Wojciech Zielonka

AI Research Scientist · Meta — Health, Agents & Robotics · Pittsburgh, PA

I am a researcher specializing in conversational agents, LLMs, and digital humans, with a focus on human motion, behavior, reasoning and appearance.

Previously TUM Niessner Lab · MPI-IS NCS Group · Meta Reality Labs · Google Syntec
News
1

How to Build Digital Humans? From Priors to Photorealistic Avatars

A comprehensive state-of-the-art report on building digital humans, covering the full pipeline from 3D morphable models and parametric body priors to photorealistic neural avatars. We survey methods for face, head, and full-body avatar creation, including mesh-based representations, neural radiance fields, and 3D Gaussian splatting. The report provides a structured overview of capture setups, geometric priors, appearance modeling, and animation techniques, serving as a guide for researchers and practitioners working on digital human technology.

Code ... arXiv
Eurographics2026
2

PhysHead: Simulation-Ready Gaussian Head Avatars

Realistic digital avatars require expressive and dynamic hair motion; however, most existing head avatar methods assume rigid hair movement. These methods often fail to disentangle hair from the head, representing it as a simple outer shell and failing to capture its natural volumetric behavior. In this paper, we address these limitations by introducing PhysHead, a hybrid representation for animatable head avatars with realistic hair dynamics learned from multi-view video. At the core is a 3D Gaussian-based layered representation of the...

Code ... arXiv Video
CVPR2026
3

SynShot - Synthetic Prior for Few-Shot Drivable Head Avatar Inversion

We present SynShot, a novel method for the few-shot inversion of a drivable head avatar based on a synthetic prior. We tackle two major challenges. First, state-of-the-art monocular avatar models struggle to generalize to new views and expressions, lacking a strong prior and often overfitting to a specific viewpoint distribution. Second, training a controllable 3D generative network requires a large number of diverse sequences, for which pairs of images and high-quality tracked meshes are not always available. Moreover, under General...

Code ... arXiv Video
CVPR2025
4

GEM - Gaussian Eigen Models for Human Heads

Current personalized neural head avatars face a trade-off: lightweight models lack detail and realism, while high-quality, animatable avatars require significant computational resources, making them unsuitable for commodity devices. To address this gap, we introduce Gaussian Eigen Models (GEM), which provide high-quality, lightweight, and easily controllable head avatars. GEM utilizes 3D Gaussian primitives for representing the appearance combined with Gaussian splatting for rendering. Building on the success of mesh-based 3D morphable face models (3DMM), we define GEM as an ensemble of...

Code ... arXiv Video
CVPR2025
5

D3GA - Drivable 3D Gaussian Avatars

We present Drivable 3D Gaussian Avatars (D3GA), a multi-layered 3D controllable model for human bodies that utilizes 3D Gaussian primitives embedded into tetrahedral cages. The advantage of using cages compared to commonly employed linear blend skinning (LBS) is that primitives like 3D Gaussians are naturally re-oriented and their kernels are stretched via the deformation gradients of the encapsulating tetrahedron. Additional offsets are modeled for the tetrahedron vertices, effectively decoupling the low-dimensional driving poses from the extensive set of primitives to...

Code ... arXiv Video
3DV2025
6

GAN-Avatar - Controllable Personalized GAN-based Human Head Avatar

Digital humans and, especially, 3D facial avatars have raised a lot of attention in the past years, as they are the backbone of several applications like immersive telepresence in AR or VR. Despite the progress, facial avatars reconstructed from commodity hardware are incomplete and miss out on parts of the side and back of the head, severely limiting the usability of the avatar. This limitation in prior work stems from their requirement of face tracking, which fails for profile and...

Code ... arXiv Video
3DV2024
7

INSTA - Instant Volumetric Head Avatars

For immersive telepresence in AR or VR, we aim for digital humans (avatars) that mimic the motions and facial expressions of the actual subjects participating in a meeting. Besides the motion, these avatars should reflect the human's shape and appearance. INSTA enables avatar reconstruction within approximately 10 minutes from a single RGB camera video, driven at interactive frame rates. We optimize a deformable neural radiance field embedded in a multi-resolution grid around a 3D face model that guides deformations, supporting...

Code ... arXiv Video
CVPR2023
8

MICA - Towards Metrical Reconstruction of Human Faces

Face reconstruction and tracking is a building block of numerous applications in AR/VR, human-machine interaction, as well as medical applications. Most of these applications rely on a metrically correct prediction of the shape, especially when the reconstructed subject is put into a metrical context. We present MICA, a metrical face shape estimation method that leverages a face recognition network pretrained on large-scale 2D data for robust features. By training a face shape estimator in a supervised fashion on a unified...

Code ... arXiv Video
ECCV2022
9

Vulkan Path Tracer Sandbox

This project is an engine where you can create/test new shaders in real-time physically based ray tracing system. The main idea is that users can test their own integrators in predefined environment with assets like: meshes, materials, lights and textures. The Vulkan pipeline is abstracted in a shaders form and the only concern for a user is providing either their own scene description or using already existing one. Vulkan Path Tracer Sandbox Wojciech Zielonka Technical University of Munich Code ......

Code ...
10

Software Rasterizer

CPU implementation of a very efficient rasterizer which utilizes AVX2 instructions and lock-free multi-threading programming. Using tiled rendering the program consumes 8 pixels at the same time, which together with std::execution threading library makes it very fast and robust. The viewer of the project is a very simple OpenGL applicatoin which renders a texture to a quad. The texture is generated by the CPU Renderer.h class. The pipeline contains most of the modern GPU rasterizers stages. Software Rasterizer Wojciech Zielonka...

Code ...
11

Parallel Mesh Simplification For Planar Surfaces

This work elaborates a parallel algorithm based on quadric error metric and adaptive thresholding to simplify a triangle mesh. The approach emphasizes planar surfaces as a target to simplify. The main goal was to create a framework able to produce high-quality progressive meshes based on reconstructed ones from the environment. Those meshes are characterized by constant resolution of reconstruction, therefore storing vertices for planar surfaces evenly across the shape, which is not necessary. Parallel Mesh Simplification For Planar Surfaces Wojciech...

Code ... Report
12

Audio-driven Neural Rendering of Portrait Videos

This work is a master thesis which proposes a novel way of synthesizing audio-driven portrait videos. We show that photo-realistic images can be rendered based on a small, fully connected neural network with the positional encoding of 3D face surface and additional audio-features extracted from an arbitrary English speech. The method is based on the intermediate geometry of 3DMMs. However, it is restricted neither by any face model in particular nor by its expression or identity space. The pipeline predicts...

Master Thesis
13

RGB Face Tracking and Reconstruction

Photometric based markerless facial tracking using analysis-by-synthesis approach for 3DMMs face parameters optimization. All the computational parts of our pipeline were ran on GPU using CUDA and OpenGL. Our energy function consists of a sparse landmark term, a dense photometric term and a regularizer term. This energy function is minimized using iteratively reweighted least squares method (IRLS). Each Gauss-Newton update is solved using preconditioned conjugate gradients method (PCG). RGB Face Tracking and Reconstruction Mustafa Işık, Patrick Radner, Wojciech Zielonka Technical...

Code ...