1

Robust Visual Tracking via Hierarchical Convolutional Features Supplementary Document Chao Ma, Jia-Bin Huang, Xiaokang Yang, and Ming-Hsuan Yang

F

The current object tracking benchmarks, such as the OTB [1], [2] and the VOT [3], [4], do not provide separate validation and test sets. To address the lack of validation set in the OTB [1], [2] and the VOT [3], [4] datasets, we use a separate set of video sequences (that are not in OTB and VOT) introduced by the MEEM method (http://cs-people.bu.edu/jmzhang/MEEM/MEEM.html) as our validation dataset for our hyper-parameter selection. The MEEM dataset contains 10 challenging sequences with featured attributes, such as heavy occlusion (dance, boxing1, boxing2, ped1, and ped2), abrupt illumination changes (carRace, and billieJean), low contrast (ball, ped2, rocky, and billieJean), and significant non-rigid deformation (latin, ball, carRace, dance, and billieJean). Figure 1 shows sample frames from the MEEM dataset, which contains approximately 7500 frames in total. In the following, we describe how we choose the three important threshold values regarding 1) 2) 3)

the re-detection threshold T0 for activating the detection module; the acceptance threshold (= 1.5T0 ) for adopting detection results; and the stability threshold (= T0 ) for conservatively updating the long-term filter.

In evaluating our tracker on the MEEM dataset, we disable our re-detection module of the HCFT* so that we can obtain more samples of failure cases. We refer this baseline implementation by HCFT*-no-det. We use the same long-term filter of the HCFT* to compute the tracking confidence scores of the HCFT*-no-det method. Over the 10 video sequences in the MEEM dataset, we observe that the when tracking failures occur (i.e., the overlap success scores are below 0.5), the confidence scores computed by the long-term filter are generally smaller than 0.2. We present in Figure 2 sample results showing the confidence scores (blue curves) as well as the overlap success scores in terms of Intersection over Union (red curves) on all the MEEM sequences. For the ball, carRace, and ped1 sequences, the HCFT*-no-det method fails to track target objects after a few hundreds of frames due to heavy occlusion or the out-of-the-view movement. We thus empirically set the re-detection threshold T0 to 0.2. For setting the acceptance threshold, we use a larger value than the re-detection threshold T0 to conservatively accept the detection results. Specifically, we use a parameter sweep scheme by setting the re-detection threshold to K ∗ T0 , where K ∈ [1, 2] with an increment of 0.1 (i.e., K = 1, 1.1, · · · , 1.9, 2.0). We find that setting the acceptance threshold to 1.5T0 gives the best averaged performance on the 10 MEEM video sequences. For the stability threshold, we start by setting this threshold to 0.2 as the long-term filter has to be updated with high confidence. We then sweep the stability threshold from 0.2 to 0.5. However, we find that the performance is not sensitive to the stability threshold on all the 10 validation sequences. Figure 3 (left) shows an example on the ball sequence. We thus set the stability threshold to equal T0 as 0.2. In addition, we validate the appropriateness of the stability threshold on the test sequences in the OTB dataset. We have the same observation that tracking confidence scores are consistent when the stability threshold between 0.2 and 0.5, as an example of the test sequence lemming shown in Figure 12 (bottom right) of the manuscript. For convenience, we present Figure 12 (bottom right) of the manuscript in Figure 3 (right) in this response. We first use the MEEM dataset as our validation set to find the three threshold values. Then we fix the hyper-parameters in all our experiments in OTB and VOT sequences. Our results show that the tracker work well on the testing datasets (OTB and VOT). For example, we observe that when tracking failures occur (i.e., the overlap success scores are below 0.5), the confidence scores computed by the long-term filter are generally smaller than 0.2 on most of the OTB sequences. Figure 4 shows the tracking confidence scores with the comparison to tracking success scores on all the 100 OTB sequences.

R EFERENCES [1] [2] [3] [4]

Y. Wu, J. Lim, and M.-H. Yang, “Online object tracking: A benchmark,” in Proc. of IEEE Conf. on Computer Vision and Pattern Recognition, 2013. 1 ——, “Object tracking benchmark,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 37, no. 9, pp. 1834–1848, 2015. 1 M. Kristan et al., “The visual object tracking VOT2014 challenge results,” in Proc. of European Conf. on Computer Vision Workshop, 2014. 1 M. Kristan, J. Matas, A. Leonardis, M. Felsberg, L. Cehovin, G. Fern´andez, T. Voj´ır, G. H¨ager, G. Nebehay, and R. P. Pflugfelder, “The visual object tracking VOT2015 challenge results,” in Proc. of IEEE Int. Conf. on Computer Vision Workshop, 2015. 1 [5] J. Zhang, S. Ma, and S. Sclaroff, “MEEM: Robust Tracking via Multiple Experts Using Entropy Minimization,” in Proc. of European Conf. on Computer Vision, 2014.

