LumiGauss: Relightable Gaussian Splatting in the Wild
We propose a 2D Gaussian Splatting model tht decouples the environment lightning and intrinsic properties of the captured subject. After training, we can relight the scene using a …
Joanna Kaleta
Professional Summary
About Me
I received PhD degree with distinctions on the 30th September 2025 at Warsaw University of Technology. I’m currently working at Cantina Labs as Machine Learning Engineer/Researcher.
My research revolves around 3D human animation and reconstruction using radiance fields and Gaussan splatting. My interests include computer graphics and application of machine learning methods in common 3D algorithms.
Education
PhD in Artificial Intelligence
Warsaw University of Technology
PhD in Artificial Intelligence (Discontinued)
Wrocław University of Science and Technology
MSc in Data Science
Wrocław University of Science and Technology
BSc Artificial Intelligence
Wrocław University of Science and Technology
Interests
Computer Graphics
Computer Vision
Machine Learning
Featured Publications
•
•
1 min read
BlendFields: Few-Shot Example-Driven Facial Modeling
We generate realistic 3D avatars from just as few as five multi-view frames. Compared to the previous approaches, which may use up to 1 million images, our approach is …
CoNeRF: Controllable Neural Radiance Fields
We extend neural 3D representations to allow for intuitive and interpretable user control beyond novel view rendering (i.e. camera control). We demonstrate, to the best of our …
Publications
(2025).
LumiGauss: Relightable Gaussian Splatting in the Wild.
In WACV 2025.
(2023).
BlendFields: Few-Shot Example-Driven Facial Modeling.
In CVPR 2023.
(2021).
CoNeRF: Controllable Neural Radiance Fields.
In CVPR 2022.
(2021).
TrajeVAE: Controllable Human Motion Generation from Trajectories.
In arXiv.
(2021).
Representing Point Clouds with Generative Conditional Invertible Flow Networks.
In Pattern Recongition Letters.
Experience
Machine Learning Engineer
CantinaWorking on the lipsyncing part of our product. I’m responsible for the planning and experimenting with our approach.Research Scientist Intern
Reality Labs at MetaDeveloped an algorithm for Continuous Levels of Detail for the Gaussian Splatting represention. We submitted our work to the CVPR 2025 conference, and it’s currently under a review.Research Scientist Intern
Microsoft Research CambridgeDesigned an approach to modeling facial wrinkles from a few, multi-view static frames of person. Our work was accepted to CVPR 2023.Read about our work hereVisiting Internation Research Student
University of British ColumbiaAn exciting collaboration with Kwang Moo Yi and Andrea Tagliasacchi which lasts to this day. We managed to extend learnable radiance fields with controllability capabilities for a post-hoc artistic editions. Our work was accepted to CVPR 2022.Read about our work here
Education
PhD in Artificial Intelligence
Warsaw University of TechnologyWorking on “Few-Shot Neural Human Rendering from Sparse Information”, supervised by Tomasz Trzcinski, Marek Kowalski in collaboration with Andrea Tagliasacchi and Kwang Moo Yi.PhD in Artificial Intelligence (Discontinued)
Wrocław University of Science and TechnologyWorking on 3D reconstruction from single-view or point clouds. During that time, I published our paper: “UCSG-Net - Unsupervised Discovering of Constructive Solid Geometry Tree” at NeurIPS 2020.Read about our work hereMSc in Data Science
Wrocław University of Science and TechnologyThesis on alternative synaptic signal aggregation (other than typical product-sum).BSc Artificial Intelligence
Wrocław University of Science and TechnologyThesis on American Sign Language (ASL) recognition from a webcam. Developed an algorithm that was published later at ICAISC 2018.Read about our work here
Honours
Serving as a reviewer at other conferences
CVPR 2025 NeurIPS 2024 CVPR 2024 3DV 2024 ICCV 2023
•
1 min read
Teaching