Skip to main content

Table 2 Performance comparison of multiple rodent tracking on all frames in dataset B

From: EDDSN-MRT: multiple rodent tracking based on ear detection and dual siamese network for rodent social behavior analysis

  

Toxtrac [25]

idtracker.ai [24]

EDDSN-MRT (Ours)

ID

No

Miss

Switch

Drift

MOTA

ICR

Miss

Switch

Drift

MOTA

ICR

Miss

Switch

Drift

MOTA(P)

ICR(P)

B1

2

8730

1796

0

72.63%

67.11%

0

234

349

97.79%

98.18%

0

0

135

99.16% ( < 0.001)***

99.58% ( < 0.001)***

B2

2

32,441

12,522

0

55.42%

38.35%

0

1170

1101

96.95%

96.89%

0

16

730

98.00% ( < 0.001)***

98.98% ( < 0.001)***

B3

2

11,257

7195

0

72.95%

55.83%

252

3800

491

97.01%

89.12%

0

10

451

97.84% ( < 0.001)***

98.90% ( < 0.001)***

B4

2

7907

13,903

0

80.95%

47.81%

66

332

180

98.96%

98.62%

0

0

185

99.11%(0.117)

99.56% ( < 0.001)***

B5

4

64,370

59,250

0

68.19%

39.34%

2175

13,726

0

98.93%

92.20%

0

14

832

99.18% ( < 0.001)***

99.58% ( < 0.001)***

B6(1)1

4

None2

None

None

None

None

9040

7420

0

96.43%

89.02%

0

24

1240

98.34% ( < 0.001)***

99.16% ( < 0.001)***

B6(2)1

4

None2

None

None

None

None

6800

4120

0

96.90%

92.54%

0

0

1261

98.28% ( < 0.001)***

99.14% ( < 0.001)***

  1. 1Since there is an occluded fragment in video B6 caused by manual operation, individual information cannot be obtained in this fragment. Therefore, video B6 is divided into two parts to implement our method after deleting this fragment. idtracker.ai is not affected by this occlusion
  2. 2Toxtrac does not run on video B6, therefore no results are presented
  3. MOTA, Multiple Object Tracking Accuracy; ICR, ID Correct Rate. No., the number of rodents in the video. ***means P < 0.001