HRT Transgender Dataset

Created in 2013, the HRT Transgender Dataset helps facial recognition systems to identify users of Hormone Replacement Therapy (HRT) transitioning or who have already transitioned from one gender to another. 

The dataset comprises 10,000 images of 38 people, with an average of 278 images per subject taken from publicly available YouTube videos 'under real-world conditions, including variations in pose, illumination, expression, and occlusion'. 

The dataset gained notoriety in August 2017 for data scraping without the knowledge of permission of those whose data was included. 

Dataset creator Karl Ricanek, a professor of computer science at the University of North Carolina at Wilmington, claimed the set was developed in order to protect against the possibility of terrorists using HRT to avoid facial recognition and sneak across borders undetected.

In July 2022, researchers Os Keyes and Jeanie Austin published a peer-reviewed audit of the project's background and practices in Big Data & Society which took issue with a number of Ricanek's practices and claims, including:

• That the real reason for the dataset was to strengthen national security - which they deride as 'ludicrous'.

• That Ricanek gained the consent of all those people whose videos he used - which appears not to be the case.

• That the data was shared not commercially - even though Ricanek's research is funded by the FBI and US Army.

• That only the dataset images were distributed to third parties - but the videos were available via an unprotected Dropbox URL, including those that had been made private or deleted.

• That he stopped giving access to the dataset in 2017 - though it was still accessible on Dropbox.

• That the researchers must hacked Dropbox to access the files - in fact, they gain acess via a UNCW public records request.