2

#153

#247

#473

#323

#500

#219

#136

#078

#287

#112

Fig. 1: Overview of the MEEM dataset. Red: Tracking results of HCFT*-no-det. Green: Ground-truth bounding boxes.

Tracking confidence plot - ball

1

Tracking confidence plot - boxing2

1

Tracking confidence score Overlap ratio (IoU)

0.9

0.7

0.7

0.7

0.7

0.5 0.4 0.3

0.5 0.4 0.3

0.6 0.5 0.4 0.3

0.6 0.5 0.4 0.3

0.2

0.2

0.2

0.2

0.1

0.1

0.1

0.1

0

0

0

0

100

200

300

400

500

0

Frame number Tracking confidence plot - dance

1

200

300

400

500

600

700

800

0

Frame number Tracking confidence plot - latin

1

Tracking confidence score Overlap ratio (IoU)

0.9

100

1000

1500

0.5 0.4 0.3

0.1 0 200

400

600

800

1000

0

Frame number Tracking confidence plot - ped2

1

Tracking confidence score Overlap ratio (IoU)

0.9

0.6

0.2

0

Frame number Tracking confidence plot - ped1

1

Tracking confidence score Overlap ratio (IoU)

0.9

500

Confidence score

0.7

Confidence score

0.8

Confidence score

0.8

0.6

Tracking confidence score Overlap ratio (IoU)

0.9

0.7

0.7

0.7

0.7

0.4 0.3

0.5 0.4 0.3

0.6 0.5 0.4 0.3

0.6 0.5 0.4 0.3

0.2

0.2

0.2

0.2

0.1

0.1

0.1

0.1

0

0 0

100

200

300

400

500

600

700

0 0

50

100

Frame number

150

200

250

300

350

400

450

Confidence score

0.7

Confidence score

0.8

Confidence score

0.8

Confidence score

0.8

0.6

50

Frame number

100

150

200

250

1500

Tracking confidence score Overlap ratio (IoU)

0.6 0.5 0.4 0.3 0.2 0.1

0 0

1000

0.9

0.8

0.5

500

Frame number Tracking confidence plot - rocky

1

0.8

0.6

Tracking confidence score Overlap ratio (IoU)

0.9

0.8

0.6

Tracking confidence plot - carRace

1

Tracking confidence score Overlap ratio (IoU)

0.9

0.8

Confidence score

Confidence score

Tracking confidence plot - boxing1

1 Tracking confidence score Overlap ratio (IoU)

0.9

0.8

0

Confidence score

Tracking confidence plot - billieJean

1

Tracking confidence score Overlap ratio (IoU)

0.9

0 0

100

200

Frame number

300

400

500

600

700

800

900

0

Frame number

50

100

150

200

250

300

350

Frame number

Fig. 2: Tracking confidence scores and overlap ratio scores of HCFT*-no-det. Note that when tracking failures occur (i.e., the overlap ratio scores are below 0.5), the confidence scores are generally below 0.2.

Tracking confidence plot - ball

1 0.9 0.8

T=0.2 T=0.3 T=0.4 T=0.5

0.9 0.8 0.7

Confidence score

0.7

Confidence score

Tracking confidence plot - Lemming

1 T=0.2 T=0.3 T=0.4 T=0.5

0.6 0.5 0.4

0.6 0.5 0.4

0.3

0.3

0.2

0.2

0.1

0.1

0

100

200

300

Frame number

400

500

0

200

400

600

800

1000

1200

1400

Frame number

Fig. 3: Tracking confidence scores on the ball and lemming sequences. We test the stability threshold for updating the long-term filter. Confidence scores are consistent with threshold values between 0.2 and 0.5.

