The dataset card
Everything a builder needs to decide whether to use this, and everything a contributor needs to see what they are part of. One page, no marketing.
AI Community Library
Dataset cardComposition
What is inside
Breadth matters more than volume. A thousand portraits of the same five people teaches a model very little.
Range of contributors
Children, young people, adults and elders. Different skin and hair presentations, different features, different body types.
Actual breakdown
Ordinary life
At work, in class, on a pitch, at home, in the street. Not a clinical backdrop, because that is exactly the association we are trying to break.
Scenes covered
Light and exposure
Very low contrast skin is genuinely hard to photograph and hard to detect. Varied lighting is what makes the set useful rather than decorative.
Lighting conditions covered
Roadmap
What comes next
The first release is a starting point. The library gets more useful as it covers more of the range of people who actually live with albinism. Target for the next release
Richer, more consistent descriptions so a model learns the right features from each image. Labelling standard in use
A held-back set builders can test against, so “we support albinism” becomes a measurable claim rather than a press line. Confirm whether this exists
Text, audio and video have the same gap. Confirm whether this is on the roadmap