metadata

license: apache-2.0

Dataset Information

TAL_OCR_Composed_37K is a composed set of TAL OCR datasets. The original images and labels are downloaded from TAL website (https://ai.100tal.com/dataset), includes mainly

K12 Handwritten Chinese texts, English texts, Math formulas.
Printed K12 materials.

Images are tiled randomly to have a more compact view. There are total 32K images and text pairs after processing:

TAL_OCR_CHN/composed (645 images)
TAL_OCR_ENG/composed (399 images)
TAL_OCR_MATH/composed (420 images)
TAL_K12_PRINT/composed (17109 images)
HME100K/composed (18625 images)

Label format

{"image": "path_to_image", "ocr" : "text", "data_name": "xx"}

Citation

Credits to the TAL team (https://github.com/Phymond/HME100K).

@article{yuan2022syntax,
  title={Syntax-Aware Network for Handwritten Mathematical Expression Recognition},
  author={Yuan, Ye and Liu, Xiao and Dikubab, Wondimu and Liu, Hui and Ji, Zhilong and Wu, Zhongqin and Bai, Xiang},
  journal={arXiv preprint arXiv:2203.01601},
  year={2022}
}