24 million image-caption pairs and 30 million image-references extracted from 6 million unique open-source biomedical articles. The dataset is annotated with over 27 unique metadata fields per data point, including article-level information. It was created by BIOMEDICA and last updated on Hugging Face in January 2025.
Use Cases
- Training biomedical vision-language models based on the large-scale image-caption pairs.
- Developing image retrieval systems for scientific literature based on the 30 million image-references.
- Fine-tuning models for biomedical concept recognition based on the extensive metadata annotations.
- Conducting research on scientific figure analysis based on the dataset's scale and source material.
Strengths
- Large scale with over 24 million image-caption pairs.
- Extensive metadata with over 27 unique fields per data point.
- Sourced from a broad base of 6 million unique open-source articles.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- BIOMEDICA, aggregating from open-source biomedical articles.
- Collection Method
- Likely web-scraped and processed from scientific publishing platforms.
- Time Range
- null
- Freshness
- Last updated 2025-01-22 01:06:11; freshness should be verified.
- Geography
- null