DocVQA Test Subsampled: 500 Document Image QA Pairs

A manually annotated test set of 500 question-answer pairs based on document images. The data originates from the UCSF Industry Documents Library and was curated for benchmarking by subsampling the original DocVQA test set. The dataset was last updated on June 20, 2025.

Use Cases

• Benchmarking document visual question answering models based on manually annotated QA pairs.
• Training models to answer questions about document content based on associated images.
• Evaluating the performance of multimodal AI systems on real-world document understanding tasks.

Strengths

• Contains 500 manually annotated question-answer pairs, ensuring a curated benchmark size.
• Data is sourced from the UCSF Industry Documents Library, providing a basis in real-world documents.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is limited to 500 subsampled pairs, which may not represent the full original test distribution.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

UCSF Industry Documents Library (original images); DocVQA dataset.

Collection Method

Images collected from source library; questions and answers manually annotated.

Freshness

Last updated 2025-06-20 15:49:00.