PD12M: a fully open image training dataset with community governance

Spawning has released PD12M, a fully open dataset consisting of 12.4 million image-caption pairs. The dataset exclusively consists of public domain and CC0 licensed images that have been obtained from Wikimedia Commons, a large number of cultural heritage organizations, and the iNaturalist website. From the paper accompanying the release:

We present Public Domain 12M (PD12M), a dataset of 12.4 million high-quality public domain and CC0-licensed images with synthetic captions, designed for training text-to-image models. PD12M is the largest public domain image-text dataset to date, with sufficient size to train foundation models while minimizing copyright concerns. Through the Source.Plus platform, we also introduce novel, community-driven dataset governance mechanisms that reduce harm and support reproducibility over time.

The release of PD12M is remarkable not only given the size of the fully open dataset but also because of the holistic approach that Spawning has taken. Via the source.plus platform, Spawning provides community-based governance mechanisms. In addition, the platform also provides an exemplary level of transparency regarding the sources of the images included in the dataset.

The release of PD12M is exciting not only because it builds on our ideas for a public data commons but also because Spawning sees the release of the dataset as a first step towards offering a foundational public domain image model with no IP concerns, that will help artists to fine-tune, and own, their own models on their own terms.