 cores @ 2.5 Ghz (Python)
195 AOG code 29.81 % 33.28 % 23.91 % 3 s 4 cores @ 2.5 Ghz (Matlab)
T. Wu, B. Li and S. Zhu: Learning And-Or Models to Represent Context and Occlusion for Car Detection and Viewpoint Estimation. TPAMI 2016.
B. Li, T. Wu and S. Zhu: Integrating Context and Occlusion for Car Detection by Hierarchical And-Or Model. ECCV 2014.
196 m-prcnn
This method uses stereo information.
 29.62 % 34.80 % 22.79 % 0.43 s 1 core @ 2.5 Ghz (Python)
197 Multi-task DG 29.49 % 36.06 % 26.06 % 0.06 s GPU @ 2.5 Ghz (Python)
198 fasterrcnn 28.42 % 30.28 % 24.95 % 0.2 s 4 cores @ 2.5 Ghz (Python)
199 RFBnet 27.91 % 34.44 % 25.24 % 0.2 s 4 cores @ 2.5 Ghz (Python)
200 E-VoxelNet 26.87 % 27.66 % 24.05 % 0.1 s GPU @ 2.5 Ghz (Python)
201 SubCat48LDCF code 26.68 % 34.33 % 19.44 % 0.5 s 8 cores @ 3.5 Ghz (Matlab + C/C++)
A. Fernández-Baldera, J. Buenaposada and L. Baumela: BAdaCost: Multi-class Boosting with Costs . Pattern Recognition 2018.
202 Lidar_ROI+Yolo(UJS) 25.33 % 30.36 % 22.20 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
203 RADNet-Mono 24.78 % 28.55 % 22.84 % 0.1 s 1 core @ 2.5 Ghz (Python)
204 100Frcnn 23.32 % 32.81 % 19.45 % 2 s 4 cores @ 2.5 Ghz (Python + C/C++)
205 RT3DStereo
This method uses stereo information.
 21.41 % 25.58 % 17.52 % 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.
206 CSoR
This method makes use of Velodyne laser scans.
 code 20.82 % 30.65 % 17.14 % 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.
207 FailNet-Mono 19.63 % 25.13 % 17.19 % 0.1 s 1 core @ 2.5 Ghz (Python)
208 RT3D
This method makes use of Velodyne laser scans.
 18.96 % 24.41 % 19.85 % 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.
209 softyolo 18.31 % 26.80 % 15.28 % 0.16 s 4 cores @ 2.5 Ghz (Python)
210 Licar
This method makes use of Velodyne laser scans.
 16.16 % 18.56 % 15.59 % 0.09 s GPU @ 2.0 Ghz (Python)
211 VoxelJones code 15.41 % 17.83 % 14.13 % .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.
212 KD53-20 13.76 % 20.58 % 11.91 % 0.19 s 4 cores @ 2.5 Ghz (Python)
213 MRF 1.75 % 0.63 % 2.14 % 0.05 s GPU @ 1.5 Ghz (Python + C/C++)
214 MP 1.51 % 0.63 % 2.03 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
215 PiP 1.45 % 0.56 % 1.85 % 0.05 s 1 core @ 2.5 Ghz (Python)
216 SPA 1.25 % 0.59 % 1.64 % 0.1 s 1 core @ 2.5 Ghz (Python)
217 Associate-3Ddet
This method makes use of Velodyne laser scans.
 1.20 % 0.52 % 1.38 % 0.05 s 1 core @ 2.5 Ghz (Python + C/C++)
218 FCPP 0.06 % 0.00 % 0.07 % 0.02 s 1 core @ 2.0 Ghz (Python + C/C++)
219 JSyolo 0.00 % 0.00 % 0.00 % 0.16 s 4 cores @ 2.5 Ghz (Python)
Table as LaTeX | Only published Methods


Pedestrians


Method Setting Code Moderate Easy Hard Runtime Environment
1 THICV-YDM 69.07 % 83.00 % 62.54 % 0.06 s GPU @ 2.5 Ghz (Python)
2 VMVS
This method makes use of Velodyne laser scans.
 68.19 % 79.98 % 63.18 % 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.
3 Sogo_MM 67.31 % 80.02 % 61.99 % 1.5 s GPU @ 2.5 Ghz (C/C++)
4 SubCNN 66.70 % 79.65 % 61.35 % 2 s GPU @ 3.5 Ghz (Python + C/C++)
Y. Xiang, W. Choi, Y. Lin and S. Savarese: Subcategory-aware Convolutional Neural Networks for Object Proposals and Detection. IEEE Winter Conference on Applications of Computer Vision (WACV) 2017.
