FACET
lastmod 2023-09-05created 2023-09-051 edits
Meta releases the FACET dataset for examining bias in computer vision models
Features
- Fairness in Computer Vision EvaluaTion
- A new comprehensive benchmark for evaluating the fairness of computer vision models across classification, detection, instance segmentation, and visual grounding tasks
- Consists of 32,000 images containing 50,000 people
- Labeled by expert human annotators for demographic attributes (e.g. gender, age), additional physical attributes (e.g. skin tone, hairstyle), and person-related classes (e.g. basketball player, doctor)
- Also includes 69,000 person, hair, and clothing labels from the SA-1B (Segment Anything) dataset
- FACET is for research evaluation purposes only and cannot be used for training
- Provided together with a dataset explorer
- Alongside this, the previously released vision model DINOv2 was relicensed under Apache 2.0