Public Domain Image-Text Pairs for Foundation Model Training

12.4 million image-caption pairs constitute the largest public domain image-text dataset for training foundation models. The dataset was created by Spawning and released in October 2024, as indicated by the associated arXiv paper identifier. It features community-driven governance mechanisms aimed at reducing harm and supporting reproducibility.

Use Cases

• Train CLIP-style contrastive models on public-domain image and text caption pairs.
• Fine-tune text-to-image generation models using the paired caption data.
• Benchmark multimodal alignment algorithms on a large-scale, copyright-cleared corpus.
• Analyze caption quality and diversity across millions of image-text samples.

Strengths

• 12.4 million pairs provides scale sufficient for foundation model training.
• Public domain licensing minimizes copyright concerns for commercial use.
• Includes governance mechanisms for community-driven dataset maintenance.

Limitations

• Specific image sources, caption quality metrics, and demographic or geographic biases are not detailed in the provided description.
• The dataset's internal structure, column names, and data splits are unknown.

Provenance

Source

Spawning, via the Source.Plus platform.

Collection Method

Collection method for the 12.4 million public domain image-caption pairs is not specified in the provided description.

Time Range

null

Freshness

Dataset page was last updated on January 9, 2025.

Geography

null