5 F-ConvNet
This method makes use of Velodyne laser scans.
 code 63.87 % 75.19 % 58.57 % 0.47 s GPU @ 2.5 Ghz (Python + C/C++)
Z. Wang and K. Jia: Frustum ConvNet: Sliding Frustums to Aggregate Local Point-Wise Features for Amodal 3D Object Detection. IROS 2019.
6 3DOP
This method uses stereo information.
 code 61.48 % 74.22 % 55.89 % 3s GPU @ 2.5 Ghz (Matlab + C/C++)
X. Chen, K. Kundu, Y. Zhu, A. Berneshawi, H. Ma, S. Fidler and R. Urtasun: 3D Object Proposals for Accurate Object Class Detection. NIPS 2015.
7 DeepStereoOP 60.15 % 73.76 % 55.30 % 3.4 s GPU @ 3.5 Ghz (Matlab + C/C++)
C. Pham and J. Jeon: Robust Object Proposals Re-ranking for Object Detection in Autonomous Driving Using Convolutional Neural Networks. Signal Processing: Image Communiation 2017.
8 Pose-RCNN 59.84 % 76.24 % 53.59 % 2 s >8 cores @ 2.5 Ghz (Python)
M. Braun, Q. Rao, Y. Wang and F. Flohr: Pose-RCNN: Joint object detection and pose estimation using 3D object proposals. Intelligent Transportation Systems (ITSC), 2016 IEEE 19th International Conference on 2016.
9 FFNet code 58.87 % 69.24 % 53.75 % 1.07 s GPU @ 1.5 Ghz (Python)
C. Zhao, Y. Qian and M. Yang: Monocular Pedestrian Orientation Estimation Based on Deep 2D-3D Feedforward. Pattern Recognition 2019.
10 Mono3D code 58.66 % 71.19 % 53.94 % 4.2 s GPU @ 2.5 Ghz (Matlab + C/C++)
X. Chen, K. Kundu, Z. Zhang, H. Ma, S. Fidler and R. Urtasun: Monocular 3D Object Detection for Autonomous Driving. CVPR 2016.
11 OHS-Direct 58.45 % 67.82 % 55.33 % 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.
12 OHS-Dense 56.89 % 66.97 % 52.75 % 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.
13 MonoPSR code 54.65 % 68.98 % 50.07 % 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.
14 MMLab-PartA^2
This method makes use of Velodyne laser scans.
 52.20 % 63.51 % 48.27 % 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.
15 FRCNN+Or code 52.15 % 67.03 % 47.14 % 0.09 s Titan Xp GPU
C. Guindel, D. Martin and J. Armingol: Fast Joint Object Detection and Viewpoint Estimation for Traffic Scene Understanding. IEEE Intelligent Transportation Systems Magazine 2018.
C. Guindel, D. Martin and J. Armingol: Joint Object Detection and Viewpoint Estimation using CNN features. IEEE International Conference on Vehicular Electronics and Safety (ICVES) 2017.
16 PointPainting
This method makes use of Velodyne laser scans.
 50.22 % 59.25 % 46.95 % 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 ARPNET 48.49 % 60.47 % 45.02 % 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.
18 PointPillars
This method makes use of Velodyne laser scans.
 code 48.05 % 57.47 % 45.40 % 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 PPFNet code 47.73 % 55.78 % 44.56 % 0.1 s 1 core @ 2.5 Ghz (Python + C/C++)
20 MMLab-PointRCNN
This method makes use of Velodyne laser scans.
 code 47.33 % 57.19 % 44.31 % 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.
21 Shift R-CNN (mono) code 46.56 % 64.73 % 41.86 % 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.
22 VOXEL_FPN_HR 45.65 % 56.17 % 42.10 % 0.12 s 8 cores @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
23 SS3D_HW 44.43 % 59.56 % 38.77 % 0.4 s GPU @ 2.5 Ghz (Python)
24 FOFNet
This method makes use of Velodyne laser scans.
 44.33 % 55.61 % 40.85 % 0.04 s GPU @ 2.5 Ghz (Python)
25 AVOD-FPN
This method makes use of Velodyne laser scans.
 code 43.99 % 53.48 % 41.56 % 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.
26 DGIST-CellBox 43.86 % 48.68 % 41.52 % 0.1 s GPU @ 2.5 Ghz (Java + C/C++)
27 DDB
This method makes use of Velodyne laser scans.
