Meta recently launchedfacet The AI application database (FAirness in Computer Vision Evaluation) will be open-sourced for external use, allowing different AI visual recognition models to analyze and determine whether there are biases such as racial discrimination.
The FACET database consists of 3.2 photos, including images of 5 people of different races. Each photo is labeled by human experts, covering physiological characteristics such as age, gender, skin color, hairstyle, and is also labeled and classified according to types such as occupation.
In addition, this database also contains racial attribute labels of approximately 1 people wearing masks from the OpenDataLab open source data platform SA-6.9B, allowing more artificial intelligence models to learn and train through this open source database.
Meta created this database in the hope of improving AI recognition and preventing biases such as racial discrimination in AI applications. By making this database open source, Meta believes it will help researchers label similar data and promote equal access to AI technology.
