/OCR-AEC5k

Dataset of AAAI 2020 paper "Accurate Structured-Text Spotting for Arithmetical Exercise Correction"

Creative Commons Attribution 4.0 InternationalCC-BY-4.0

Introduction

AEC-5k is a dataset focusing on text spotting of arithmetic exercises. The exercise images cover the entire grades of Chinese primary students (usually from age 6 to 12). This dataset contains 77, 098 expressions. Annotation of each expression has two attributes: 1) a bounding box that covers the entire AE even if it is multi-lined, and 2) a char-level text annotation.

Paper: https://ojs.aaai.org/index.php/AAAI/article/view/5410/5266 [Link]

Data download link: Link code: u1uw

Some Notes:

  • Annotations of several expressions on the edge of images are deleted, which aims to ensure the semantic compactness of expressions.

Annotation Format:

The annotation format for each bounding box is: (CHARS, x1, y1, x2, y2, 1), which indicates 1) character sequences, 2) coordinates of top left corner, 3) coordinates of bottom right corner and 4) a placeholder. Supported expressions are listed as follows:

  • Single-lined expressions, like 1+1=2, 3米=300厘米.
  • Vertical expressions, like: !30#×31#&!30#30!#&930, here ! indicates a space symbol, & indicates a horizontal line and # stands for a newline symbol.
  • Multi-lined expressions, like 1+2+3#=4+5#=6.
  • Fractions, like \frac{1}{3}/\frac{1}{3}=1.
  • Formulas, like x+1=2.

Supported Symbols

  • 1234567890()[].+-$*÷=><≈≥≤≠□
  • x, y, z
  • \sqrt、\frac、#、&、!、|
  • '年', '月', '星期', '日', 天', '时', '分', '秒', '毫秒'
  • '吨', '千克', '克', '毫克'
  • '元', '角', '分'
  • '亿', '万', '千', '百', '十', '个'
  • '千米', '公里', '米', '分米', '厘米', '毫米'
  • '平方千米', '平方公里', '公顷', '平方米', '平方分米', '平方厘米', '平方毫米'
  • '升', '毫升', '立方米', '立方分米', '立方厘米'
  • 't', 'kg', 'g', 'mg'
  • 'km', 'm', 'dm', 'cm', 'mm', 'L', 'mL' and related square / cube operations.

Release Updates

  • Sep 23, 2021: release v1.0

Copyright

Tencent is pleased to support the open source community by making the dataset available. Copyright (C) 2021 THL A29 Limited, a Tencent company.
The images and related annotations are licensed under a Creative Commons Attribution 4.0 International License.

Citation

If you find AEC-5k useful in your research, please consider citing:

@inproceedings{hu2020accurate,
  title={Accurate Structured-Text Spotting for Arithmetical Exercise Correction},
  author={Hu, Yiqing and Zheng, Yan and Liu, Hao and Jiang, Deqiang and Liu, Yinsong and Ren, Bo},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={34},
  number={01},
  pages={686--693},
  year={2020}
}

Contact

For any question, please file an issue or contact

Yiqing Hu: hooverhu@tencent.com
Yan Zheng: neoyzheng@tencent.com