Skip to content

Latest commit

 

History

History
executable file
·
222 lines (184 loc) · 23.9 KB

Comparison.md

File metadata and controls

executable file
·
222 lines (184 loc) · 23.9 KB
Raw

Leaderboard on various pedestrian datasets

  • If you find a new paper about pedestrian detection, please feel free to contact us for adding it.
  • If you find any error about performance, please feel free to contact us to fix it.

Table of Contents

  1. Caltech test set
  2. Citypersons validation set
  3. Citypersons test set
  4. TJU-Ped-campus validation set
  5. TJU-Ped-traffic validation set
  6. KITTI test set
  7. KAIST test set

Caltech test set

  • The original annotations, while the bottom table uses the new annotations.
  • CNN indicates whether or not deep features are used.
  • P. Dollár, C. Wojek, B. Schiele, and P. Perona, Pedestrian Detection: An Evaluation of the State of the Art, IEEE TPAMI 2010.
Method publication CNN R HO R+HO A link
ACF PAMI2014 no 44.2 90.2 54.6 79.6 Paper
SpatialPooling ECCV2014 no 29.2 84.1 41.7.6 74.0 Paper
LDCF NIPS2014 no 24.8 81.3 37.7 71.2 Paper
Katamari ECCV2014 no 22.5 84.4 36.2 71.3 Paper
DeepCascade BMVC2015 yes 31.1 81.7 42.4 74.1 Paper
SCCPriors BMVC2015 no 21.9 80.9 35.1 70.3 Paper
TA-CNN CVPR2015 no 20.9 70.4 33.3 71.2 Paper
CCF ICCV2015 yes 18.7 72.4 30.6 66.7 Paper
Checkerboards ICCV2015 yes 18.5 77.5 31.8 68.7 Paper
DeepParts ICCV2015 yes 11.9 60.4 22.8 64.8 Paper
CompACT-Deep BMVC2015 yes 11.7 65.8 24.6 64.4 Paper
NNNF CVPR2016 no 16.2 74.9 - - Paper
MS-CNN ECCV2016 yes 10.0 59.9 21.5 60.9 Paper
RPN+BF ECCV2016 yes 9.6 74.3 24.0 64.7 Paper
MCF TIP2017 yes 10.4 66.7 - - Paper
F-DNN WACV2017 yes 8.6 55.1 19.3 50.6 Paper
PCN BMVC2017 yes 8.4 55.8 19.2 61.9 Paper
PDOE ECCV2018 yes 7.6 44.4 - - Paper
UDN+ PAMI2018 yes 11.5 70.3 24.7 64.8 Paper
FRCNN+ATT CVPR2018 yes 10.3 45.2 18.2 54.5 Paper
SA-FRCNN TMM2018 yes 9.7 64.4 21.9 62.6 Paper
ADM TIP2018 yes 8.6 30.4 13.7 42.3 Paper
GDFL ECCV2018 yes 7.8 43.2 15.6 48.1 Paper
TLL-TFA ECCV2018 yes 7.4 28.7 12.3 38.2 Paper
AR-Ped CVPR2019 yes 6.5 48.8 16.1 58.9 Paper
FRCNN+A+DT ICCV2019 yes 8.0 37.9 - - Paper
MGAN ICCV2019 yes 6.8 38.1 13.8 - Paper
TFAN CVPR2020 yes 6.7 30.9 12.4 - Paper
Method publication CNN R HO R+HO A link
HyperLearner CVPR2017 yes 5.5 - - - Paper
RepLoss CVPR2018 yes 4.0 - - - Paper
ALFNet ECCV2018 yes 4.5 - - - Paper
BGRNet ACM-MM2020 yes 4.5 - - - Paper
OR-CNN ECCV2018 yes 4.1 - - - Paper
HGPD ACM-MM2020 yes 3.78 - - - Paper
SML ACMMM2020 yes 3.7 - - - Paper
JointDet AAAI2020 yes 3.0 - - - Paper
PedHutter AAAI2020 yes 2.3 - - - Paper

