Dataset · HuggingFaceM4
Race-balanced FairFace facial dataset
HuggingFaceM4/FairFace
108,501 facial images annotated by race, gender, and age to combat bias in computer vision.
What's inside
The dataset contains 108,501 face photos collected from the YFCC-100M collection on Flickr. The main feature of the set is a balanced distribution across 7 racial groups (Caucasian, Black, Indian, East Asian, Southeast Asian, Middle Eastern, and Latino). Each image is annotated by race, gender, and age group.
What you can do with it
- Training demographic attribute classifiers (gender, age, race) from photos
- Auditing and reducing algorithmic bias in facial recognition systems
- Validating generative image models for representativeness
- Size
- 100K–1M rows
- Downloads this month
- 6.2K
- Likes
- 40
Models trained on it
- LITVA2006/chihuahua-monk-skin-toneImage classification 49
- bykof/peekaboolean-450mImage text → text 18
- LITVA2006/shinshilla-monk-skin-toneImage classification 5
- evaks/fairface-age-detection-model-cardImage classification 0
- snakajima/face-lensImage classification 0
- wesjos/fairface-halfpint-cnano-3task-112Image classification 0
- wesjos/fairface-papercarousel-cnano-3task-224Image classification 0
- wesjos/fairface-arrowhead-cnano-race7-224Image classification 0
How to load it
from datasets import load_dataset
ds = load_dataset("HuggingFaceM4/FairFace")