Mindtech’s Digital Humans 2.0 aims to mitigate bias in synthetic datasets

Mindtech, the developer of a synthetic data creation platform for training computer vision systems, has announced the launch of its Digital Humans 2.0 as part of its DataOps Platform.

Mindtech creates digital humans with varied digital DNA elements like age, physical size and features associated with ethnicity such as facial build and features. It uses the Monk skin tone scale to measure diversity in its skin variance.

Synthetic data like the avatars created by generative AI models is produced from an algorithm and is still subject to potential bias. The models may be trained on datasets of a lower visual quality or that have inherent biases that can affect performance.

“The ability to create diverse digital humans in precise and controlled ways is revolutionary for the creation of data sets,” said Mindtech VP of product Chris Longstaff. “Our platform, analyses and identifies bias in existing datasets, and allows data scientists to rapidly address diverse digital humans, improving AI model accuracy.”

Some potential use cases include in-cabin automotive testing, retail spaces like Just Walk Out stores, and identifying lost children in a shopping mall, the release states.

Article Topics

biometric-bias  |  biometrics  |  computer vision  |  dataset  |  Mindtech  |  synthetic data