Table 5 Ablation results for the proposed method.

From: Multi-target detection and tracking based on CRF network and spatio-temporal attention for sports videos

Model

Datasets

MOT2015

MOT2016

PET2009

MS COCO

SportMOT

MAE

MAPE(%)

RMSE

MSE

MAE

MAPE(%)

RMSE

MSE

MAE

MAPE(%)

RMSE

MSE

MAE

MAPE(%)

RMSE

MSE

MAE

MAPE(%)

RMSE

MSE

CRF

49.42

12.66

7.79

25.27

43.28

14.23

6.62

14.16

40.64

14.19

4.30

15.33

25.60

9.26

6.93

27.00

25.22

13.11

6.16

17.73

LAC filter

21.30

14.48

7.60

25.45

29.15

11.86

7.00

27.42

41.46

9.07

4.69

25.37

28.63

12.65

7.13

21.15

41.02

13.89

6.94

15.42

S-T Attention

34.28

13.48

7.73

13.54

42.34

12.36

7.41

18.50

22.60

8.94

7.92

12.27

29.18

9.04

4.95

28.88

33.39

15.12

7.30

28.34

CRF+LAC filter

34.56

10.82

6.74

22.28

42.38

11.17

7.74

29.28

22.46

15.46

5.54

24.81

36.16

13.74

8.04

24.00

38.40

9.98

6.90

14.17

CRF+S-T Attention

33.25

12.13

5.51

17.08

47.48

9.22

4.71

26.51

37.80

11.10

4.92

18.97

22.97

12.71

8.42

18.74

34.43

14.84

7.04

22.04

LAC filter+S-T Attention

47.38

9.23

6.62

19.04

45.33

10.79

6.74

25.27

31.54

8.56

5.85

27.42

39.64

12.14

6.67

16.76

29.56

15.42

5.37

12.54

Ours

17.50

5.63

4.77

6.78

14.89

5.14

3.97

4.83

16.54

4.86

3.87

4.63

15.81

4.63

3.15

4.22

13.87

3.53

2.95

3.67

Back to article page