DocVQA 2026: Multimodal Reasoning Across 8 Diverse Document Domains

Created by VLR-CVC for the ICDAR2026 competition, this multimodal dataset contains approximately 1,000 document-question pairs focused on complex reasoning. It spans eight distinct domains including business reports, scientific papers, slides, posters, maps, comics, infographics, and engineering drawings.

Use Cases

• Developing visual question answering models for technical engineering drawings
• Testing multimodal reasoning on non-linear layouts like comics and infographics
• Benchmarking document understanding across business and scientific domains

Strengths

• Includes 8 diverse document domains
• Features rare document types like engineering drawings and maps
• ICDAR2026 competition-grade reasoning challenges

Limitations

• Small sample size of ~1,000 records
• Lack of detailed column metadata in the current release

Provenance

Source

VLR-CVC

Collection Method

Annotated for the ICDAR2026 competition

Freshness

Last updated March 2026.