Locations
- Mountain View, CA
Research Topics
Neural Rendering
We look beyond “traditional” rendering techniques to generate convincing, photorealistic Artificial Humans. We use neural networks-based methodologies with various intermediate representations to deliver photorealistic streams.
Few-shot Learning
Few-shot learning enables the creation of personalized virtual beings with just a few images of a person. We’ve developed viable solutions to robustly train our research pipelines in a fast and cost-effective manner, while still delivering visual quality comparable to high data-demanding solutions.
3D Reconstruction
We use state-of-the-art approaches and, build on recent advancements in multi-texture mapping and reconstruction technology to recreate and render photo-realistic virtual humans. Our novel 3D fitting and reconstruction system yields the combined realism and fluid interactivity of our NEONs.
Behavior Modeling and Generation
Virtual humans shouldn’t only “look” realistic, they also need to “act” and “react" convincingly. Our research architecture contemplates how a virtual human should behave and move in believable ways. It takes advantage of the full gamut of NLP advancements and probes into the fields of computer graphics and animation.
Conversational AI
Every interaction is a unique, joint activity co-constructed by the participants. Our conversational engine combines several AI technologies to facilitate more contextual and personalized interactions with NEONs. When paired with behavior modeling and generation, conversational AI-powered NEONs can facilitate wholistic and human-like interactions.
