MegaHan97K Dataset

• We introduce MegaHan97K, a mega-category, large-scale dataset that contains the largest 97,455 Chinese character categories.
• MegaHan97K includes Chinese characters of 97,455 categories, which significantly surpasses existing datasets with at least six times larger categories and holds the largest volume.
• MegaHan97K pioneers to support the latest Chinese GB18030-2022 standard, ensuring the most comprehensive coverage and compatibility with modern Chinese processing systems.
• MegaHan97K contains four distinct subsets: handwritten, mouse-written, historical, and synthetic. Each subset contains a greater number of character categories compared to existing datasets, resulting in remarkable scale and diversity advantages.
• MegaHan97K effectively mitigates long-tail distribution issues by providing a balanced and sufficient number of samples for each category, ensuring robust training and validation of CCR models.

🔥 Download

Setting	Dataset	status
General CCR	GoogleDrive / BaiduYun:mbns	Released
Zero-Shot CCR	GoogleDrive / BaiduYun:6pxi	Released

🛠️ Usage

• Clone this repo:

git clone https://github.com/SCUT-DLVCLab/MegaHan97K.git

• Execute the following command to obtain example samples from the MegaHan97K dataset.

python MegaHan_Dataloader.py

Note: If you wish to access the entire dataset, please contact us via the email of the first author listed in the paper to obtain the decryption password.

• To access the entire dataset, please first download it, update the data_root in the python MegaHan_Dataloader.py script and then execute

python MegaHan_Dataloader.py

🌄 Gallery

• Illustration of the handwritten-original data in MegaHan97K

• Illustration of the handwritten-augmented data in MegaHan97K

• Illustration of the M⁵HisDoc data in MegaHan97K

• Illustration of the Kangxi dictionary data in MegaHan97K

• Illustration of the mouse-written-original data in MegaHan97K

• Illustration of the mouse-written-augmented data in MegaHan97K

• Illustration of the synthetic data in MegaHan97K

💙 Acknowledgement

• M⁵HisDoc
• FontDiffuser

License

MegaHan97K should be used and distributed under Creative Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) License for non-commercial research purposes.

Copyright

• This repository can only be used for non-commercial research purposes.
• For commercial use, please contact Prof. Lianwen Jin (eelwjin@scut.edu.cn).
• Copyright 2024, Deep Learning and Vision Computing Lab (DLVC-Lab), South China University of Technology.