MOT16 Results

Click on a measure to sort the table accordingly. See below for a more detailed description.


Showing only entries that use public detections!

TrackerAvg RankMOTAMOTPFAFMTMLFPFNID Sw.Frag HzDetector
DP_NMS
1. using public detections
18.4
32.2
±9.8
76.40.25.4% 62.1% 1,123121,579972 (29.2)944 (28.3)212.6Public
H. Pirsiavash, D. Ramanan, C. Fowlkes. Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects. In CVPR, 2011.
LP2D
2. using public detections
17.8
35.7
±10.1
75.80.98.7% 50.7% 5,084111,163915 (23.4)1,264 (32.4)49.3Public
MOT baseline: Linear programming on 2D image coordinates.
JPDA_m
3. using public detections
17.3
26.2
±6.1
76.30.64.1% 67.5% 3,689130,549365 (12.9)638 (22.5)22.2Public
H. Rezatofighi, A. Milan, Z. Zhang, Q. Shi, A. Dick, I. Reid. Joint Probabilistic Data Association Revisited. In ICCV, 2015.
DeepAC
4. online method using public detections
19.7
38.8
±9.3
74.90.99.1% 42.8% 5,444103,1742,886 (66.5)6,592 (151.9)21.1Public
Anonymous submission
TSSRC
5. online method using public detections
18.0
42.4
±11.8
76.82.512.8% 44.9% 14,68589,654739 (14.5)1,368 (26.9)16.8Public
Anonymous submission
JCmin_MOT
6. online method using public detections
17.0
36.7
±9.1
75.90.57.5% 54.4% 2,936111,890667 (17.3)831 (21.5)14.8Public
Joint Cost Minimization for Multi-Object Tracking
GMPHD_HDA
7. online method using public detections
19.8
30.5
±6.9
75.40.94.6% 59.7% 5,169120,970539 (16.0)731 (21.7)13.6Public
Y. Song, M. Jeon. Online Multiple Object Tracking with the Hierarchically Adopted GM-PHD Filter using Motion and Appearance. In IEEE/IEIE The International Conference on Consumer Electronics (ICCE) Asia, 2016.
EAMTT_pub
8. online method using public detections
22.8
38.8
±8.5
75.11.47.9% 49.1% 8,114102,452965 (22.0)1,657 (37.8)11.8Public
R. Sanchez-Matilla, F. Poiesi, A. Cavallaro "Multi-target tracking with strong and weak detections" in BMTT ECCVw 2016
LRIM
9. online method using public detections
22.2
30.9
±15.8
76.10.45.4% 54.9% 2,375119,4804,075 (118.2)5,484 (159.1)10.0Public
Anonymous submission
DQNTracker
10. online method using public detections
22.9
33.7
±13.7
75.40.96.9% 59.3% 5,210113,8651,744 (46.4)4,184 (111.4)9.9Public
Anonymous submission
TrackerAvg RankMOTAMOTPFAFMTMLFPFNID Sw.Frag HzDetector
DACTracker
11. online method using public detections
23.7
38.2
±9.5
74.81.28.4% 45.8% 7,079103,3942,228 (51.5)5,969 (137.9)9.9Public
Anonymous submission
NLLMPa
12. using public detections
9.4
47.6
±10.6
78.51.017.0% 40.4% 5,84489,093629 (12.3)768 (15.0)8.3Public
E. Levinkov, J. Uhrig, S. Tang, M. Omran, E. Insafutdinov, A. Kirillov, C. Rother, T. Brox, B. Schiele, B. Andres. Joint Graph Decomposition and Node Labeling: Problem, Algorithms, Applications. In CVPR, 2017.
LINF1
13. using public detections
19.6
41.0
±9.5
74.81.311.6% 51.3% 7,89699,224430 (9.4)963 (21.1)4.2Public
L. Fagot-Bouquet, R. Audigier, Y. Dhome, F. Lerasle. Improving Multi-Frame Data Association with Sparse Representations for Robust Near-Online Multi-Object Tracking. In ECCV, 2016.
PT_JMC
14. using public detections
16.8
45.2
±8.4
74.82.117.7% 38.3% 12,20487,081681 (13.0)1,152 (22.1)3.8Public
Anonymous submission
MLMRF_DL61
15. online method using public detections
13.3
48.4
±9.4
74.31.318.2% 39.5% 7,84985,719491 (9.3)873 (16.5)3.0Public
Anonymous submission
NOMT
16. using public detections
10.8
46.4
±9.9
76.61.618.3% 41.4% 9,75387,565359 (6.9)504 (9.7)2.6Public
W. Choi. Near-Online Multi-target Tracking with Aggregated Local Flow Descriptor. In ICCV, 2015.
QuadMOT16
17. using public detections
14.8
44.1
±9.4
76.41.114.6% 44.9% 6,38894,775745 (15.5)1,096 (22.8)1.8Public
J. Son, M. Baek, M. Cho, B. Han. Multi-Object Tracking with Quadruplet Convolutional Neural Networks. In CVPR, 2017.