 43.21 % 52.02 % 40.81 % 0.05 s GPU @ 2.5 Ghz (Python)
28 PiP 42.76 % 51.23 % 40.06 % 0.05 s 1 core @ 2.5 Ghz (Python)
29 HWFD 42.45 % 48.31 % 39.66 % 0.21 s one 1080Ti
30 MonoPair 42.38 % 55.26 % 38.53 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
31 SCANet 42.12 % 54.48 % 38.64 % 0.17 s >8 cores @ 2.5 Ghz (Python)
32 Faster RCNN + Gr + A 40.92 % 47.81 % 37.89 % 1.29 s GPU @ 2.5 Ghz (Python)
33 HR-SECOND code 40.81 % 51.12 % 37.48 % 0.11 s 1 core @ 2.5 Ghz (Python + C/C++)
34 Faster RCNN + G 40.49 % 47.16 % 37.57 % 1.1 s GPU @ 2.5 Ghz (Python)
35 Faster RCNN + A 39.95 % 47.52 % 37.08 % 0.19 s GPU @ 2.5 Ghz (Python)
36 CentrNet-v1
This method makes use of Velodyne laser scans.
 39.83 % 46.21 % 38.05 % 0.03 s GPU @ 2.5 Ghz (Python)
37 AB3DMOT
This method makes use of Velodyne laser scans.
This is an online method (no batch processing).
 code 39.76 % 50.30 % 36.90 % 0.0047s 1 core @ 2.5 Ghz (python)
X. Weng and K. Kitani: A Baseline for 3D Multi-Object Tracking. arXiv:1907.03961 2019.
38 SS3D 39.60 % 53.72 % 35.40 % 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.
39 Faster RCNN + A 39.44 % 46.80 % 36.46 % 0.19 s GPU @ 2.5 Ghz (Python + C/C++)
40 RTM3D code 38.75 % 54.73 % 34.52 % 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.
41 CSFADet 38.41 % 46.75 % 35.44 % 0.05 s GPU @ 2.5 Ghz (Python)
42 DPM-VOC+VP 37.79 % 52.91 % 33.27 % 8 s 1 core @ 2.5 Ghz (C/C++)
B. Pepik, M. Stark, P. Gehler and B. Schiele: Multi-view and 3D Deformable Part Models. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015.
43 Tencent_ADlab_Lidar
This method makes use of Velodyne laser scans.
 37.23 % 44.01 % 35.54 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
44 dgist_multiDetNet 37.10 % 45.90 % 33.80 % 0.05 s 1 core @ 2.5 Ghz (C/C++)
45 A-VoxelNet 36.24 % 42.48 % 34.36 % 0.029 s GPU @ 2.5 Ghz (Python)
46 PP-3D 36.22 % 44.49 % 34.19 % 0.1 s 1 core @ 2.5 Ghz (Python)
47 TANet code 36.21 % 42.54 % 34.39 % 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.
48 SAANet 36.08 % 46.09 % 34.14 % 0.10 s 1 core @ 2.5 Ghz (Python)
49 AtrousDet 35.85 % 44.79 % 32.12 % 0.05 s TITAN X
50 SCNet
This method makes use of Velodyne laser scans.
 35.49 % 44.50 % 33.38 % 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.
51 CRCNNA 34.88 % 43.18 % 31.90 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
52 sensekitti code 34.26 % 41.03 % 31.51 % 4.5 s GPU @ 2.5 Ghz (Python + C/C++)
B. Yang, J. Yan, Z. Lei and S. Li: Craft Objects from Images. CVPR 2016.
53 merge12-12 34.10 % 43.60 % 30.43 % 0.2 s 4 cores @ 2.5 Ghz (Python)
54 cas+res+soft 34.01 % 43.51 % 30.28 % 0.2 s 4 cores @ 2.5 Ghz (Python)
55 cas_retina 33.98 % 43.80 % 31.12 % 0.2 s 4 cores @ 2.5 Ghz (Python)
56 cas_retina_1_13 33.87 % 43.55 % 30.99 % 0.03 s 4 cores @ 2.5 Ghz (Python)
57 D4LCN code 33.62 % 46.73 % 28.71 % 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.
58 JSU-NET 33.55 % 45.79 % 30.72 % 0.1 s 1 core @ 2.5 Ghz (Python)
59 SparsePool code 33.35 % 43.86 % 29.99 % 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 SparsePool code 33.29 % 43.52 % 30.01 % 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 LSVM-MDPM-sv 33.01 % 45.60 % 29.27 % 10 s 4 cores @ 3.0 Ghz (C/C++)
P. Felzenszwalb, R. Girshick, D. McAllester and D. Ramanan: Object Detection with Discriminatively Trained Part-Based Models. PAMI 2010.
A. Geiger, C. Wojek and R. Urtasun: Joint 3D Estimation of Objects and Scene Layout. NIPS 2011.
62 PG-MonoNet 32.67 % 44.75 % 29.33 % 0.19 s GPU @ 2.5 Ghz (Python)
63 cascadercnn 32.59 % 43.37 % 29.73 % 0.36 s 4 cores @ 2.5 Ghz (Python)
64 ReSqueeze 32.47 % 38.49 % 30.04 % 0.03 s GPU @ >3.5 Ghz (Python)
65 AVOD
This method makes use of Velodyne laser scans.
 code 32.19 % 42.54 % 29.09 % 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.
