# Wojciech Zielonka - Personal Academic Website > Personal website of Wojciech Zielonka, AI Research Scientist at Meta (Health, Agents & Robotics) in Pittsburgh, PA. The site showcases research publications on 3D digital humans, neural avatars, conversational agents, LLMs, multimodal AI, embodied AI, and human motion, behavior, reasoning and appearance. ## About - Name: Wojciech Zielonka - Role: AI Research Scientist at Meta — Health, Agents & Robotics, Pittsburgh, PA - Ph.D.: "Capturing and Synthesis of 3D Digital Humans", TU Darmstadt (defended August 2025) - M.Sc.: "Audio-driven Neural Rendering of Portrait Videos", TU Munich (March 2021) - Previous affiliations: MPI-IS Tubingen, TU Munich (Niessner Lab), Google (Thabo Beeler's Syntec group), Meta Reality Labs (Michael Zollhofer's group) - Website: https://wojciechzielonka.com - GitHub: https://github.com/zielon - Google Scholar: https://scholar.google.com/citations?user=aITcRswAAAAJ&hl=en - LinkedIn: https://www.linkedin.com/in/wojciech-zielonka-99011a352 - Email: wojtas.zielonka@gmail.com ## Research Areas 3D digital humans, head avatars, 3D Gaussian splatting, neural radiance fields, face reconstruction, 3D morphable models, neural rendering, physically based rendering ## Pages - [Home](https://wojciechzielonka.com/): Project listing with research publications and engineering projects - [Documents](https://wojciechzielonka.com/documents/): Interactive PDF archive of papers, posters, theses, and talks ## Publications ### How to Build Digital Humans? From Priors to Photorealistic Avatars - URL: https://wojciechzielonka.com/how-to-build-digital-humans/ - Conference: Eurographics 2026 (State-of-the-Art Report) - Authors: Wojciech Zielonka, Tobias Kirschstein, Timo Bolkart, Simon Giebenhain, Vanessa Sklyarova, Xiang Deng, Donglai Xiang, Shunsuke Saito, Yebin Liu, Matthias Niessner, Justus Thies - Code: https://github.com/Zielon/how-to-build-digital-humans - Description: Comprehensive survey covering the full pipeline from 3D morphable models and parametric body priors to photorealistic neural avatars. Covers face, head, and full-body avatar creation including mesh-based representations, neural radiance fields, and 3D Gaussian splatting. ### SynShot - Synthetic Prior for Few-Shot Drivable Head Avatar Inversion - URL: https://wojciechzielonka.com/synshot/ - Conference: CVPR 2025 - Authors: Wojciech Zielonka, Stephan J. Garbin, Alexandros Lattas, George Kopanas, Paulo Gotardo, Thabo Beeler, Justus Thies, Timo Bolkart - Code: https://github.com/Zielon/synshot - arXiv: https://arxiv.org/abs/2501.06903 - Video: https://youtu.be/4KQQatkaSgc - Description: Few-shot inversion of drivable head avatars using a synthetic prior. Fine-tunes a pretrained synthetic prior with 3-5 input images to create photorealistic avatars that generalize to novel expressions and viewpoints. Addresses GDPR compliance by using synthetic data. ### GEM - Gaussian Eigen Models for Human Heads - URL: https://wojciechzielonka.com/gem/ - Conference: CVPR 2025 - Authors: Wojciech Zielonka, Timo Bolkart, Thabo Beeler, Justus Thies - Code: https://github.com/Zielon/GEM - arXiv: http://arxiv.org/abs/2407.04545 - Video: https://youtu.be/sgG4K4I1TGw - Description: Lightweight and controllable head avatars using 3D Gaussian primitives. Defines an ensemble of linear eigenbases for head appearance via PCA distillation of CNN-based Gaussian avatar models. ### D3GA - Drivable 3D Gaussian Avatars - URL: https://wojciechzielonka.com/d3ga/ - Conference: 3DV 2025 - Authors: Wojciech Zielonka, Timur Bagautdinov, Shunsuke Saito, Michael Zollhofer, Justus Thies, Javier Romero - Code: https://github.com/facebookresearch/D3GA - arXiv: https://arxiv.org/abs/2311.08581 - Video: https://youtu.be/C4IT1gnkaF0 - Description: Multi-layered 3D controllable model for human bodies using 3D Gaussian primitives in tetrahedral cages. Compositional pipeline decomposes avatars into layers (garments, hands, faces). ### GAN-Avatar - Controllable Personalized GAN-based Human Head Avatar - URL: https://wojciechzielonka.com/ganavatar/ - Conference: 3DV 2024 - Authors: Berna Kabadayi, Wojciech Zielonka, Bharat Lal Bhatnagar, Gerard Pons-Moll, Justus Thies - Code: https://github.com/bernakabadayi/ganavatar - arXiv: https://arxiv.org/abs/2311.13655 - Video: https://youtu.be/uAi5IVrzzZY - Description: Person-specific animatable avatars from images without precise facial expression tracking. Maps 3DMM parameters to GAN latent space for animation control. ### INSTA - Instant Volumetric Head Avatars - URL: https://wojciechzielonka.com/insta/ - Conference: CVPR 2023 - Authors: Wojciech Zielonka, Timo Bolkart, Justus Thies - Code: https://github.com/Zielon/INSTA - arXiv: https://arxiv.org/abs/2211.12499 - Video: https://youtu.be/HOgaeWTih7Q - Description: Avatar reconstruction in ~10 minutes from a single RGB video, driven at interactive frame rates. Optimizes a deformable neural radiance field in an Instant-NGP multi-resolution grid around a 3D face model. ### MICA - Towards Metrical Reconstruction of Human Faces - URL: https://wojciechzielonka.com/mica/ - Conference: ECCV 2022 - Authors: Wojciech Zielonka, Timo Bolkart, Justus Thies - Code: https://github.com/Zielon/MICA - arXiv: https://arxiv.org/abs/2204.06607 - Video: https://youtu.be/vzzEbvv08VA - Description: Metrical face shape estimation from a single image. Supervised training on ~2300 identities using face recognition features. Outperforms SOTA by 15-24% on the NoW benchmark. Also provides a metrical face tracker. - Additional code: https://github.com/Zielon/metrical-tracker ## Engineering Projects ### Audio-driven Neural Rendering of Portrait Videos - URL: https://wojciechzielonka.com/face-neural-rendering/ - Description: Master thesis on synthesizing audio-driven portrait videos using a neural network with positional encoding of 3D face surface and audio features. ### Parallel Mesh Simplification For Planar Surfaces - URL: https://wojciechzielonka.com/mesh-simplification/ - Code: https://github.com/Zielon/ParallelQSlim - Description: Parallel algorithm based on quadric error metric for simplifying triangle meshes, emphasizing planar surfaces. ### Vulkan Path Tracer Sandbox - URL: https://wojciechzielonka.com/vulkan/ - Code: https://github.com/Zielon/PBRVulkan - Description: Real-time physically based ray tracing engine using Vulkan for creating and testing shaders. ### Software Rasterizer - URL: https://wojciechzielonka.com/rasterizer/ - Code: https://github.com/Zielon/CPURasterizer - Description: CPU rasterizer using AVX2 instructions and lock-free multi-threading with tiled rendering. ### RGB Face Tracking and Reconstruction - URL: https://wojciechzielonka.com/face2face/ - Code: https://github.com/isikmustafa/face-tracking - Description: Photometric-based markerless facial tracking using analysis-by-synthesis with CUDA and OpenGL.