3

Tracking confidence plot - Basketball

1.2

Tracking confidence plot - Biker

1.2

0.8

0.8

0.4

0.4

0.2

0.2

0

0

0

0

100

200

300

400

500

600

700

800

0

Frame number Tracking confidence plot - BlurCar1

50

100

150

0

Frame number Tracking confidence plot - BlurCar2

1.2

50

100

150

200

250

300

350

400

450

0.6

0.4

0.2

0 0

Frame number Tracking confidence plot - BlurCar3

1.2

Tracking confidence score Overlap ratio (IoU)

20

40

60

80

100

0

Frame number Tracking confidence plot - BlurCar4

1.2

Tracking confidence score Overlap ratio (IoU)

1

0.8

0.4

0.6

0.4

0.6

0.4

0.2

0.2

0.2

0.2

0

0

0

0

100

200

300

400

500

600

700

800

0

Frame number Tracking confidence plot - BlurOwl

100

200

300

400

500

600

0

Frame number Tracking confidence plot - Board

1.2

Tracking confidence score Overlap ratio (IoU)

50

100

150

200

250

300

350

400

Tracking confidence score Overlap ratio (IoU)

100

150

200

250

300

350

400

0

0.8

0.6

0.4

0.2

0.2

0.2

0.2

0

0

0

0

100

200

300

400

500

600

700

0

100

200

Frame number Tracking confidence plot - Boy

300

400

500

600

700

0

Frame number Tracking confidence plot - Car1

1.2 Tracking confidence score Overlap ratio (IoU)

50

100

150

200

250

300

350

Tracking confidence score Overlap ratio (IoU)

0.6

0.4

0

50

100

150

200

250

300

0

Frame number Tracking confidence plot - Car24

1.2

Tracking confidence score Overlap ratio (IoU)

0.6

0.4

0.6

0.4

0.2

0.2

0.2

0.2

0

0

0

0

0

100

200

300

400

500

600

700

0

Frame number Tracking confidence plot - CarDark

1.2

200

400

600

800

1000

1200

0

Frame number Tracking confidence plot - CarScale

1.2

Tracking confidence score Overlap ratio (IoU)

200

400

600

800

1000

Tracking confidence score Overlap ratio (IoU)

0 500

1000

1500

2000

2500

3000

0

3500

Frame number Tracking confidence plot - Coke

1.2

Tracking confidence score Overlap ratio (IoU)

0.8

0.4

0.6

0.4

0.2

0.2

0.2

0.2

0

0

0

0

0

50

100

150

200

250

300

350

400

0

Frame number Tracking confidence plot - Coupon

50

100

150

200

250

300

0

Frame number Tracking confidence plot - Crossing

1.2

Tracking confidence score Overlap ratio (IoU)

100

200

300

400

500

Tracking confidence score Overlap ratio (IoU)

100

150

200

250

300

0

Frame number Tracking confidence plot - Dancer

0.6

0.4

0.6

0.4

0.2

0.2

0.2

0.2

0

0

0

0

100

150

200

250

300

350

0

1.2

20

40

60

80

100

120

0

Frame number Tracking confidence plot - David2

1.2

Tracking confidence score Overlap ratio (IoU)

50

100

150

200

250

300

350

Tracking confidence score Overlap ratio (IoU)

100

150

200

250

0

Frame number Tracking confidence plot - Deer

0.8

0.8

0.6

0.4

0.2

0.2

0.2

0.2

0

0

0

0

100

200

300

Frame number

400

500

0

100

200

300

Frame number

400

500

600

0

50

100

150

Frame number

200

250

300

Confidence score

0.8

Confidence score

0.8

Confidence score

0.8

Confidence score

1

0.4

150

Tracking confidence score Overlap ratio (IoU)

1

0

100

Tracking confidence score Overlap ratio (IoU)

Tracking confidence score Overlap ratio (IoU)

0.4

50

Frame number Tracking confidence plot - Diving

1.2

1

0.4

140

0 50

1.2

0.6

120

0.4

1

0.6

100

0.6

1

0.6

80

0.2

0

Frame number Tracking confidence plot - David3

1.2

Confidence score

Confidence score

1

0.8

Confidence score

1

0.8

Confidence score

1

Frame number Tracking confidence plot - David

60

Tracking confidence score Overlap ratio (IoU)

0.8

50

40

Tracking confidence score Overlap ratio (IoU)

1

0

20

Frame number Tracking confidence plot - Dancer2

1.2

0.8

0.4

700

0 50

1.2

Tracking confidence score Overlap ratio (IoU)