66 Complexer-YOLO
This method makes use of Velodyne laser scans.
 32.13 % 37.32 % 28.94 % 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.
67 yolo800 32.12 % 40.53 % 28.83 % 0.13 s 4 cores @ 2.5 Ghz (Python)
68 RPN+BF code 32.12 % 41.19 % 28.83 % 0.6 s GPU @ 2.5 Ghz (Matlab + C/C++)
L. Zhang, L. Lin, X. Liang and K. He: Is Faster R-CNN Doing Well for Pedestrian Detection?. ECCV 2016.
69 bin 31.94 % 36.94 % 29.50 % 15ms s GPU @ >3.5 Ghz (Python)
70 M3D-RPN code 31.88 % 44.33 % 28.55 % 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 .
71 Point-GNN
This method makes use of Velodyne laser scans.
 31.86 % 39.16 % 29.65 % 0.6 s GPU @ 2.5 Ghz (Python)
72 SubCat 31.26 % 42.31 % 27.39 % 1.2 s 6 cores @ 2.5 Ghz (Matlab + C/C++)
E. Ohn-Bar and M. Trivedi: Fast and Robust Object Detection Using Visual Subcategories. Computer Vision and Pattern Recognition Workshops Mobile Vision 2014.
73 ZKNet 31.21 % 39.55 % 28.61 % 0.01 s GPU @ 2.0 Ghz (Python)
74 RFCN 30.97 % 40.51 % 27.45 % 0.2 s 4 cores @ 2.5 Ghz (Python)
75 LPN 30.84 % 38.60 % 28.49 % 0.2 s GPU @ 2.5 Ghz (Python + C/C++)
76 CHTTL MMF 30.45 % 41.08 % 27.57 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
77 RFCN_RFB 29.91 % 38.71 % 26.50 % 0.2 s 4 cores @ 2.5 Ghz (Python)
78 deprecated 29.74 % 37.71 % 27.25 % 0.05 s GPU @ 2.0 Ghz (Python)
79 NM code 29.60 % 38.81 % 26.99 % 0.01 s GPU @ 2.5 Ghz (Python)
80 fasterrcnn 29.48 % 38.63 % 26.89 % 0.2 s 4 cores @ 2.5 Ghz (Python)
81 FCY
This method makes use of Velodyne laser scans.
 29.07 % 40.31 % 25.89 % 0.02 s GPU @ 2.5 Ghz (Python)
82 Multi-task DG 28.71 % 38.97 % 26.13 % 0.06 s GPU @ 2.5 Ghz (Python)
83 FD2 28.40 % 35.59 % 25.75 % 0.01 s GPU @ >3.5 Ghz (Python + C/C++)
84 MTDP 28.24 % 37.49 % 25.57 % 0.15 s GPU @ 2.0 Ghz (Python)
85 centernet 27.53 % 37.41 % 24.35 % 0.01 s GPU @ 2.5 Ghz (Python)
86 cascade_gw 26.32 % 36.41 % 23.73 % 0.2 s 4 cores @ 2.5 Ghz (Python)
87 Cmerge 25.09 % 34.53 % 22.43 % 0.2 s 4 cores @ 2.5 Ghz (Python)
88 DSGN
This method uses stereo information.
 code 24.32 % 31.21 % 23.09 % 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.
89 ACF 24.31 % 32.23 % 21.70 % 1 s 1 core @ 3.5 Ghz (Matlab + C/C++)
P. Doll\'ar, R. Appel, S. Belongie and P. Perona: Fast Feature Pyramids for Object Detection. PAMI 2014.
90 Resnet101Faster rcnn 23.70 % 30.19 % 21.55 % 1 s 1 core @ 2.5 Ghz (Python)
91 multi-task CNN 22.80 % 30.30 % 20.47 % 25.1 ms GPU @ 2.0 Ghz (Python)
M. Oeljeklaus, F. Hoffmann and T. Bertram: A Fast Multi-Task CNN for Spatial Understanding of Traffic Scenes. IEEE Intelligent Transportation Systems Conference 2018.
92 ACF-MR 22.61 % 29.23 % 20.08 % 0.6 s 1 core @ 3.5 Ghz (C/C++)
R. Rajaram, E. Ohn-Bar and M. Trivedi: Looking at Pedestrians at Different Scales: A Multi-resolution Approach and Evaluations. T-ITS 2016.
93 OC Stereo
This method uses stereo information.
