WikiVQABench: Human-Curated Visual Question Answering Benchmark

WikiVQABench is a human-curated benchmark for knowledge-grounded visual question answering. IBM Research constructed it by systematically combining Wikipedia images, article captions, and structured knowledge from Wikidata. Candidate multiple-choice questions were generated by large language models and then reviewed by human annotators for factual correctness and visual-text consistency.

Use Cases

• Benchmarking multimodal AI models based on knowledge-grounded visual question answering tasks.
• Training models to answer questions requiring external knowledge based on Wikipedia images and captions.
• Evaluating the factual correctness and visual-text consistency of AI-generated content.
• Researching the integration of structured knowledge from Wikidata with visual and textual data.

Strengths

• Human-reviewed for factual correctness and visual-text consistency.
• Systematically constructed from Wikipedia images, captions, and Wikidata knowledge.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Wikipedia images and captions, Wikidata.

Collection Method

Combined systematically; candidate questions generated by LLMs and curated by human annotators.

Freshness

Last updated 2026-05-04 23:47:11; freshness should be verified.