Large-scale CelebFaces Attributes (CelebA) dataset

The Large-scale CelebFaces Attributes (CelebA) Dataset is a facial dataset developed by a team of researchers at the Chinese University of Hong Kong to help train and test computer vision applications such as facial analysis, facial recognition, and facial detection.

Released late 2015, the dataset consists of 202,599 images of over 10,000 mostly western celebrities, each annotated with 40 attributes such as moustache, beard, spectacles, and the shape of face and nose.

CelebA became a commonly used dataset and is seen to have helped make facial recognition and analysis tools more accurate. It has been referred to and cited in hundreds of academic studies and tests. 

Risks and harms 🛑

CelebA has been criticised for violating privacy and amplifying bias, and for its potential misuse to develop unauthorised facial recognition or deepfake applications which could lead to discrimination, identity fraud, and the perpetuation of harmful stereotypes.

Transparency and accountability 🙈

The Large-scale CelebFaces Attributes (CelebA) dataset is seen to suffer from several transparency limitations.

• Unclear data collection methods. The researchers behind CelebA have not disclosed how the dataset was compiled, whether licensing was complied with, or consent given by people appearing in the images.

• Labeling accuracy. Information about the process and accuracy of attribute labels - such as hair colour, and facial expressions - is poorly documented, meaning users may find it difficult to mitigate potential biases or errors.

• Potential for misuse. Other than banning commercial use, the creators of CelebA do not provide guidance on how the dataset should - and should not be used - meaning it could potentially be used to train inappropriate or unethical facial recognition systems or deepfake algorithms without the consent of the individuals in the images, raising privacy, copyright, disinformation and other concerns.

Research, advocacy 🧮

• Lingenfelter, B., Davis, S.R., Hand, E.M. (2022). A Quantitative Analysis of Labeling Issues in the CelebA Dataset

• Böhlen, M., Chandola, V., Salunkhe, A. (2017). Server, server in the cloud. Who is the fairest in the crowd?

News, commentary, analysis 🗞️

• https://www.technologyreview.com/2020/12/10/1013617/racism-data-science-artificial-intelligence-ai-opinion/

• https://www.biometricupdate.com/202011/report-says-lack-of-diversity-in-face-biometrics-datasets-extends-to-expression-emotion

• https://venturebeat.com/ai/researchers-find-evidence-of-bias-in-facial-expression-data-sets/

• https://www.nbcnews.com/tech/internet/facial-recognition-s-dirty-little-secret-millions-online-photos-scraped-n981921

• https://www.dailymail.co.uk/sciencetech/article-5035045/Creepy-video-shows-AI-faces-fake-celebrities.html