 22.02 % 31.36 % 20.20 % 0.35 s 1 core @ 2.5 Ghz (Python + C/C++)
94 Retinanet100 21.71 % 29.72 % 19.12 % 0.2 s 4 cores @ 2.5 Ghz (Python)
95 softyolo 21.56 % 30.46 % 20.01 % 0.16 s 4 cores @ 2.5 Ghz (Python)
96 Lidar_ROI+Yolo(UJS) 19.43 % 26.83 % 17.14 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
97 KD53-20 19.36 % 25.10 % 17.54 % 0.19 s 4 cores @ 2.5 Ghz (Python)
98 DPM-C8B1
This method uses stereo information.
 19.17 % 27.79 % 16.48 % 15 s 4 cores @ 2.5 Ghz (Matlab + C/C++)
J. Yebes, L. Bergasa and M. García-Garrido: Visual Object Recognition with 3D-Aware Features in KITTI Urban Scenes. Sensors 2015.
J. Yebes, L. Bergasa, R. Arroyo and A. Lázaro: Supervised learning and evaluation of KITTI's cars detector with DPM. IV 2014.
99 RefinedMPL 17.26 % 25.83 % 15.41 % 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.
100 BirdNet
This method makes use of Velodyne laser scans.
 16.45 % 21.07 % 15.65 % 0.11 s Titan Xp GPU
J. Beltrán, C. Guindel, F. Moreno, D. Cruzado, F. García and A. Escalera: BirdNet: A 3D Object Detection Framework from LiDAR Information. 2018 21st International Conference on Intelligent Transportation Systems (ITSC) 2018.
101 RT3DStereo
This method uses stereo information.
 15.34 % 21.41 % 13.23 % 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.
102 100Frcnn 12.37 % 19.41 % 10.92 % 2 s 4 cores @ 2.5 Ghz (Python + C/C++)
103 MP 5.39 % 6.41 % 5.14 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
104 CBNet 0.72 % 0.56 % 0.75 % 1 s 4 cores @ 2.5 Ghz (Python)
105 softretina 0.13 % 0.10 % 0.14 % 0.16 s 4 cores @ 2.5 Ghz (Python)
106 JSyolo 0.06 % 0.11 % 0.07 % 0.16 s 4 cores @ 2.5 Ghz (Python)
Table as LaTeX | Only published Methods


Cyclists


Method Setting Code Moderate Easy Hard Runtime Environment
1 MMLab PV-RCNN
This method makes use of Velodyne laser scans.
 79.70 % 86.43 % 72.96 % 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 OHS-Dense 78.23 % 85.65 % 71.60 % 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.
3 MMLab-PartA^2
This method makes use of Velodyne laser scans.
 77.52 % 88.70 % 70.41 % 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.
4 PointPainting
This method makes use of Velodyne laser scans.
 76.92 % 87.33 % 68.21 % 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.
5 F-ConvNet
This method makes use of Velodyne laser scans.
 code 76.71 % 86.39 % 66.92 % 0.47 s GPU @ 2.5 Ghz (Python + C/C++)
Z. Wang and K. Jia: Frustum ConvNet: Sliding Frustums to Aggregate Local Point-Wise Features for Amodal 3D Object Detection. IROS 2019.
6 OHS-Direct 75.59 % 83.45 % 68.42 % 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.
7 VOXEL_FPN_HR 74.77 % 87.41 % 68.16 % 0.12 s 8 cores @ 2.5 Ghz (Python)
ERROR: Wrong syntax in BIBTEX file.
8 MMLab-PointRCNN
This method makes use of Velodyne laser scans.
 code 72.81 % 85.94 % 65.84 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
S. Shi, X. Wang and H. Li: Pointrcnn: 3d object proposal generation and detection from point cloud. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition 2019.
9 FOFNet
This method makes use of Velodyne laser scans.
 72.48 % 86.89 % 65.63 % 0.04 s GPU @ 2.5 Ghz (Python)
10 PiP 71.10 % 82.83 % 64.88 % 0.05 s 1 core @ 2.5 Ghz (Python)
11 HR-SECOND code 69.60 % 82.42 % 62.47 % 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 69.54 % 82.18 % 62.98 % 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 ARPNET 68.72 % 82.61 % 62.00 % 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.
14 PointPillars
This method makes use of Velodyne laser scans.
 code 68.55 % 83.79 % 61.71 % 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.
15 TANet code 66.37 % 81.15 % 60.10 % 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 A-VoxelNet 66.17 % 80.73 % 58.96 % 0.029 s GPU @ 2.5 Ghz (Python)
17 Tencent_ADlab_Lidar
This method makes use of Velodyne laser scans.
