vAIsual introduces consented biometric video series to dataset shop for AI training
U.S.-based training data provider vAIsual has launched a new biometric video dataset series to supply fully consented material to fuel the development of machine learning algorithms.
The new datasets are made up of videos with 4k resolution of full-body biometrics and postures from models who have signed releases for their biometric data.
Non-compliance with existing privacy and copyright laws are the AI industry’s greatest challenge, according to vAIsual CEO, Michael Osterrieder.
“Cleanwashing has (become) the norm amongst generative AI companies who are trying to overcome the challenges of training their AI models on scraped data. We are the only company who can provide full dataset disclosure that demonstrates 100 percent legal compliance,” says Osterrieder in the announcement.
The videos have been added to vAIsual’s Dataset Shop, which up until now has offered only still images for training facial recognition and other AI algorithms.
“Adding video opens up a whole range of options for AI teams,” comments Osterrieder. “To begin with, the video itself captures thousands of individual frames that can be extracted into separate images. These stills provide a huge amount of variations in expression and movement.”
The datasets can be used for training generative and diagnostic algorithms, in addition to biometric ones, the company notes in the announcement. The biometric image datasets are designed specifically to be added to workflows for AI training by engineers.
vAIsual also recently signed a deal to balance the demographic makeup of its datasets with stock photos from Asia.