Synthetic Chinese Document Images For OCR-Free Understanding

Encompassing 0.5 million synthetic Chinese document images generated by the SynthDoG tool for training the Donut model. It is part of a multi-language collection created by naver-clova-ix and was last updated in January 2024.

Use Cases

• Train the Donut transformer model on synthetic Chinese document images for OCR-free understanding tasks.
• Fine-tune vision-language models using the Chinese-language image-text pairs generated by SynthDoG.
• Benchmark document understanding models on a large-scale synthetic Chinese dataset of 0.5M samples.

Strengths

• Contains 0.5 million synthetic samples, providing substantial scale for model training.
• Part of a multi-language collection including English, Japanese, and Korean variants, enabling comparative studies.
• Specifically designed for the Donut (OCR-Free Document Understanding Transformer) model, ensuring task relevance.

Limitations

• Data is entirely synthetic, which may not fully capture the noise and variation of real-world documents.
• The specific content, document types, and annotation schema within the images are not described.
• No information is provided on the distribution of document layouts or text complexity.

Provenance

Source

naver-clova-ix

Collection Method

Synthetically generated by the SynthDoG tool.

Freshness

Last updated on 2024-01-31.

Geography

Content is in Chinese, but geographic origin of source templates is unspecified.