Bias in AI Computer Vision Systems Dramatically Reduced With Synthetic Data ‘Digital Humans’

Mindtech’s Digital DNA technology creates a diverse range of Digital Humans that allow for statistically fair representation of the eight billion real humans on the planet

Mindtech, the developer of the world’s leading end-to-end synthetic data creation platform for training AI vision systems, has announced the launch of its Digital Humans 2.0 within its Data Ops Platform.

To improve accuracy in understanding and identifying human behaviours, massive training dataset diversity is required. Mindtech’s digital humans vary the “Digital DNA” elements such as skin tone (using Monk skin tone scale to measure diversity), ethnicity (especially facial build and features), age, physical sizes, as well as clothing. This creates training data that can surpass the performance of real-world data alone.

Recommended AI News: Bluescape Streamlines Generative AI Workflows With Enhanced API Updates

Digital DNA technology has significant advantages compared to the “avatars” created by Generative AI models. These may be trained on datasets of unknown provenance and rights, have inherent biases, or often suffer from visual artefacts that severely impair model performance.

Recommended AI News: Scorpion Launches Industry-Leading AI Chat Solution Designed For Local Services Businesses

Mindtechs Digital Humans 2.0 can be used in numerous use cases, such as:

• In-cabin automotive testing, where companies can test a range of variations of heights, faces and behaviours.

• Retail spaces, such as ‘just walk out’ stores, to ensure algorithms are not biassed against age, race or sex.

• Help in cases such as lost child identification in a shopping mall.

Recommended AI News: Francisco Partners to Acquire The Weather Company Assets from IBM

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

[To share your insights with us, please write to sghosh@martechseries.com]