Hi, I'm Dimitrios, but most people, including me, prefer to call me Dimitris. I am a PhD candidate in the
Department of Computing at
Imperial College London, supervised by
Professor Stefanos Zafeiriou.
My research lies at the intersection of Machine Learning, Computer Vision, and Graphics, with a particular focus on
Digital Humans.
3D Human ModelingDiffusion ModelsVideo ModelsDistillation
Research Interests
Controllable Limited-View 3D Avatars
Creating animatable 3D avatars from sparse visual observations, with a particular emphasis on single, in-the-wild
images. I am interested in combining modern 3D representations with strong generative priors to reconstruct observed
appearance while plausibly completing unobserved regions. A central goal is to produce highly realistic avatars that
remain controllable and can reproduce facial expressions, speech-driven motion, head pose, and broader human motion.
Diffusion Models
Large-scale (foundation) image diffusion models as powerful generative priors, as well as diffusion models that operate
directly in 3D representation spaces, including meshes and other structured geometric representations. I am particularly
interested in how their learned priors can be transferred to downstream avatar generation tasks.
Video Models
Controllable generative video models for human-centric tasks such as animation and reenactment, driven by explicit conditioning
signals such as audio and parametric head-model parameters, as well as learned implicit features, with a focus on their potential
for real-time, streaming 2D avatars.
Distillation
Distilling the knowledge of strong image and video diffusion models into 3D, through both optimization-based approaches such as score
distillation sampling and self-distillation into efficient generative models for 3D representations, so their potential carries over
to 3D generation.
PII-Free Solutions
Human data is inherently sensitive. One of my priorities is reducing reliance on real-person data by fine-tuning large generative priors on
synthetic human data, aiming to achieve realistic results while minimizing the use of personally identifiable data and bridging the synthetic-to-real domain gap.
News
Latest first
Feb 2026Started as a Student Researcher at Google AR/VR, London!
Jun 2025Joined Meta Reality Labs, London, as a Research Scientist Intern!
@article{gerogiannis2025arc2avatar,
title={Arc2Avatar: Generating Expressive 3D Avatars from a Single Image via ID Guidance},
author={Gerogiannis, Dimitrios and Papantoniou, Foivos Paraperas and Potamias, Rolandos Alexandros and Lattas, Alexandros and Zafeiriou, Stefanos},
journal={arXiv preprint arXiv:2501.05379},
year={2025}
}
@inproceedings{gerogiannis2024animateme,
title={AnimateMe: 4D facial expressions via diffusion models},
author={Gerogiannis, Dimitrios and Papantoniou, Foivos Paraperas and Potamias, Rolandos Alexandros and Lattas, Alexandros and Moschoglou, Stylianos and Ploumpis, Stylianos and Zafeiriou, Stefanos},
booktitle={European Conference on Computer Vision},
pages={270--287},
year={2024},
organization={Springer}
}
