Table 2 An overview of existing oracle bone inscription datasets categorized by tasks

From: Oracle bone inscriptions information processing: a comprehensive survey

Task

Dataset

Year

Type

Source

# Samples

# Classes

Resolution

Available

F2

Recognition

*YinQiWenYuandetection

2020

Rubbings

Book

9823

–

540 × 776

✓

jpg

 

OracleBone-800033

2020

Rubbings

Book

128,770

–

–

✗

–

 

ACCID52

2023

Handprinted

Book

15,085

2892

–

✗

–

 

O2BR2

2025

Original

Website

800

–

1529 × 1192

✓

jpg

Rejoining

OB-Rejoin34

2022

Rubbings

Book

998

249

–

✗

–

 

COBD54

2023

Rubbings

–

960

–

–

✗

–

 

OBI-rejoin2

2025

Orig. & Rubb.

Website

483

200

265 × 313

✓

png

 

OBFI35

2025

Original

Website

5374

110

721 × 816

✓

jpg

Classification & Retrieval

Oracle-20k43

2016

Handprinted

Website

20,039

261

–

✗

–

 

OBC30669

2019

Rubbings

Book

309,551

306

382 × 478

✓

bmp

 

Oracle-AYNU44

2019

Handprinted

–

39,062

2583

64 × 64

✗

–

 

HWOBC70

2020

Handprinted

Website

83,245

3881

400 × 400

✓

png

 

Oracle-50k71

2020

Handprinted

Website

59,081

2668

50 × 50

✓

jpg

 

*OBI-IJDH

2020

Rubbings

–

655

29

64 × 96

✓

png

 

Oracle-25072

2020

Handprinted

Website

92,160

250

–

✗

–

 

Radical-14872

2020

Handprinted

Website

108,989

148

–

✗

–

 

OBI12573

2022

Rubbings

Book

4257

125

278 × 473

✓

jpg

 

OBI-10030

2022

Handprinted

Book

4748

100

–

✗

–

 

Oracle-24175

2022

Handpri. & Rubb.

Website

78,565

241

588 × 700

✓

dmp

 

ORCD76

2022

Handpri. & Rubb.

–

6700

64

–

✗

–

 

OCCD76

2022

Handprinted

Website

62,186

1,320

–

✗

–

 

OracleRC51

2023

Rubbings

Book

2005

202

–

✗

–

 

Oracle-MNIST78

2024

Rubbings

Website

30,222

10

28 × 28

✓

idx

 

OBI component 2079

2024

Handprinted

Website

10,257

20

413 × 401

✓

png

Deciphering

OBI-ECC80

2022

Handprinted

Website

4860

972

105 × 105

✓

png

 

EVOBC81

2024

Handpri. & Rubb.

Website & Book

229,170

13,714

465 × 857

✓

png

 

HUST-OBC82

2024

Handprinted

Website & Book

140,053

10,999

400 × 520

✓

png

 

ACCP83

2024

Handpri. & Rubb.

Website & Book

346,344

88,901

103 × 129

✓

jpg & png

 

OracleSem60

2024

Handprinted

Website & Book

–

1762

–

✗

–

 

GEVOBC84

2025

Handprinted

Website & Book

3,780

765

105 × 105

✓

png

 

PD-OBS62

2025

Handpri. & Rubb.

Website & Book

211,796

–

209 × 236

✓

jpg & png

 

PictOBI-20k5

2025

Handprinted

Website & Book

15,175

80

512 × 512

✓

jpg

Emerging

RCRN85

2022

Rubbings

Book

1606

362

520 × 668

✓

png

 

OBIMD86

2024

Handpri. & Rubb.

Book

10,077

–

518 × 842

✓

jpg

 

RMOBS61

2025

Handprinted

Website

> 20,000

900

–

✗

–

 

Oracle-P15k3

2025

Handpri. & Rubb.

OBC306

14,542

239

128 × 128

✓

png

  1. Since both the classification and retrieval tasks of OBI rely on the category labels of individual characters, we merge the datasets of these two tasks. We report the average resolution of the images in the ‘Resolution’ column. ‘F2’ denotes the file format. We differentiate recognition datasets from classification datasets according to whether bounding-box annotations are provided. *The website references of YinQiWenYuandetection and OBI-IJDH are (https://jgw.aynu.edu.cn/home/down/detail/index.html?sysid=3) and (http://www.ihpc.se.ritsumei.ac.jp/OBIdataseIJDH.zip), respectively.