metadata

language:
  - zh
language_bcp47:
  - zh-tw
license: mit
task_categories:
  - image-to-text
  - document-question-answering
pretty_name: SynthDoG Traditional Chinese Dataset
size_categories:
  - 10K<n<100K
tags:
  - ocr
  - synthetic-data
  - traditional-chinese

SynthDoG Traditional Chinese Dataset

This dataset contains synthetic document-ground truth pairs for Traditional Chinese text recognition training. The dataset is generated using the SynthDoG (Synthetic Document Generation) framework, which creates realistic document images with Traditional Chinese text.

Dataset Structure

The dataset is organized into three splits:

train/: Training data
validation/: Validation data
test/: Test data

Each split contains:

Image files (*.jpg): Synthetic document images with Traditional Chinese text
metadata.jsonl: Ground truth annotations for each image in JSONL format

File Format

Images

Format: JPEG
Resolution: Various sizes, optimized for document recognition
Content: Synthetic documents with Traditional Chinese text
Features: Includes various document layouts, fonts, and text styles

Annotations (metadata.jsonl)

The metadata file contains annotations for each image in JSONL format, including:

Text content
Text regions
Layout information

Usage

This dataset is designed for:

Training OCR models for Traditional Chinese text recognition
Fine-tuning document understanding models
Testing document layout analysis systems

Loading the Dataset

You can load this dataset using the Hugging Face datasets library:

from datasets import load_dataset

dataset = load_dataset("LeeTung/synthdoc-zh-tw-dataset")

License

MIT License. Please refer to the original SynthDoG repository for additional license information.

Citation

If you use this dataset in your research, please cite the original SynthDoG paper and this dataset:

@misc{synthdoc-zh-tw-dataset,
  title={SynthDoG Traditional Chinese Dataset},
  author={Lee Tung},
  year={2024},
  publisher={Hugging Face}
}