Scanned Document Images for OCR and VLM Fine-Tuning Across 10 Categories

Voxel51 provides 3,482 scanned document images across 10 diverse categories. The dataset is designed for training Optical Character Recognition systems and fine-tuning Vision-Language Models, featuring real-world documents with varied layouts and scanning quality. It was last updated on February 5, 2026.

Use Cases

• Fine-tuning Vision-Language Models based on the pairing of scanned document images and their textual content.
• Training Optical Character Recognition systems based on real-world scanned documents with varied layouts.
• Benchmarking document layout analysis algorithms based on the 10 diverse document categories.
• Improving document image preprocessing models based on the varied scanning quality present.

Strengths

• Contains 3,482 scanned document images, providing a substantial collection for model training.
• Covers 10 diverse document categories, suggesting breadth in document types.
• Features real-world scanned documents with varied layouts and scanning quality, likely improving model robustness.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for any associated metadata is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Voxel51

Freshness

Last updated 2026-02-05 18:46:43.