 65.85 % 81.05 % 59.17 % 0.1 s GPU @ 2.5 Ghz (Python + C/C++)
18 SAANet 65.52 % 82.29 % 58.81 % 0.10 s 1 core @ 2.5 Ghz (Python)
19 Sogo_MM 63.50 % 71.57 % 55.24 % 1.5 s GPU @ 2.5 Ghz (C/C++)
20 SubCNN 63.36 % 71.97 % 55.42 % 2 s GPU @ 3.5 Ghz (Python + C/C++)
Y. Xiang, W. Choi, Y. Lin and S. Savarese: Subcategory-aware Convolutional Neural Networks for Object Proposals and Detection. IEEE Winter Conference on Applications of Computer Vision (WACV) 2017.
21 deprecated 63.08 % 83.73 % 53.51 % 0.05 s GPU @ 2.0 Ghz (Python)
22 CentrNet-v1
This method makes use of Velodyne laser scans.
 62.11 % 78.10 % 55.54 % 0.03 s GPU @ 2.5 Ghz (Python)
23 Pose-RCNN 62.02 % 75.74 % 53.99 % 2 s >8 cores @ 2.5 Ghz (Python)
M. Braun, Q. Rao, Y. Wang and F. Flohr: Pose-RCNN: Joint object detection and pose estimation using 3D object proposals. Intelligent Transportation Systems (ITSC), 2016 IEEE 19th International Conference on 2016.
24 SCNet
This method makes use of Velodyne laser scans.
 61.11 % 77.77 % 54.82 % 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.
25 SCANet 60.84 % 75.16 % 54.70 % 0.17 s >8 cores @ 2.5 Ghz (Python)
26 PP-3D 60.09 % 76.73 % 53.41 % 0.1 s 1 core @ 2.5 Ghz (Python)
27 AVOD-FPN
This method makes use of Velodyne laser scans.
 code 58.70 % 69.21 % 53.47 % 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.
28 DDB
This method makes use of Velodyne laser scans.
 58.65 % 75.36 % 52.85 % 0.05 s GPU @ 2.5 Ghz (Python)
29 Deep3DBox 58.56 % 68.31 % 50.30 % 1.5 s GPU @ 2.5 Ghz (C/C++)
A. Mousavian, D. Anguelov, J. Flynn and J. Kosecka: 3D Bounding Box Estimation Using Deep Learning and Geometry. CVPR 2017.
30 3DOP
This method uses stereo information.
 code 58.45 % 72.24 % 51.91 % 3s GPU @ 2.5 Ghz (Matlab + C/C++)
X. Chen, K. Kundu, Y. Zhu, A. Berneshawi, H. Ma, S. Fidler and R. Urtasun: 3D Object Proposals for Accurate Object Class Detection. NIPS 2015.
31 Complexer-YOLO
This method makes use of Velodyne laser scans.
 58.28 % 65.41 % 54.27 % 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.
32 DeepStereoOP 56.55 % 69.36 % 49.37 % 3.4 s GPU @ 3.5 Ghz (Matlab + C/C++)
C. Pham and J. Jeon: Robust Object Proposals Re-ranking for Object Detection in Autonomous Driving Using Convolutional Neural Networks. Signal Processing: Image Communiation 2017.
33 Mono3D code 53.96 % 67.33 % 47.91 % 4.2 s GPU @ 2.5 Ghz (Matlab + C/C++)
X. Chen, K. Kundu, Z. Zhang, H. Ma, S. Fidler and R. Urtasun: Monocular 3D Object Detection for Autonomous Driving. CVPR 2016.
34 AVOD
This method makes use of Velodyne laser scans.
 code 51.05 % 64.81 % 45.12 % 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.
35 FRCNN+Or code 49.53 % 63.45 % 43.65 % 0.09 s Titan Xp GPU
C. Guindel, D. Martin and J. Armingol: Fast Joint Object Detection and Viewpoint Estimation for Traffic Scene Understanding. IEEE Intelligent Transportation Systems Magazine 2018.
C. Guindel, D. Martin and J. Armingol: Joint Object Detection and Viewpoint Estimation using CNN features. IEEE International Conference on Vehicular Electronics and Safety (ICVES) 2017.
36 MonoPSR code 49.32 % 58.63 % 43.05 % 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.
37 FCY
This method makes use of Velodyne laser scans.
