datasets.md 4.55 KB
Newer Older
root's avatar
root committed
1
2
3
4
## 数据集
这里整理了常用中文数据集,持续更新中,欢迎各位小伙伴贡献数据集~
- [ICDAR2019-LSVT](#ICDAR2019-LSVT)
- [ICDAR2017-RCTW-17](#ICDAR2017-RCTW-17)
Yipeng's avatar
Yipeng committed
5
- [中文街景文字识别](#中文街景文字识别)
root's avatar
root committed
6
7
8
- [中文文档文字识别](#中文文档文字识别)
- [ICDAR2019-ArT](#ICDAR2019-ArT)

MissPenguin's avatar
MissPenguin committed
9
除了开源数据,用户还可使用合成工具自行合成,可参考的合成工具包括[text_renderer](https://github.com/Sanster/text_renderer)[SynthText](https://github.com/ankush-me/SynthText)[SynthText_Chinese_version](https://github.com/JarveeLee/SynthText_Chinese_version)[TextRecognitionDataGenerator](https://github.com/Belval/TextRecognitionDataGenerator)等。
root's avatar
root committed
10
11
12
13

<a name="ICDAR2019-LSVT"></a>
#### 1、ICDAR2019-LSVT
- **数据来源**:https://ai.baidu.com/broad/introduction?dataset=lsvt
MissPenguin's avatar
MissPenguin committed
14
- **数据简介**: 共45w中文街景图像,包含5w(2w测试+3w训练)全标注数据(文本坐标+文本内容),40w弱标注数据(仅文本内容),如下图所示:  
MissPenguin's avatar
MissPenguin committed
15
16
17
18
    ![](../datasets/LSVT_1.jpg)  
    (a) 全标注数据  
    ![](../datasets/LSVT_2.jpg)  
    (b) 弱标注数据  
root's avatar
root committed
19
20
21
22
23
24
- **下载地址**:https://ai.baidu.com/broad/download?dataset=lsvt

<a name="ICDAR2017-RCTW-17"></a>
#### 2、ICDAR2017-RCTW-17
- **数据来源**:https://rctw.vlrlab.net/
- **数据简介**:共包含12,000+图像,大部分图片是通过手机摄像头在野外采集的。有些是截图。这些图片展示了各种各样的场景,包括街景、海报、菜单、室内场景和手机应用程序的截图。
25
    ![](../datasets/rctw.jpg)
root's avatar
root committed
26
27
28
- **下载地址**:https://rctw.vlrlab.net/dataset/

<a name="中文街景文字识别"></a>
Yipeng's avatar
Yipeng committed
29
#### 3、中文街景文字识别 
root's avatar
root committed
30
- **数据来源**:https://aistudio.baidu.com/aistudio/competition/detail/8
MissPenguin's avatar
MissPenguin committed
31
- **数据简介**:ICDAR2019-LSVT行识别任务,共包括29万张图片,其中21万张图片作为训练集(带标注),8万张作为测试集(无标注)。数据集采自中国街景,并由街景图片中的文字行区域(例如店铺标牌、地标等等)截取出来而形成。所有图像都经过一些预处理,将文字区域利用仿射变化,等比映射为一张高为48像素的图片,如图所示:  
32
    ![](../datasets/ch_street_rec_1.png)  
root's avatar
root committed
33
    (a) 标注:魅派集成吊顶  
34
    ![](../datasets/ch_street_rec_2.png)  
root's avatar
root committed
35
    (b) 标注:母婴用品连锁  
root's avatar
root committed
36
37
38
39
40
- **下载地址**
https://aistudio.baidu.com/aistudio/datasetdetail/8429

<a name="中文文档文字识别"></a>
#### 4、中文文档文字识别
root's avatar
root committed
41
42
- **数据来源**:https://github.com/YCG09/chinese_ocr  
- **数据简介**
root's avatar
root committed
43
44
45
46
    - 共约364万张图片,按照99:1划分成训练集和验证集。
    - 数据利用中文语料库(新闻 + 文言文),通过字体、大小、灰度、模糊、透视、拉伸等变化随机生成
    - 包含汉字、英文字母、数字和标点共5990个字符(字符集合:https://github.com/YCG09/chinese_ocr/blob/master/train/char_std_5990.txt )
    - 每个样本固定10个字符,字符随机截取自语料库中的句子
root's avatar
root committed
47
    - 图片分辨率统一为280x32  
48
49
50
    ![](../datasets/ch_doc1.jpg)  
    ![](../datasets/ch_doc2.jpg)  
    ![](../datasets/ch_doc3.jpg)  
root's avatar
root committed
51
52
53
54
55
- **下载地址**:https://pan.baidu.com/s/1QkI7kjah8SPHwOQ40rS1Pw (密码:lu7m)

<a name="ICDAR2019-ArT"></a>
#### 5、ICDAR2019-ArT
- **数据来源**:https://ai.baidu.com/broad/introduction?dataset=art
Yipeng's avatar
Yipeng committed
56
- **数据简介**:共包含10,166张图像,训练集5603图,测试集4563图。由Total-Text、SCUT-CTW1500、Baidu Curved Scene Text (ICDAR2019-LSVT部分弯曲数据) 三部分组成,包含水平、多方向和弯曲等多种形状的文本。
57
58
    ![](../datasets/ArT.jpg)
- **下载地址**:https://ai.baidu.com/broad/download?dataset=art
Yipeng's avatar
Yipeng committed
59

Yipeng's avatar
Yipeng committed
60
## 参考文献
Yipeng's avatar
Yipeng committed
61
**ICDAR 2019-LSVT Challenge**
Yipeng's avatar
Yipeng committed
62
```
Yipeng's avatar
Yipeng committed
63
64
65
66
67
68
@article{sun2019icdar,
  title={ICDAR 2019 Competition on Large-scale Street View Text with Partial Labeling--RRC-LSVT},
  author={Sun, Yipeng and Ni, Zihan and Chng, Chee-Kheng and Liu, Yuliang and Luo, Canjie and Ng, Chun Chet and Han, Junyu and Ding, Errui and Liu, Jingtuo and Karatzas, Dimosthenis and others},
  journal={arXiv preprint arXiv:1909.07741},
  year={2019}
}
Yipeng's avatar
Yipeng committed
69
```
Yipeng's avatar
Yipeng committed
70

Yipeng's avatar
Yipeng committed
71
**ICDAR 2019-ArT Challenge**
Yipeng's avatar
Yipeng committed
72
```
Yipeng's avatar
Yipeng committed
73
74
75
76
77
78
@article{chng2019icdar2019,
  title={ICDAR2019 Robust Reading Challenge on Arbitrary-Shaped Text (RRC-ArT)},
  author={Chng, Chee-Kheng and Liu, Yuliang and Sun, Yipeng and Ng, Chun Chet and Luo, Canjie and Ni, Zihan and Fang, ChuanMing and Zhang, Shuaitao and Han, Junyu and Ding, Errui and others},
  journal={arXiv preprint arXiv:1909.07145},
  year={2019}
}
Yipeng's avatar
Yipeng committed
79
```