Old Book Illustrations Collection With Multilingual Captions

4,172 scanned illustrations sourced from old books. The collection was curated by the Old Book Illustrations website team for the BigLAM initiative. Captions and descriptions are primarily in English but include other languages.

Use Cases

• Train an image captioning model using the 4,172 illustrations and their associated multilingual text descriptions.
• Fine-tune a text-to-image generation model on the paired illustration and caption data.
• Analyze the distribution of languages (e.g., English, French, German) present in the captions and descriptions.
• Perform image-to-image translation tasks using the historical illustration style as a source domain.

Strengths

• Contains 4,172 unique scanned illustrations from a curated historical source.
• Data includes multilingual text annotations (English, French, German) for image-text tasks.
• Annotations were expert-generated, suggesting a degree of quality control.

Limitations

• The dataset size of 4,172 images is relatively small for training large vision-language models from scratch.
• The specific time period and geographic origin of the source books are not detailed, limiting historical analysis.
• Language distribution in captions is not quantified, which may lead to class imbalance in multilingual modeling.

Provenance

Source

Old Book Illustrations website, curated by its team.

Collection Method

Scraped from the website with permission for the BigLAM initiative.

Freshness

Last updated on the platform in December 2023.