0.6

600

0.4

1

0.4

500

0.6

0.8

0.6

400

0.2

0

Frame number Tracking confidence plot - Crowds

1.2

Confidence score

0.8

Confidence score

0.8

Confidence score

0.8

Confidence score

0.8

1.2

300

Tracking confidence score Overlap ratio (IoU)

1

0.4

200

Frame number Tracking confidence plot - Couple

Tracking confidence score Overlap ratio (IoU)

1

0.4

100

1.2

1

0.6

1200

0.4

1

0.6

1000

0.6

1

0.6

800

0.2

0

Frame number Tracking confidence plot - ClifBar

1.2

Confidence score

1

0.8

Confidence score

1

0.8

Confidence score

1

0.8

Confidence score

1

0.4

600

Tracking confidence score Overlap ratio (IoU)

0.8

0.6

400

Frame number Tracking confidence plot - Car4

Tracking confidence score Overlap ratio (IoU)

1

0.4

200

1.2

0.8

0.6

500

0.2

0

Frame number Tracking confidence plot - Car2

1.2

Confidence score

0.8

Confidence score

0.8

Confidence score

0.8

Confidence score

0.8

0.4

400

Tracking confidence score Overlap ratio (IoU)

1

0

300

Tracking confidence score Overlap ratio (IoU)

1

1.2

200

Frame number Tracking confidence plot - Box

1.2

Tracking confidence score Overlap ratio (IoU)

0.4

100

Frame number Tracking confidence plot - Bolt2

1

0.4

350

0 50

1.2

0.6

300

0.4

1

0.6

250

0.6

1

0.6

200

0.2

0

Frame number Tracking confidence plot - Bolt

1.2

Confidence score

1

0.8

Confidence score

1

0.8

Confidence score

1

0.6

150

Tracking confidence score Overlap ratio (IoU)

0.8

0.4

100

Frame number Tracking confidence plot - BlurFace

Tracking confidence score Overlap ratio (IoU)

1

0.6

50

1.2

0.8

Confidence score

Confidence score

0.4

0.2

0

Confidence score

0.4

0.6

0.2

1.2

Confidence score

0.6

Confidence score

0.8

0.6

Confidence score

0.8

Confidence score

0.8

Tracking confidence score Overlap ratio (IoU)

Confidence score

Tracking confidence score Overlap ratio (IoU)

1

0.6

Tracking confidence plot - BlurBody

1.2

Tracking confidence score Overlap ratio (IoU)

1

0

Confidence score

Tracking confidence plot - Bird2

1.2

Tracking confidence score Overlap ratio (IoU)

1

1.2

Confidence score

Tracking confidence plot - Bird1

1.2

Tracking confidence score Overlap ratio (IoU)

1

Confidence score

Confidence score

Tracking confidence score Overlap ratio (IoU)

1

0.6

0.4

0.2

0

0

10

20

30

40

Frame number

50

60

70

0

50

100

150

200

250

Frame number

Fig. 4: Tracking confidence scores and overlap ratio scores of HCFT* on the OTB-2015 dataset. Note that when tracking failures occur (i.e., the overlap ratio scores are below 0.5), the confidence scores are generally below 0.2.

4

Tracking confidence plot - Dog

1.2

Tracking confidence plot - Dog1

1.2

0.8

0.8

0.4

0.4

0.2

40

60

80

100

120

140

200

400

600

800

1000

500

1000

1500

2000

2500

3000

3500

Frame number Tracking confidence plot - Fish

1.2

Tracking confidence score Overlap ratio (IoU)

0

0

4000

0.4

0.2

0 0

0.6

20

40

60

80

100

120

0

Frame number Tracking confidence plot - FleetFace

1.2

Tracking confidence score Overlap ratio (IoU)

0.8

0.8

0.8

0.4

0.6

0.4

0.6

0.4

0.2

0.2

0.2

0.2

0

0

0

0

0

100

200

300

400

500

600

700

800

900

0

100

Frame number Tracking confidence plot - Football1

200

300

400

500

600

700

800

900

0

Frame number Tracking confidence plot - Freeman1

1.2 Tracking confidence score Overlap ratio (IoU)

100

200

300

400

500

Tracking confidence score Overlap ratio (IoU)

0 100

200

300

400

500

600

700

800

0

Frame number Tracking confidence plot - Freeman4

1.2

Tracking confidence score Overlap ratio (IoU)

0.8

0.4

0.6

0.4

0.2

0.2

0.2

0.2

0

0

0

0

0

10

20

30

40

50

60

70

80

0

Frame number Tracking confidence plot - Girl2

50

100

150

200

250

300

350

0

100

Frame number Tracking confidence plot - Gym

1.2

200

300

Tracking confidence score Overlap ratio (IoU)

