Table 3 Detailed accuracy for each incremental session on the DB100 dataset

From: DFS: Dual-branch forward-looking simulation network for incremental learning of ancient Chinese characters

Method

Accuracy in each session↑

  
 

0

1

2

3

4

5

6

7

8

PD↓

ΔPD

Finetune

93.256

85.562

76.395

73.761

69.002

64.82

61.116

57.812

54.848

38.408

+33.698

CEC25

96.067

95.118

95.611

94.949

94.985

93.885

92.957

91.228

90.447

5.62

+0.91

CLOM46

97.07

96.18

95.66

94.33

94.08

92.18

90.36

88.08

86.28

10.79

+6.08

S3C26

95.341

94.585

93.697

90.795

89.885

87.441

84.608

82.689

82.255

13.086

+8.376

FACT7

96.906

96.602

95.981

94.844

94.59

92.973

91.652

89.84

89.644

7.262

+2.552

SAVC24

95.804

95.296

94.506

92.229

92.06

90.153

88.50

87.176

87.299

8.505

+3.795

Ours

96.639

96.612

96.284

95.425

95.325

94.311

93.50

91.946

91.899

4.74

 
  1. The bold values in the last row represent the recognition accuracy of our proposed method in the incremental stage, and the bold values in the last column represent the improved accuracy of our proposed method in the forgetting rate indicator compared with other methods.
Back to article page