4th May 2019
AI generates entire photorealistic people
DataGrid, a Japanese startup company based at Kyoto University, has developed a new AI system capable of smoothly and automatically generating high-quality and high-resolution (1024 x 1024) photorealistic models of entire people.
This was achieved with Generative Adversarial Networks (GANs) – a type of machine learning technique, which consists of two neural nets that "compete" against each other, creating a feedback loop of increasingly higher-quality output.
Previous attempts to create photorealistic people with GANs have focused on just their faces. This new system, however, can generate complete bodies including faces, hair, limbs, clothes and shoes. The images will be used as virtual models for the advertising and fashion industries, though this technology will inevitably raise concerns about its potential for deepfakes.
"We will further improve the accuracy of the whole-body automatic model generation, and research and develop motion-generation AI," the DataGrid team state on their website. "We will conduct demonstration experiments with advertising and apparel companies to develop functions required for operation."