Tracking confidence score Overlap ratio (IoU)

400

500

100

150

200

250

300

0

Frame number Tracking confidence plot - Human3

0

500

1000

Frame number Tracking confidence plot - Human5

1.2

100

200

300

400

500

600

700

800

1.2

Tracking confidence score Overlap ratio (IoU)

Confidence score

0.4

200

400

600

800

1000

1200

Tracking confidence score Overlap ratio (IoU)

0.4

200

400

600

800

1000

1200

1400

1600

0

1800

Frame number Tracking confidence plot - Human8

1.2

Tracking confidence score Overlap ratio (IoU)

0.8

0.8

0.6

0.4

0.6

0.4

0.2

0.2

0.2

0.2

0

0

0

0

0

100

200

300

400

500

600

700

800

0

Frame number Tracking confidence plot - Ironman

100

200

300

400

500

600

700

800

0

Frame number Tracking confidence plot - Jogging.1

1.2

Tracking confidence score Overlap ratio (IoU)

50

100

150

200

250

Tracking confidence score Overlap ratio (IoU)

40

60

80

100

120

140

0

Frame number Tracking confidence plot - Jump

0.6

0.4

0.6

0.4

0.2

0.2

0.2

0.2

0

0

0

0

40

60

80

100

120

140

160

180

0

1.2

50

100

150

200

250

300

350

0

50

100

Frame number

Tracking confidence score Overlap ratio (IoU)

200

250

300

350

Tracking confidence score Overlap ratio (IoU)

20

40

60

80

100

120

140

0

Frame number Tracking confidence plot - Man

1.2

0.8

0.4

0.6

0.4

0.2

0.2

0.2

0.2

0

0

0

0

10

20

30

40

50

Frame number

60

70

80

90

0

200

400

600

800

Frame number

1000

1200

1400

0

200

400

600

800

1000

Frame number

1200

1400

1600

1800

Confidence score

0.8

Confidence score

0.8

Confidence score

0.8

Confidence score

0.8

0

150

200

250

300

350

Tracking confidence score Overlap ratio (IoU)

1

0.4

100

Tracking confidence score Overlap ratio (IoU)

Tracking confidence score Overlap ratio (IoU)

1

0.4

50

Frame number Tracking confidence plot - Matrix

1.2

1

0.6

350

0.4

1

0.6

300

0.6

1

0.6

250

0 0

Tracking confidence plot - Liquor

1.2

200

0.2

Frame number

Tracking confidence plot - Lemming

1.2

150

Confidence score

Confidence score

1

0.8

Confidence score

1

0.8

Confidence score

1

Frame number Tracking confidence plot - KiteSurf

150

Tracking confidence score Overlap ratio (IoU)

0.8

20

100

Frame number Tracking confidence plot - Jumping

Tracking confidence score Overlap ratio (IoU)

1

0

50

1.2

0.8

0.4

700

0 20

1.2

Tracking confidence score Overlap ratio (IoU)

0.6

600

0.4

1

0.4

500

0.6

0.8

0.6

400

0.2

0

Frame number Tracking confidence plot - Jogging.2

1.2

Confidence score

0.8

Confidence score

0.8

Confidence score

0.8

Confidence score

1

1.2

300

Tracking confidence score Overlap ratio (IoU)

1

0.4

200

Frame number Tracking confidence plot - Human9

Tracking confidence score Overlap ratio (IoU)

1

0.4

100

1.2

1

0.6

500

0.6

1

0.6

400

0 0

Frame number Tracking confidence plot - Human7

1.2

300

0.2

0 0

Frame number Tracking confidence plot - Human6

0.6

0.2

0

0

1500

0.4

0.2

0

0

Confidence score

1

0.8

Confidence score

1

0.8

Confidence score

1

0.8

0.2

200

Tracking confidence score Overlap ratio (IoU)

1

0.2

100

Tracking confidence score Overlap ratio (IoU)

0.6

400

Frame number Tracking confidence plot - Human4

1.2

0.8

0.4

350

0 50

1.2

Tracking confidence score Overlap ratio (IoU)

0.6

300

0.4

1

0.4

250

0.6

0.8

0.6

200

0.2

0

Frame number Tracking confidence plot - Human2

1.2

Confidence score

0.8

Confidence score

0.8

Confidence score

0.8

Confidence score

0.8

1.2

150

Tracking confidence score Overlap ratio (IoU)