Citypersons validation set

  • Usually, HO represents pedestrians over 50 pixels in height with 35-80% occlusion.
  • † indicates the pedestrians over 50 pixels in height with more than 35% occlusion. Thus, † suggest higher difficulty.
  • Shanshan Zhang, Rodrigo Benenson, Bernt Schiele, CityPersons: A Diverse Dataset for Pedestrian Detection, CVPR 2017.
Method publication scale R HO link
Adapted FRCNN CVPR2017 1.0x 15.4 - Paper
RepLoss CVPR2018 1.0x 13.7 56.9† Paper
FRCNN+ATT CVPR2018 1.0x 16.0 56.7 Paper
TLL+MRF ECCV2018 1.0x 14.4 52.0† Paper
OR-CNN ECCV2018 1.0x 12.8 55.7† Paper
ALFNet ECCV2018 1.0x 12.0 51.9† Paper
Cascade RCNN CVPR2018 1.0x 12.0 49.4 Paper
LBST TIP2019 1.0x 12.6 48.7 Paper
CSP CVPR2019 1.0x 11.0 49.3† Paper
Adaptive-NMS CVPR2019 1.0x 11.9 55.2† Paper
EGCL arXiv2021 1.0x 11.5 51.1† Paper
MGAN ICCV2019 1.0x 11.5 51.7 Paper
HGPD ACM-MM2020 1.0x 11.3 51.7 Paper
AutoPedestrian TIP2021 1.0x 11.3 50.5† Paper
R2NMS CVPR2020 1.0x 11.1 53.3 Paper
PRNet ECCV2020 1.0x 10.8 42.0 Paper
APD TMM2021 1.0x 10.6 46.6† Paper
CaSe ECCV2020 1.0x 11.0 50.3 Paper
SADet ACM-MM2020 1.0x 9.7 52.8† Paper
BGCNet ACM-MM2020 1.0x 9.4 45.9† Paper
Method publication scale R HO link
Adapted FRCNN CVPR2017 1.3x 12.8 - Paper
RepLoss CVPR2018 1.3x 11.6 55.3† Paper
OR-CNN ECCV2018 1.3x 11.0 51.3† Paper
PDOE ECCV2018 1.3x 11.2 44.2 Paper
LBST TIP2019 1.3x 11.4 45.2 Paper
FRCNN+A+DT CVPR2019 1.3x 11.1 44.3 Paper
Adaptive-NMS CVPR2019 1.3x 10.8 54.2† Paper
HGPD ACM-MM2020 1.3x 10.9 40.9 Paper
IoU+Sign ICIP2019 1.3x 10.8 54.3† Paper
NOH-NMS ACM-MM2020 1.3x 10.8 53.0† Paper
CrowdDet CVPR2020 1.3x 10.7 - Paper
SML ACMMM2020 1.3x 10.6 - Paper
EGCL arXiv2021 1.3x 10.5 45.3† Paper
MGAN ICCV2019 1.3x 10.3 49.6 Paper
AutoPedestrian TIP2021 1.3x 10.3 49.4† Paper
CaSe ECCV2020 1.3x 9.6 48.2 Paper
JointDet CVPR2019 1.3x 10.2 - Paper
0.5-stage WACV2020 1.3x 8.1 - Paper
PedHutter AAAI2020 1.3x 8.3 43.5† Paper

Citypersons test set

  • Shanshan Zhang, Rodrigo Benenson, and Bernt Schiele, CityPersons: A Diverse Dataset for Pedestrian Detection, CVPR 2017.
Method publication R RS HO A link
MS-CNN ECCV2016 13.32 15.86 51.88 39.94 Paper
Adapted FRCNN CVPR2017 12.97 37.24 50.47 43.86 Paper
Cascade MS-CNN CVPR2018 11.62 13.64 47.14 37.63 Paper
RepLoss CVPR2018 11.48 15.67 52.59 39.17 Paper
Adaptive-NMS CVPR2019 11.40 13.64 46.99 38.89 Paper
OR-CNN ECCV2018 11.32 14.19 51.43 40.19 Paper
MHN TMM2019 12.92 17.24 46.72 39.16 Paper
HBA-RCNN - 11.26 15.68 39.54 38.77 -
DVRNet - 10.99 15.68 43.77 41.48 -
HGPD ACM-MM2020 10.17 - 38.65 38.24 Paper
MGAN ICCV2019 9.29 11.38 40.97 38.86 Paper
STNet - 8.92 11.13 34.31 29.54 -
YT-PedDet - 8.41 10.60 37.88 37.22 -
APD arXiv2019 8.27 11.03 35.45 35.65 Paper
Pedestron arXiv2020 7.69 9.16 27.08 28.33 Paper
  • RS represents the pedestrians over 50 pixels and under 75 pixels with less than 0.35 occlusion, while A the pedestrians over 20 pixels with less than 0.8 occlusion.

