Biz Doc VQA: 624 Japanese Receipt Question-Answer Pairs

624 Japanese-language Visual Question Answering annotations across 116 receipt images for business document OCR evaluation. Created by icoxfog417 and updated in March 2026, the collection focuses on extracting structured data from financial documents.

Use Cases

• Training models to extract financial data from receipts using the 'question' and 'answers' fields
• Evaluating spatial localization of OCR results using 'answer_bbox' coordinates
• Fine-tuning Japanese-language Document VQA models

Strengths

• 624 question-answer pairs for targeted document analysis
• Normalized bounding box coordinates (0-1 range) provided in 'answer_bbox' for precise localization
• Specialized Japanese language support for document OCR tasks

Limitations

• Small sample size of 116 images may limit model generalization
• Restricted to Japanese language text only
• Lack of diversity in document types as all samples are receipts

Provenance

Source

icoxfog417 via Hugging Face

Collection Method

Annotated

Freshness

Last updated March 2026.

Geography

Japan