1

0.4

100

Frame number Tracking confidence plot - Girl

Tracking confidence score Overlap ratio (IoU)

1

0.4

50

1.2

1

0.6

1200

0.4

1

0.6

1000

0.6

1

0.6

800

0.2

0

Frame number Tracking confidence plot - Freeman3

1.2

Confidence score

0.8

Confidence score

0.8

Confidence score

1

0.4

600

Tracking confidence score Overlap ratio (IoU)

1

0.6

400

Frame number Tracking confidence plot - Football

Tracking confidence score Overlap ratio (IoU)

1

0.6

200

1.2

1

Confidence score

Confidence score

0.4

1

1.2

Confidence score

1400

Frame number Tracking confidence plot - FaceOcc2

1.2

Tracking confidence score Overlap ratio (IoU)

Confidence score

1200

0.6

0.2

0 0

Frame number Tracking confidence plot - FaceOcc1

1.2

0.4

0.2

0

20

0.6

Confidence score

0.8

0.6

Confidence score

0.8

Confidence score

0.8

0

Confidence score

Tracking confidence score Overlap ratio (IoU)

1

0.6

Tracking confidence plot - Dudek

1.2 Tracking confidence score Overlap ratio (IoU)

1

0

Confidence score

Tracking confidence plot - DragonBaby

1.2

Tracking confidence score Overlap ratio (IoU)

1

0.2

Confidence score

Tracking confidence plot - Doll

1.2

Tracking confidence score Overlap ratio (IoU)

1

Confidence score

Confidence score

Tracking confidence score Overlap ratio (IoU)

1

0.6

0.4

0.2

0 0

20

40

60

80

Frame number

100

120

140

0

20

40

60

80

100

Frame number

Fig. 4: (Continue) Tracking confidence scores and overlap ratio scores of HCFT* on the OTB-2015 dataset. Note that when tracking failures occur (i.e., the overlap ratio scores are below 0.5), the confidence scores are generally below 0.2.

5

Tracking confidence plot - Mhyang

1.2

Tracking confidence plot - MotorRolling

1.2

0.8

0.8

0.8

0.4

0.6

0.4

0.4

0.2

0.2

0.2

0

0

0

0

500

1000

0

1500

Frame number Tracking confidence plot - Rubik

20

40

60

80

100

120

140

160

180

0

Frame number Tracking confidence plot - Shaking

1.2

50

100

150

200

250

0.6

0.4

0.2

0 0

Frame number Tracking confidence plot - Singer1

1.2

Tracking confidence score Overlap ratio (IoU)

200

400

600

800

1000

0

Frame number Tracking confidence plot - Singer2

1.2

Tracking confidence score Overlap ratio (IoU)

1

0.8

0.4

0.2

0

500

1000

1500

Frame number Tracking confidence plot - Skater2

1.2

50

100

150

200

250

300

350

400

Tracking confidence score Overlap ratio (IoU)

0.4

50

100

150

200

250

300

350

Tracking confidence score Overlap ratio (IoU)

0.6

0.4

0 0

Frame number Tracking confidence plot - Skating2.1

1.2

50

100

150

200

250

300

350

400

0

Frame number Tracking confidence plot - Skating2.2

1.2

Tracking confidence score Overlap ratio (IoU)

0.8

0.8

0.8

0.6

0.4

0.6

0.4

0.2

0.2

0.2

0.2

0

0

0

0

0

50

100

150

200

250

300

350

400

450

0

Frame number Tracking confidence plot - Soccer

1.2

50

100

150

200

250

300

350

400

0

Frame number Tracking confidence plot - Subway

1.2

Tracking confidence score Overlap ratio (IoU)

100

200

300

400

Confidence score

0.8

Confidence score

0.8

Confidence score

1

Confidence score

1

0.4

500

100

200

300

400

0

500

Frame number Tracking confidence plot - Suv

Tracking confidence score Overlap ratio (IoU)

0.8

0.2

0

0.2

0 0

50

100

150

200

250

300

350

400

20

40

60

80

100

120

140

160

180

Tracking confidence score Overlap ratio (IoU)

0.4

50

100

150

200

250

300

350

400

Tracking confidence score Overlap ratio (IoU)

200

400

600

800

1000

0

Frame number Tracking confidence plot - Trans

1.2

Tracking confidence score Overlap ratio (IoU)

