Machine learning algorithms developed to artificially generate realistic whole-body images of people

Technology creates high-resolution 1024 x 1024 moving images of models against a white background

Researchers at a Japanese startup have developed machine learning algorithms that can create realistic whole-body images of artificially created people.

DataGrid is a spin-out from Japan's Kyoto University. The company recently released a video on its website to reveal the capabilities of its newly-developed technology.

The tool morphs and presents model after model with dissimilar faces as their outfits transform from winter coats to bomber jackets in the video.

DataGrid revealed that both of the algorithms it has developed are first trained on real-life images of human models.

According to researchers, the algorithms are a type of Generative Adversarial Network (GAN) - a technology commonly used to create imitations of something that exists in the real world. DataGrid uses GAN to automatically create high-resolution 1024 x 1024 pictures of models against a white background.

While similar tools have been used in the past to produce facial images, this is the first time that any algorithm has generated high-precision whole-body images, including faces, hairstyles, and clothes of artificially generated people.

AI & Machine Learning Live is returning to London on 3rd July 2019. Hear about the latest developments in AI and machine learning from the Met Office's Charles Ewen, AutoTrader lead data scientist Dr David Hoyle and the BBC's Noriko Matsuoka, among many others. Attendance is free to qualifying IT leaders and senior IT pros, but places are limited, so reserve yours now.

DataGrid revealed that both of the algorithms it has developed are first trained on real-life images of human models. After the initial training is completed, one algorithm is tasked to repeatedly generate pictures of human models, while its counterpart works to distinguish the original image from the fake image.

Then, the first algorithm refines its fakes until the second algorithm becomes unable to identify differences between the real image and the AI-generated image.

According to DataGrid, the new technology could be licensed to fashion and advertising firms looking to create photogenic models - saving money on models and the cost of photo shoots - to exhibit their new products.

The research team is also working to further refine their algorithms to increase the range of motions for virtual models.

"We will further improve the accuracy of the whole-body model automatic generation AI and research and develop the motion generation AI," claimed the company in an official announcement.

"In addition, we will conduct demonstration experiments with advertising and apparel companies to develop functions required for actual operation."

Computing and CRN have united to present the Women in Tech Festival UK 2019, on 17 September in London.

The event will celebrate successful women in the IT industry, enabling attendes to hear about, and to share, personal experiences of professional journeys and challenges.

Whether you're the ‘Next Generation', an ‘Inspirational Leader', or an ‘Innovator of Tech' this event will offer inspiration on not only how to improve yourself, but how to help others too. The event is FREE for qualifying IT pros, but places will go fast