TJU-Ped-campus validation set

  • Yanwei Pang, Jiale Cao, Yazhao Li, Jin Xie, Hanqing Sun, and Jinfeng Gong, TJU-DHD: A Diverse High-Resolution Dataset for Object Detection, IEEE TIP2021.
Method publication R RS HO R+HO A link
RetinaNet ICCV2017 34.73 82.99 71.31 42.26 44.34 Paper
FCOS ICCV2019 31.89 69.04 81.28 39.38 41.62 Paper
FPN ICCV2017 27.92 67.52 73.14 35.67 38.08 Paper
CrowdDet CVPR2020 25.73 - 66.38 33.63 35.90 Paper
EGCL arXiv2021 24.84 - 65.27 32.39 34.87 Paper

TJU-Ped-traffic validation set

  • Yanwei Pang, Jiale Cao, Yazhao Li, Jin Xie, Hanqing Sun, Jinfeng Gong, TJU-DHD: A Diverse High-Resolution Dataset for Object Detection, IEEE TIP2021.
Method publication R RS HO R+HO A link
RetinaNet ICCV2017 23.89 37.92 61.60 28.45 41.40 Paper
FCOS ICCV2019 24.35 37.40 63.73 28.86 40.02 Paper
FPN ICCV2017 22.30 35.19 60.30 26.71 37.78 Paper
CrowdDet CVPR2020 20.82 - 61.22 25.28 36.94 Paper
EGCL arXiv2021 19.73 - 60.05 24.19 35.76 Paper

KITTI test set

  • Andreas Geiger, Philip Lenz, and Raquel Urtasun, Are we ready for Autonomous Driving? The KITTI Vision Benchmark Suite, CVPR 2012.
Method publication Medium Easy Hard link
ACF PAMI2014 39.81 44.49 37.21 Paper
Checkerboards CVPR2015 56.75 67.65 51.12 Paper
DeepParts ICCV2015 58.67 70.49 52.78 Paper
CompACT-Deep ICCV2015 58.74 70.69 52.71 Paper
Regionlets PAMI2015 60.83 73.79 54.72 Paper
NNNF CVPR2016 58.01 69.16 52.77 Paper
MCF TIP2016 59.45 70.87 54.28 Paper
RPN+BF ECCV2016 61.29 75.45 56.08 Paper
SDP+RPN CVPR2016 70.42 82.07 65.09 Paper
IVA ACCV2016 71.37 84.61 64.90 Paper
MS-CNN ECCV2016 74.89 85.71 68.99 Paper
SubCNN WACV2017 72.77 84.88 66.82 Paper
PCN BMVC2017 63.41 80.08 58.55 Paper
GN PRL2017 72.29 82.93 65.56 Paper
RRC CVPR2017 76.61 85.98 71.47 Paper
CFM TCSVT2018 62.84 74.76 56.06 Paper
SAF R-CNN TMM2018 65.01 77.93 60.42 Paper
SJTU-HW ICIP2018 75.81 87.17 69.86 Paper
GDFL ECCV2018 68.62 84.61 66.86 Paper
MonoPSR CVPR2019 68.56 85.60 63.34 Paper
FFNet PR2019 75.99 87.21 69.86 Paper
MHN TCSVT2019 75.99 87.21 69.50 Paper
Aston-EAS TITS2019 76.07 86.71 70.02 Paper
AR-Ped CVPR2019 73.44 83.66 68.12 Paper

KAIST test set

  • Soonmin Hwang, Jaesik Park, Namil Kim, Yukyung Choi, and In So Kweon, Multispectral Pedestrian Detection: Benchmark Dataset and Baselines, CVPR 2015.
Method publication MR(All) MR(Day) MR(Nighy) link
ACF CVPR2015 47.32 42.57 56.17 Paper
Halfway Fusion BMVC2016 25.75 24.88 26.59 Paper
IAF-RCNN PR2019 15.73 14.55 18.26 Paper
IATDNN+IAMSS IF2019 14.95 14.67 15.72 Paper
CIAN IF2019 14.12 14.77 11.13 Paper
MSDS-RCNN BMVC2018 11.34 10.53 12.94 Paper
AR-CNN ICCV2019 9.34 9.94 8.38 Paper
MBNet ECCV2020 8.13 8.28 7.86 Paper
SCDN ArXiv2021 8.07 8.16 7.51 Paper
UGCML TCSVT2022 7.89 8.18 6.96 Paper