DSR
18. using public detections
14.4
42.8
±8.8
77.61.112.8% 45.8% 6,37297,214688 (14.7)756 (16.2)1.4Public
Anonymous submission
TBD
19. using public detections
22.7
33.7
±9.2
76.51.07.2% 54.2% 5,804112,5872,418 (63.2)2,252 (58.9)1.3Public
A. Geiger, M. Lauer, C. Wojek, C. Stiller, R. Urtasun. 3D Traffic Scene Understanding from Movable Platforms. In Pattern Analysis and Machine Intelligence (PAMI), 2014.
MDPNN16
20. online method using public detections
13.2
47.2
±7.7
75.80.514.0% 41.6% 2,68192,856774 (15.8)1,675 (34.1)1.0Public
A. Sadeghian, A. Alahi, S. Savarese. Tracking The Untrackable: Learning To Track Multiple Cues with Long-Term Dependencies. In arXiv preprint arXiv:1701.01909, 2017.
TrackerAvg RankMOTAMOTPFAFMTMLFPFNID Sw.Frag HzDetector
JMC
21. using public detections
13.8
46.3
±9.0
75.71.115.5% 39.7% 6,37390,914657 (13.1)1,114 (22.2)0.8Public
S. Tang, B. Andres, M. Andriluka, B. Schiele. Multi-Person Tracking by Multicuts and Deep Matching. In BMTT, 2016.
MHT_DAM
22. using public detections
13.4
42.9
±8.9
76.61.013.6% 46.9% 5,66897,919499 (10.8)659 (14.2)0.8Public
C. Kim, F. Li, A. Ciptadi, J. Rehg. Multiple Hypothesis Tracking Revisited. In ICCV, 2015.
HCC
23. using public detections
6.8
49.3
±10.2
79.00.917.8% 39.9% 5,33386,795391 (7.5)535 (10.2)0.8Public
Anonymous submission
MCjoint
24. using public detections
12.0
47.1
±10.8
76.31.120.4% 46.9% 6,70389,368370 (7.3)598 (11.7)0.6Public
Anonymous submission
FWT
25. using public detections
22.3
42.3
±9.0
74.51.414.1% 40.4% 8,48195,6431,032 (21.7)2,612 (54.9)0.6Public
Anonymous submission
YGT
26. using public detections
16.5
44.7
±9.0
75.22.118.6% 46.5% 12,49187,855404 (7.8)709 (13.7)0.6Public
Anonymous submission
LTTSC-CRF
27. using public detections
22.4
37.6
±9.9
75.92.09.6% 55.2% 11,969101,343481 (10.8)1,012 (22.8)0.6Public
N. Le, A. Heili, M. Odobez. Long-Term Time-Sensitive Costs for CRF-Based Tracking by Detection. In ECCVw, 2016.
LMP
28. using public detections
9.6
48.8
±9.8
79.01.118.2% 40.1% 6,65486,245481 (9.1)595 (11.3)0.5Public
Anonymous submission
CDA_DDALv2
29. online method using public detections
19.6
43.9
±7.8
74.71.110.7% 44.4% 6,45095,175676 (14.1)1,795 (37.6)0.5Public
S. Bae and K. Yoon, Under review.
GMCSS
30. online method using public detections
26.6
37.5
±9.0
75.02.58.2% 48.0% 14,60698,511838 (18.2)2,057 (44.7)0.4Public
Anonymous submission
TrackerAvg RankMOTAMOTPFAFMTMLFPFNID Sw.Frag HzDetector
oICF
31. online method using public detections
19.9
43.2
±10.2
74.31.111.3% 48.5% 6,65196,515381 (8.1)1,404 (29.8)0.4Public
H. Kieritz, S. Becker, W. Hübner, M. Arens. Online Multi-Person Tracking using Integral Channel Features. In IEEE Advanced Video and Signal-based Surveillance (AVSS) 2016, 2016.
CEM
32. using public detections
22.9
33.2
±7.9
75.81.27.8% 54.4% 6,837114,322642 (17.2)731 (19.6)0.3Public
A. Milan, S. Roth, K. Schindler. Continuous Energy Minimization for Multitarget Tracking. In IEEE TPAMI, 2014.
NHL
33. using public detections
21.3
45.1
±8.5
72.52.115.9% 37.3% 12,60585,6911,747 (33.0)2,033 (38.4)0.3Public
Anonymous submission
OVBT
34. online method using public detections
26.2
38.4
±8.8
75.41.97.5% 47.3% 11,51799,4631,321 (29.1)2,140 (47.1)0.3Public
Y. Ban, S. Ba, X. Alameda-Pineda, R. Horaud. Tracking Multiple Persons Based on a Variational Bayesian Model. In BMTT 2016, .
SMOT
35. using public detections
31.3
29.7
±7.3
75.22.95.3% 47.7% 17,426107,5523,108 (75.8)4,483 (109.3)0.2Public
C. Dicle, O. Camps, M. Sznaier. The Way They Move: Tracking Targets with Similar Appearance. In ICCV, 2013.
PTBFPT
36. using public detections
26.8
10.5
±6.4
66.71.40.1% 90.0% 8,106154,754303 (20.0)293 (19.4)0.0Public
Anonymous submission

