Meta introduced an artificial intelligence database called FACET

Recently, Meta introduced an artificial intelligence database called FACET (FAirness in Computer Vision EvaluaTion) and made it publicly accessible as an open-source tool, allowing various AI visual recognition models to be scrutinized for biases, particularly racial discrimination.

Comprising 32,000 images, the FACET database encompasses representations of 50,000 individuals across varied racial backgrounds. Each image has been meticulously annotated by human experts, denoting physiological characteristics such as age, gender, skin tone, and hair type. Additionally, classifications based on professions have been designated.

Moreover, this repository incorporates data from OpenDataLab’s open-source platform, SA-1B, featuring annotations of approximately 69,000 individuals donning masks. This integration seeks to facilitate a broader spectrum of AI models to benefit from training using this open-source resource.

In launching this database, Meta’s ambition is to refine AI’s recognition capabilities, thereby mitigating any biases or racial disparities in AI technology applications. By open-sourcing the database, Meta believes it paves the way for researchers to undertake analogous data annotation, simultaneously promoting equity in AI technology applications.