Scanned Images: 3,482 Documents Across 10 Categories for OCR

Housing 3,482 scanned document images across 10 distinct categories, published by prabhats0605 in February 2026. It provides real-world document samples with varied layouts and scanning qualities specifically for OCR and Vision-Language Model (VLM) training.

Use Cases

• Fine-tuning Vision-Language Models (VLM) using the 10 document category labels
• Training OCR engines to handle varied scanning quality and document layouts
• Visual document retrieval tasks based on document type classification

Strengths

• 3,482 total images
• 10 distinct document categories
• Captures real-world scanning artifacts and varied layouts
• MIT licensed for permissive use

Limitations

• Relatively small sample size for deep learning compared to industrial-scale OCR datasets
• Lack of detailed column or metadata schema documentation

Provenance

Source

prabhats0605

Freshness

Last updated February 2026.