0.8

0.6

0.4

0.2

0.2

0.2

0.2

0

0

0

0

0

50

100

150

200

250

300

350

400

0

Frame number Tracking confidence plot - Twinnings

1.2

50

100

150

200

250

300

350

400

0

Frame number Tracking confidence plot - Vase

1.2

Tracking confidence score Overlap ratio (IoU)

50

100

150

200

250

300

Tracking confidence score Overlap ratio (IoU)

40

60

80

100

120

140

0

0.2

0

0.2

0 0

100

200

300

Frame number

400

500

50

100

150

Frame number

200

250

300

0.6

0.4

0.2

0 0

Confidence score

Confidence score

1

0.8

Confidence score

1

0.8

Confidence score

1

0.8

0.4

50

100

150

200

250

Frame number

300

350

400

200

300

450

400

500

600

0.6

0.4

0.2

0 0

1400

Tracking confidence score Overlap ratio (IoU)

1

0.2

100

Tracking confidence score Overlap ratio (IoU)

0.6

1200

Frame number Tracking confidence plot - Woman

1.2

0.8

0.4

1000

0 20

Frame number Tracking confidence plot - Walking2

1.2

Tracking confidence score Overlap ratio (IoU)

0.6

800

0.4

1

0.4

600

0.6

0.8

0.6

400

0.2

0

Frame number Tracking confidence plot - Walking

1.2

Confidence score

0.8

Confidence score

0.8

Confidence score

0.8

Confidence score

0.8

0.4

80

Tracking confidence score Overlap ratio (IoU)

1

0.4

200

Tracking confidence score Overlap ratio (IoU)

1

0.4

70

Frame number Tracking confidence plot - Trellis

1.2

1

0.6

60

0.4

1

0.6

50

0.6

1

0.6

40

0 0

Frame number Tracking confidence plot - Toy

1.2

30

0.2

0

0

Frame number Tracking confidence plot - Tiger2

1.2

0.6

0.2

0 0

Frame number Tracking confidence plot - Tiger1

1.2

0.4

Confidence score

0.8

Confidence score

0.8

Confidence score

0.8

Confidence score

0.8

0.2

20

Tracking confidence score Overlap ratio (IoU)

1

0.4

10

Tracking confidence score Overlap ratio (IoU)

1

0.4

160

Frame number Tracking confidence plot - Sylvester

1.2

1

0.6

140

0.4

1

0.6

120

0.6

1

0.6

100

0 0

1.2

Tracking confidence score Overlap ratio (IoU)

80

0.2

Frame number Tracking confidence plot - Surfer

1.2

60

Tracking confidence score Overlap ratio (IoU)

1

0.4

40

Frame number Tracking confidence plot - Skiing

Tracking confidence score Overlap ratio (IoU)

1

0.6

20

1.2

1

0.6

2000

0.2

0 0

Frame number Tracking confidence plot - Skating1

1.2

0.6

0.2

0 0

2000

0.4

0.2

0

0

0.6

Confidence score

1

0.8

Confidence score

1

0.8

Confidence score

1

0.6

1500

Tracking confidence score Overlap ratio (IoU)

0.8

0.4

1000

Tracking confidence score Overlap ratio (IoU)

1

0.6

500

Frame number Tracking confidence plot - Skater

1.2

0.8

Confidence score

Confidence score

0.4

0.6

0.2

0.2

Confidence score

0.6

Confidence score

0.8

Confidence score

0.8

Confidence score

1

Confidence score

Confidence score

1

Tracking confidence score Overlap ratio (IoU)

Confidence score

Tracking confidence score Overlap ratio (IoU)

1

0.6

Tracking confidence plot - RedTeam

1.2

Tracking confidence score Overlap ratio (IoU)

1

0

Confidence score

Tracking confidence plot - Panda

1.2 Tracking confidence score Overlap ratio (IoU)

1

1.2

Confidence score

Tracking confidence plot - MountainBike

1.2

Tracking confidence score Overlap ratio (IoU)

Tracking confidence score Overlap ratio (IoU)

0 0

100

200

300

Frame number

400

500

0

100

200

300

400

500

600

Frame number

Fig. 4: (Continue) Tracking confidence scores and overlap ratio scores of HCFT* on the OTB-2015 dataset. Note that when tracking failures occur (i.e., the overlap ratio scores are below 0.5), the confidence scores are generally below 0.2.s