OmniMedVQA: A Medical Visual Question Answering Benchmark

OmniMedVQA is a large-scale Visual Question Answering benchmark for the medical domain, containing 118,010 images and 127,995 QA-items. The benchmark was collected from 73 different medical datasets and introduced by an author named foreverbeliever. It was last updated on 2024-04-30.

Use Cases

• Benchmarking medical VQA model performance based on the 127,995 QA-items.
• Training multimodal models for medical image interpretation based on the 118,010 images from authentic scenarios.
• Evaluating model generalization across different medical image modalities based on the 12 covered modalities.
• Studying anatomical region recognition in AI models based on references to more than 20 human anatomical regions.

Strengths

• Contains 118,010 images sourced from authentic medical scenarios.
• Includes 127,995 QA-items for comprehensive evaluation.
• Aggregates data from 73 different medical datasets.
• Covers 12 different medical image modalities and more than 20 human anatomical regions.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Freshness should be verified as the last update was 2024-04-30.

Provenance

Source

Aggregated from 73 different medical datasets.

Collection Method

Collected to create a medical VQA benchmark.

Freshness

Last updated 2024-04-30 13:34:50.