 44.14 % 61.97 % 39.79 % 0.02 s GPU @ 2.5 Ghz (Python)
38 SparsePool code 43.50 % 59.77 % 39.36 % 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 sensekitti code 41.14 % 47.48 % 35.07 % 4.5 s GPU @ 2.5 Ghz (Python + C/C++)
B. Yang, J. Yan, Z. Lei and S. Li: Craft Objects from Images. CVPR 2016.
40 RTM3D code 40.99 % 60.53 % 35.19 % 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.
41 MonoPair 39.47 % 53.36 % 33.95 % 0.06 s GPU @ 2.5 Ghz (Python + C/C++)
42 SS3D_HW 37.68 % 52.40 % 32.33 % 0.4 s GPU @ 2.5 Ghz (Python)
43 Shift R-CNN (mono) code 34.77 % 51.95 % 31.10 % 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.
44 SparsePool code 34.56 % 43.33 % 31.09 % 0.13 s 8 cores @ 2.5 Ghz (Python)
Z. Wang, W. Zhan and M. Tomizuka: Fusing bird view lidar point cloud and front view camera image for deep object detection. arXiv preprint arXiv:1711.06703 2017.
45 PG-MonoNet 34.11 % 49.05 % 28.14 % 0.19 s GPU @ 2.5 Ghz (Python)
46 Point-GNN
This method makes use of Velodyne laser scans.
 32.37 % 36.29 % 29.81 % 0.6 s GPU @ 2.5 Ghz (Python)
47 HWFD 31.75 % 35.47 % 27.72 % 0.21 s one 1080Ti
48 D4LCN code 31.70 % 48.03 % 26.99 % 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.
49 Faster RCNN + Gr + A 31.55 % 36.35 % 28.43 % 1.29 s GPU @ 2.5 Ghz (Python)
50 M3D-RPN code 31.09 % 48.11 % 26.10 % 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 .
51 Faster RCNN + A 30.81 % 36.25 % 27.51 % 0.19 s GPU @ 2.5 Ghz (Python)
52 Faster RCNN + G 30.61 % 36.19 % 27.22 % 1.1 s GPU @ 2.5 Ghz (Python)
53 DGIST-CellBox 30.34 % 35.69 % 27.10 % 0.1 s GPU @ 2.5 Ghz (Java + C/C++)
54 Faster RCNN + A 30.12 % 36.03 % 26.98 % 0.19 s GPU @ 2.5 Ghz (Python + C/C++)
55 BirdNet
This method makes use of Velodyne laser scans.
 29.65 % 41.68 % 27.21 % 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.
56 bin 29.63 % 35.40 % 25.98 % 15ms s GPU @ >3.5 Ghz (Python)
57 AtrousDet 28.26 % 34.10 % 24.69 % 0.05 s TITAN X
58 SS3D 27.79 % 42.95 % 24.26 % 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.
59 ReSqueeze 27.40 % 36.26 % 24.04 % 0.03 s GPU @ >3.5 Ghz (Python)
60 cascadercnn 26.59 % 33.81 % 23.48 % 0.36 s 4 cores @ 2.5 Ghz (Python)
61 merge12-12 26.39 % 33.49 % 22.83 % 0.2 s 4 cores @ 2.5 Ghz (Python)
62 cas+res+soft 26.32 % 33.63 % 22.75 % 0.2 s 4 cores @ 2.5 Ghz (Python)
63 GA2500 26.08 % 32.91 % 22.06 % 0.2 s 1 core @ 2.5 Ghz (Python)
64 GA_rpn500 26.08 % 32.91 % 22.06 % 1 s 1 core @ 2.5 Ghz (Python)
65 dgist_multiDetNet 25.97 % 34.56 % 22.74 % 0.05 s 1 core @ 2.5 Ghz (C/C++)
66 GA_FULLDATA 25.80 % 33.35 % 22.70 % 1 s 4 cores @ 2.5 Ghz (Python)
67 CSFADet 25.77 % 32.19 % 22.78 % 0.05 s GPU @ 2.5 Ghz (Python)
68 GA_BALANCE 25.27 % 33.79 % 22.03 % 1 s 1 core @ 2.5 Ghz (Python)
69 cas_retina 25.24 % 31.74 % 22.30 % 0.2 s 4 cores @ 2.5 Ghz (Python)
70 cas_retina_1_13 25.01 % 31.17 % 22.12 % 0.03 s 4 cores @ 2.5 Ghz (Python)
71 Multi-task DG 24.72 % 33.39 % 21.63 % 0.06 s GPU @ 2.5 Ghz (Python)
72 bigger_ga 24.64 % 31.31 % 21.06 % 1 s 1 core @ 2.5 Ghz (Python)
73 CRCNNA 23.88 % 29.91 % 20.70 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
74 FD2 23.83 % 35.75 % 20.79 % 0.01 s GPU @ >3.5 Ghz (Python + C/C++)
75 GAFM 22.84 % 31.62 % 19.88 % 0.5 s 1 core @ 2.5 Ghz (Python)
76 JSU-NET 22.83 % 31.58 % 19.81 % 0.1 s 1 core @ 2.5 Ghz (Python)
77 ga50 21.59 % 29.77 % 18.77 % 1 s 1 core @ 2.5 Ghz (Python)
78 fasterrcnn 21.52 % 28.50 % 18.86 % 0.2 s 4 cores @ 2.5 Ghz (Python)
79 ZKNet 21.51 % 28.26 % 18.83 % 0.01 s GPU @ 2.0 Ghz (Python)
80 LPN 21.11 % 27.67 % 18.82 % 0.2 s GPU @ 2.5 Ghz (Python + C/C++)
81 RFCN 20.77 % 26.80 % 18.25 % 0.2 s 4 cores @ 2.5 Ghz (Python)
82 yolo800 20.66 % 27.38 % 18.77 % 0.13 s 4 cores @ 2.5 Ghz (Python)
83 RFCN_RFB 20.40 % 26.19 % 17.91 % 0.2 s 4 cores @ 2.5 Ghz (Python)
84 DSGN
This method uses stereo information.
