OpenDocVQA: A Unified Corpus for Document Visual Question Answering

A training and evaluation corpus for VDocRAG, a retrieval-augmented generation framework designed to understand real-world documents from visual features. The dataset is a unified collection of open-domain document visual question answering data, encompassing diverse document types and formats. It was created by NTT-hil-insight and last updated on 2025-05-26.

Use Cases

• Training document-understanding models based on the described visual features of diverse real-world documents.
• Evaluating the performance of RAG frameworks like VDocRAG on open-domain visual question answering.
• Benchmarking model capabilities across different document types and formats mentioned in the description.

Strengths

• Described as the first unified collection of open-domain document VQA datasets.
• Encompasses diverse document types and formats, suggesting broad applicability.
• Specifically created for training and evaluating the VDocRAG framework.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license information are unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the source documents used for compilation.

Provenance

Source

NTT-hil-insight via Hugging Face

Collection Method

Likely compiled from multiple existing document VQA datasets to create a unified corpus.

Freshness

Last updated 2025-05-26 02:58:12; freshness should be verified.