AI will now be able to create 3D Avatars from any 2D pictures

Researchers of Imperial College London and, a startup working on machine learning shared details regarding AvatarMe in a research paper submitted to the Conference on Computer Vision and Pattern Recognition (CVPR) 2020.

According to the team, AvatarMe will be able to recreate photorealistic 3D images from ‘in-the-wild’ photos. They further said that the AI system would be able to surpass other existing protocols with a significant margin.

To render the 3D faces, the researchers took the pore-level reflectance maps of 200 people’s mugs using an LED sphere rig with 168 lights and 9 DSLR cameras. They then used it to train GANFIT, an AI model to create realistic maps from the textures and optimize for an identity match.

The researchers further said that AvatarMe didn’t show any signs of artifacts in the final renderings and were able to handle even the most extreme poses and add-ons like sunglasses.

Unfortunately, the biggest drawback of AvatarMe is its inability to reconstruct faces of dark-skinned objects. It is also able to give more accurate results if the input photograph is a high-resolution image.

Nevertheless, AvatarMe is an advancement in the art-generating AI system. And we can expect the technology to go further with innovative approaches.

Read next: Facebook reveals the results of the Deepfake Detection Challenge and shows some promising prospects
Previous Post Next Post