 code 20.28 % 29.76 % 19.13 % 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.
85 NM code 20.02 % 26.27 % 17.87 % 0.01 s GPU @ 2.5 Ghz (Python)
86 Cmerge 19.78 % 27.75 % 16.58 % 0.2 s 4 cores @ 2.5 Ghz (Python)
87 LSVM-MDPM-sv 19.15 % 26.05 % 18.02 % 10 s 4 cores @ 3.0 Ghz (C/C++)
P. Felzenszwalb, R. Girshick, D. McAllester and D. Ramanan: Object Detection with Discriminatively Trained Part-Based Models. PAMI 2010.
A. Geiger, C. Wojek and R. Urtasun: Joint 3D Estimation of Objects and Scene Layout. NIPS 2011.
88 OC Stereo
This method uses stereo information.
 18.99 % 29.07 % 16.40 % 0.35 s 1 core @ 2.5 Ghz (Python + C/C++)
89 DPM-VOC+VP 18.92 % 27.97 % 17.43 % 8 s 1 core @ 2.5 Ghz (C/C++)
B. Pepik, M. Stark, P. Gehler and B. Schiele: Multi-view and 3D Deformable Part Models. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015.
90 cascade_gw 18.74 % 27.00 % 16.35 % 0.2 s 4 cores @ 2.5 Ghz (Python)
91 MTDP 18.02 % 23.30 % 16.07 % 0.15 s GPU @ 2.0 Ghz (Python)
92 BdCost+DA+BB+MS 17.73 % 23.48 % 14.67 % TBD s 4 cores @ 2.5 Ghz (C/C++)
93 centernet 17.55 % 23.39 % 15.59 % 0.01 s GPU @ 2.5 Ghz (Python)
94 RefinedMPL 16.02 % 26.54 % 13.20 % 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.
95 DPM-C8B1
This method uses stereo information.
 14.64 % 23.93 % 13.09 % 15 s 4 cores @ 2.5 Ghz (Matlab + C/C++)
J. Yebes, L. Bergasa and M. García-Garrido: Visual Object Recognition with 3D-Aware Features in KITTI Urban Scenes. Sensors 2015.
J. Yebes, L. Bergasa, R. Arroyo and A. Lázaro: Supervised learning and evaluation of KITTI's cars detector with DPM. IV 2014.
96 Retinanet100 13.34 % 19.09 % 11.79 % 0.2 s 4 cores @ 2.5 Ghz (Python)
97 BdCost+DA+BB 13.30 % 17.22 % 11.04 % TBD s 4 cores @ 2.5 Ghz (C/C++)
98 softyolo 11.12 % 15.91 % 9.84 % 0.16 s 4 cores @ 2.5 Ghz (Python)
99 100Frcnn 11.07 % 16.90 % 9.63 % 2 s 4 cores @ 2.5 Ghz (Python + C/C++)
100 Lidar_ROI+Yolo(UJS) 8.95 % 13.15 % 7.96 % 0.1 s 1 core @ 2.5 Ghz (C/C++)
101 KD53-20 4.86 % 7.19 % 4.74 % 0.19 s 4 cores @ 2.5 Ghz (Python)
102 RT3DStereo
This method uses stereo information.
 3.88 % 5.46 % 3.54 % 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.
103 MP 0.97 % 0.62 % 0.89 % 0.2 s 1 core @ 2.5 Ghz (C/C++)
104 CBNet 0.18 % 0.11 % 0.21 % 1 s 4 cores @ 2.5 Ghz (Python)
105 softretina 0.11 % 0.07 % 0.08 % 0.16 s 4 cores @ 2.5 Ghz (Python)
106 JSyolo 0.02 % 0.01 % 0.02 % 0.16 s 4 cores @ 2.5 Ghz (Python)
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