Michael Zollhöfer

Also known as Michael Zollhoefer, M. Zollhofer, M. Zollhöfer, Michael Zollhofer

Michael Zollhöfer is a computer vision researcher whose work connects real-time scene reconstruction with digital human avatars. He earned a doctorate at Erlangen in 2014, worked at Max Planck and Stanford, and joined Facebook in 2019. His work includes Scene Representation Networks, DeepCap, AutoAvatar, Codec Telepresence, and systems for text-driven and controllable avatars. As of October 2026, he is a Director and Research Scientist at Meta Reality Labs Research.

Now at: Meta
Works: 176
Citations: 20,600
h-index: 67
Badges: Rising
Bridge

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

By 2019

From Erlangen to Facebook

He completes a PhD at the University of Erlangen-Nuremberg in 2014 for work on real-time reconstruction of static and dynamic scenes. Before joining Reality Labs Research, he works as a postdoctoral researcher at the Max Planck Institute for Informatics and a Visiting Assistant Professor at Stanford University. He joins Facebook in summer 2019; his research focuses on reconstructing and analyzing the world at frame rate from visual input.
2019

Scene Representation Networks

With Vincent Sitzmann and Gordon Wetzstein, he coauthors Scene Representation Networks (SRNs) in 2019, a continuous 3D-structure-aware representation that encodes scene geometry and appearance and trains from posed 2D images and camera poses. NeurIPS gives the paper an Outstanding New Directions Paper Honorable Mention.
2020

DeepCap earns a CVPR mention

He coauthors DeepCap in 2020, a weakly supervised method for dense human performance capture from monocular video. CVPR gives it a Best Student Paper Honorable Mention.
2022

AutoAvatar and neural fields

In 2022, he coauthors AutoAvatar, which uses autoregressive modeling of implicit neural fields to capture dynamic effects such as soft-tissue deformation from raw scans.
February 2024

Leading Codec Telepresence teams

In February 2024, he is a Director at Reality Labs Research in Pittsburgh and leads six research and engineering teams focused on Codec Telepresence. His group develops the system with computer vision, machine learning and graphics through neural reconstruction and rendering.
February 2025

Text-driven talking avatars

In February 2025, he coauthors AV-Flow, an audio-visual generative model that animates photorealistic 4D talking avatars from text and synthesizes speech with synchronized lip motion and facial expression.
March 2025

Drivable 3D Gaussian Avatars

In March 2025, he coauthors Drivable 3D Gaussian Avatars (D3GA), a multilayered 3D model that uses Gaussian primitives in tetrahedral cages to control human bodies.
As of October 2026

Director at Meta Reality Labs

In October 2026, he coauthors No Distillation Needed, which introduces FaceGAN for real-time talking heads. FaceGAN emits facial expression and head pose in one forward pass per frame. As of October 2026, he is a Director and Research Scientist at Meta Reality Labs Research, working across neural rendering, computer vision, machine learning and computer graphics.