Due to a minor bug in the export script, all results were re-evaluated on April 11, 2016. Here is the old snapshot of the leaderboard.


Benchmark Statistics

SequencesFramesTrajectoriesBoxes
75919759182326

Difficulty Analysis

Sequence difficulty (from easiest to hardest, measured by average MOTA)

MOT16-03

MOT16-03

(49.9% MOTA)

MOT16-06

MOT16-06

(42.2% MOTA)

MOT16-07

MOT16-07

(37.1% MOTA)

...

...

MOT16-08

MOT16-08

(28.6% MOTA)

MOT16-14

MOT16-14

(22.5% MOTA)


Evaluation Measures

Lower is better. Higher is better.
Measure Better Perfect Description
Avg Rank lower 1 This is the rank of each tracker averaged over all present evaluation measures.
MOTA higher 100 % Multiple Object Tracking Accuracy [1]. This measure combines three error sources: false positives, missed targets and identity switches.
MOTP higher 100 % Multiple Object Tracking Precision [1]. The misalignment between the annotated and the predicted bounding boxes.
FAF lower 0 The average number of false alarms per frame.
MT higher 100 % Mostly tracked targets. The ratio of ground-truth trajectories that are covered by a track hypothesis for at least 80% of their respective life span.
ML lower 0 % Mostly lost targets. The ratio of ground-truth trajectories that are covered by a track hypothesis for at most 20% of their respective life span.
FP lower 0 The total number of false positives.
FN lower 0 The total number of false negatives (missed targets).
ID Sw. lower 0 The total number of identity switches. Please note that we follow the stricter definition of identity switches as described in [2].
Frag lower 0 The total number of times a trajectory is fragmented (i.e. interrupted during tracking).
Hz higher Inf. Processing speed (in frames per second excluding the detector) on the benchmark.

Legend

Symbol Description
online method This is an online (causal) method, i.e. the solution is immediately available with each incoming frame and cannot be changed at any later time.
using public detections This method used the provided detection set as input.
new This entry has been submitted or updated less than a week ago.

References:


[1] Bernardin, K. & Stiefelhagen, R. Evaluating Multiple Object Tracking Performance: The CLEAR MOT Metrics. Image and Video Processing, 2008(1):1-10, 2008.
[2] Li, Y., Huang, C. & Nevatia, R. Learning to associate: HybridBoosted multi-target tracker for crowded scene. In Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2009.