The highly diverse computer vision training set reduces bias concerns by spanning genders, BMI, and ethnicities
Synthesis AI, a pioneer in synthetic data technologies, released 40,000 unique high-resolution 3D facial models. Through the company’s synthetic data-as-a-service FaceAPI solution, users can now programmatically create perfectly labelled image training data spanning 40,000 unique identities. Demonstrating Synthesis AI’s commitment to addressing ethical AI issues related to bias and privacy, this data set not only represents the largest collection of 3D facial models available anywhere, but also is the most diverse, spanning gender, ethnicity, age, and BMI.
“On the heels of our recent funding announcement, we are excited to continue this type of growth and momentum as a company,” said Yashar Behzadi, CEO of Synthesis AI. “Our goal is to address bias and privacy in AI and to democratise access to high-quality data. Making 40,000 unique identities available further strengthens this mission, while also addressing the technical, economic, and ethical issues with current approaches.”
The new capability will allow companies of any size to create high-performing and unbiased facial models with access to more robust data. Early customers include three of the top five handset manufacturers, leading teleconferencing companies, and global technology companies building the next generation of smart assistants.
Also Read: The Tale of Facial Recognition Technology
By bringing together cinematic VFX pipelines and novel generative AI models, Synthesis AI is uniquely able to produce high-quality and diverse 3D models of faces. Each identity can be modified by near-infinite variability through the combination of emotion, head pose, hair, facial hair, accessories, environments, and camera attributes. Each image comes with associated pixel-perfect labels such as segmentation, facial landmarks, depth maps, surface normals, and more. The ability to create large diverse datasets has recently enabled leading handset manufacturers to develop improved facial verification systems that work with user mask wear across environments and camera angles. Companies across industries and use-cases will be able to build more capable and less biased models, further solidifying Synthesis AI as a leader in synthetic data by providing more capabilities than competitors.
“Releasing a data set of this breadth and depth reflects Synthesis AI’s commitment to pushing the boundaries of computer vision,” said Dr Rana el Kaliouby, Co-Founder and CEO of Affectiva. “Brands have a unique opportunity to address ethical AI issues related to bias and privacy and build better, more capable models. We’re proud to be working with Synthesis AI to pioneer synthetic data technologies that will strengthen and facilitate that trust.”
