Coherent US Tax, Health, and Employment Forms for Document AI Training

A synthetic dataset of 3,000 page-level images and annotations for three US forms: IRS Form 1040, CMS-1500, and USCIS Form I-9. Created by Symage, the dataset includes FUNSD ground truth and LayoutLMv3-ready annotations. The forms are filled with consistent synthetic identity information across all three document types.

Use Cases

• Train document layout analysis models based on the provided word boxes and entity labels.
• Benchmark key-value linking performance on structured forms using the provided ground truth.
• Fine-tune multimodal models like LayoutLMv3 using the token/bbox/tag view.
• Develop synthetic data pipelines for generating coherent, multi-form document sets.

Strengths

• Contains 3,000 page-level image and annotation rows.
• Includes three distinct, commonly used US forms filled with coherent synthetic data.
• Provides multiple annotation formats: FUNSD ground truth and LayoutLMv3-ready views.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset is fully synthetic, which may not capture all nuances of real-world handwritten or scanned forms.

Provenance

Source

Symage

Collection Method

Fully synthetic generation.

Freshness

Last updated 2026-07-08 16:21:55; freshness should be verified.

Geography

United States