Table 2 Multi-metric test results for different algorithms.

From: Tennis ball detection based on YOLOv5 with tensorrt

Condition

Algorithm

P (%)

R (%)

F1 (%)

Computation time (s)

Model size (MB)

p

Normal light

YOLOX

89.73

89.74

89.74

7.35

16.37

0.001

YOLOv5-Lite

90.17

90.17

90.17

7.06

9.57

0.008

YOLOv7

92.26

91.62

91.94

5.76

78.64

0.003

YOLOv8

93.41

92.39

92.92

4.21

94.51

0.006

Our model

95.84

93.57

94.71

2.28

7.15

Bright light

YOLOX

90.37

89.41

89.89

6.52

18.36

0.001

YOLOv5-Lite

91.26

88.46

89.86

6.04

14.17

0.001

YOLOv7

92.56

90.28

91.42

5.16

84.19

0.004

YOLOv8

93.89

91.68

92.79

4.21

95.68

0.007

Our model

95.68

93.67

94.68

2.81

8.16

Low light

YOLOX

87.64

88.41

89.61

8.33

22.58

0.006

YOLOv5-Lite

88.96

85.72

88.16

7.25

19.58

0.009

YOLOv7

89.12

89.36

88.25

7.01

74.76

0.001

YOLOv8

90.17

90.28

89.71

5.29

86.11

0.003

Our model

93.22

92.37

92.99

2.98

16.01