DocVQA: 10,000+ Document Images for Visual Question Answering

DocVQA consists of 10,000 to 100,000 document images paired with question-answer sets, formatted by lmms-lab in 2024. This version is derived from the original 2020 DocVQA research to facilitate standardized evaluation of Large Multi-modality Models (LMMs). It provides a structured framework for testing how models interpret text and layout within diverse document types.

Use Cases

• Benchmarking Large Multi-modality Models (LMMs) using the question and answer pairs
• Evaluating document-based visual question answering performance on raw images
• Testing zero-shot reasoning on document layouts and embedded text

Strengths

• 10,000 to 100,000 record scale
• Apache 2.0 licensed for open redistribution
• Standardized for lmms-eval pipeline integration

Limitations

• Temporal staleness of source documents dating back to 2020
• Formatted specifically for evaluation pipelines which may require restructuring for general training use

Provenance

Source

Mathew et al. (2020), DocVQA: A Dataset for VQA on Document Images

Freshness

Last updated April 2024; source material originates from 2020.