vAIsual signs deal for Asian photos to increase biometric training data diversity
vAIsual, the company behind the largest visual dataset collection in the world, has signed a deal to access images of Asian diaspora to help companies train machine learning algorithms like those in facial recognition. The deal was made with Vietnamese stock photo production studio Dragonimages and will help companies using VAIsual’s data to increase diversity and representation of their AI output, vAIsual says in a press release on Thursday.
The company’s Dataset Shop is a marketplace for visual media designed for AI training purposes and contains thousands of biometric photos alongside synthetic images. The online store was launched in 2022 to allow companies to purchase legally clean datasets through partnerships with stock image agencies. vAIsual created the store to solve the issue of scraping photos illegally from the internet without the consent of copyright owners.
The online store initially sold the largest dataset of photos captured with consent for use in biometrics training, consisting of over 600,000 high-quality images that are custom shot for AI development.
The deal with Dragonomics, a part international holding Everypixel Labs, comes months after a similar agreement with stock photography service Middle Frame selling access to images of Middle Eastern people for use in training algorithms.
“The only way to ensure diversity in the datasets to train AI is to make sure the datasets used to train AI are as inclusive as possible,” vAIsual CEO Michael Osterrieder says in the announcement. “Working with Dragonimages helps us to broaden our scope of visual images, and represent the Asian diaspora as best we can.”
For Dragonimages, the deal represents a chance to participate in the rapidly growing text-to-image generation industry, the company said.