The Fair Human Image Benchmark, or FHIBE (pronounced fee-bee), is a human-centric image dataset that is responsibly and ethically sourced to help advance higher ethical standards for data collection in AI. It was created for the purpose of evaluating and mitigating bias in computer vision models.

FHIBE is the first publicly available, demographically diverse, fairness benchmark with appropriate informed consent, licensing, compensation, and diverse, global representation.​ The current dataset available for download is an extended version of FHIBE first detailed in research featured on the cover of Nature Magazine.

The FHIBE dataset was created as an initiative of Sony. Watch the film A Fair Reflection to learn more about the motivation and process of creating FHIBE.

More information about the dataset can be found at:​ fairnessbenchmark.ai.sony​

You can find FHIBE's data card and transparency documentation here: fairnessbenchmark.ai.sony/data-card

For support related to these datasets, including consent revocation, please contact [email protected]

Note for Previous Downloaders:

If you previously downloaded the dataset from the Sony website, you will need to register separately with Mozilla Data Collective (MDC) to download the dataset there. Your previous Sony registration will not carry over to MDC