Visual AI for Extended Reality

Overview

The international research group Visual AI for Extended Reality, led by Prof. Helge Rhodin, advances 3D computer vision, machine learning, and computer graphics to enable new levels of non-intrusive computer-human interaction.

The latest AR devices offer incredible display capabilities; we make them truly useful by complementing the display capabilities with algorithms for high-quality dynamic reconstruction. From just a video, we reconstruct the user's avatar and their environment. The goal is immersion through a tight coupling of the real and virtual worlds, enabled by a bidirectional information flow both in the device-to-user (display) and the user-to-device (reconstruction) direction.

The research is human-centric and aims at doing good. The three core research directions are privacy-preserving sensors, self-supervised learning, and how to add control to personalized human and large-scale generative models.

Follow us on X or LinkedIn to hear about job openings and the most recent research outcomes.

Directions

The three core research directions are privacy-preserving vision, self-supervised learning, and generative models for human modeling (click links for details). The group collaborates with industry and international researchers to make these algorithmic advances practical, e.g. with neuroscientists to relate laboratory animal behavior to neural activation.

Teaching

All courses taught by the Visual AI for Extended Realities group can be found in the ekVV. The group typically offers courses on the following topics:

• Visual Computing

• Neural and Geometric Modeling

• Probabilistic Learning for Visual Computing

• Video Game Programming

Theses

We always seek highly motivated and talented students to write their thesis in the Visual AI for Extended Realities group. Reach out to the member where you see the best fit, e.g. ask PhDs and Postdocs working in your interest area for possible topics.

It is recommended that you attend at least one lecture of the working group in advance.

Source Code

The group published source code and models for almost all publications. Check out our most recent publications and follow the link that is typically given in the abstract.
For instance, we provide code and examples for Neural Point Characters (NPC) and for AutoLink: Self-supervised Learning of Human Skeletons.

PhD and Postdoc Positions

Currently, only self-funded, e.g. with a DFG/DAAD scholarship, are possible. More positions are expected later this year. Follow Helge Rhodin on X and LinkedIn to be notified.

Research Assistant Positions

Reach out to Helge Rhodin.

Publications focus on graphics, vision, and machine learning venues, such as SIGGRAPH, CVPR, and NeurIPS. Visit the publication highlights on helge.rhodin.de for a graphical overview. The bibliography below lists publications in text form.