Artist and researcher Branislav Ulicny (“AlteredQualia”) developed an online demo that combines two different deep learning-based projects into spooky cursor-tracking faces.
The first is a hobby project by Mike Tyka, who works on machine learning at Google, makes use of GANs to generate realistic new faces. Using a TITAN X GPU, Tyke trained his generative model on nearly 20,000 which were then used as the base for the portraits.
“It uses a technique called “generative adversarial networks” (“GAN”) where two artificial neural networks are playing an adversarial game: one (the “Generator”) tries to generate increasingly convincing output, while a the second (the “Critic”) tries to learn to distinguish real photos from generated ones. With time, the generated output becomes increasingly realistic, as both adversaries try to outwit each other,” Tyka told Mashable.
DeepWarp developed by computer vision researchers at the Skolkovo Institute of Science and Technology (Skoltech) in Russia, is used to move the eyeballs in the portraits. The realistic gaze manipulation software was trained using a TITAN X GPU and the cuDNN-accelerated Torch deep learning framework, and when combined with Tyka’s project, it produces some haunted-looking paintings.
The interactive “eyes gaze warping” online demo features 13 different characters.
Read more >
Related resources
- GTC session: Using Omniverse to Generate First-Person Experiential Data for Humanoid Robots
- GTC session: Navigating the Photorealistic AI Revolution
- GTC session: Open-World Segmentation and Tracking in 3D
- NGC Containers: Eye Contact
- NGC Containers: Audio2face (A2F)
- SDK: NVIDIA Tokkio