MOT17 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.FragHzDetector
FWT
1. using public detections new
5.2
51.3
±13.1
77.01.421.4% 35.2% 24,101247,9212,648 (47.2)4,279 (76.3)0.2Public
R. Henschel, L. Leal-Taixé, D. Cremers, B. Rosenhahn. Improvements to Frank-Wolfe optimization for multi-detector multi-object tracking. In arXiv preprint arXiv:1705.08314, 2017.
jCC
2. using public detections
5.6
51.2
±14.5
75.91.420.7% 37.4% 24,986248,3281,851 (33.1)2,991 (53.4)1.8Public
M. Keuper, S. Tang, Y. Zhongjie, B. Andres, T. Brox, B. Schiele. A multi-cut formulation for joint segmentation and tracking of multiple objects. In arXiv preprint arXiv:1607.06317, 2016.
MHT_DAM
3. using public detections
4.6
50.7
±13.7
77.51.320.8% 36.9% 22,875252,8892,314 (41.9)2,865 (51.9)0.9Public
C. Kim, F. Li, A. Ciptadi, J. Rehg. Multiple Hypothesis Tracking Revisited. In ICCV, 2015.
EDMT17
4. using public detections
5.6
50.0
±13.9
77.31.821.6% 36.3% 32,279247,2972,264 (40.3)3,260 (58.0)0.6Public
J. Chen, H. Sheng, Y. Zhang, Z. Xiong. Enhancing Detection Model for Multiple Hypothesis Tracking. In BMTT-PETS CVPRw, 2017.
JCSTDDP
5. online method using public detections
6.7
47.6
±11.8
76.11.616.9% 34.2% 27,672264,4223,825 (72.0)8,688 (163.5)16.2Public
Anonymous submission
IOU17
6. using public detections
7.2
45.5
±13.6
76.91.115.7% 40.5% 19,993281,6435,988 (119.6)7,404 (147.8)1,522.9Public
E. Bochinski, V. Eiselein, T. Sikora. High-Speed Tracking-by-Detection Without Using Image Information. In International Workshop on Traffic and Street Surveillance for Safety and Security at IEEE AVSS 2017, 2017.
NN_LM
7. using public detections
5.4
45.1
±13.3
78.90.614.8% 46.2% 10,834296,4512,286 (48.2)2,463 (51.9)2.5Public
Anonymous submission
MEST
8. online method using public detections
7.5
44.7
±16.0
77.41.715.9% 34.9% 30,666275,8175,382 (105.3)9,210 (180.2)11.8Public
Anonymous submission
DP_NMS
9. using public detections
6.6
43.7
±14.0
77.30.612.6% 46.5% 10,048302,7284,942 (106.6)5,342 (115.3)137.7Public
Anonymous submission
TBD
10. using public detections
9.4
43.2
±14.7
77.02.516.3% 37.3% 45,131267,6417,570 (144.0)7,489 (142.5)1.5Public
Anonymous submission
TrackerAvg RankMOTAMOTPFAFMTMLFPFNID Sw.FragHzDetector
DMPO
11. online method using public detections
7.1
43.1
±13.8
78.10.711.3% 46.4% 11,949304,6214,408 (95.8)6,679 (145.2)29.7Public
Anonymous submission
ESCNN
12. online method using public detections
9.1
37.9
±15.5
76.43.417.7% 33.7% 60,287261,35428,637 (533.5)15,254 (284.2)7.8Public
Anonymous submission
SORTASMS
13. online method using public detections
12.3
25.4
±8.2
66.73.52.8% 49.6% 61,821353,0545,792 (154.8)10,096 (269.8)3.9Public
Anonymous submission
STSiam
14. online method using public detections
12.7
15.7
±19.4
69.45.22.6% 50.4% 93,155376,7945,730 (172.5)6,890 (207.4)0.8Public
Anonymous submission

Benchmark Statistics

SequencesFramesTrajectoriesBoxes
21177572355564228

Difficulty Analysis

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

MOT17-03-SDP

MOT17-03-SDP

(63.1% MOTA)

MOT17-03-FRCNN

MOT17-03-FRCNN

(51.4% MOTA)

MOT17-06-SDP

MOT17-06-SDP

(43.0% MOTA)

...

...

MOT17-14-DPM

MOT17-14-DPM

(15.8% MOTA)

MOT17-14-FRCNN

MOT17-14-FRCNN

(14.5% 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.