Edoardo A. Dominici

About

I am a research engineer at Huawei in Zürich, Switzerland currently working at the intersection of computer graphics and generative models, with a particular focus on image-based scene generation.

Before joining Huawei I worked on production renderers for animated movies: at Animal Logic (then Netflix Animation Studios) where I was part of the Rendering R&D team and at Tangent Animation on a custom fork of Blender's Cycles. You can find me in the credits of Leo, The Magician's Elephant and Maya and the Three.

I hold a MSc from the University of British Columbia, where I worked on image vectorization algorithms under the supervision of dr. Alla Sheffer. My BSc thesis was on WebGL image retargeting under the supervision of dr. Marco Tarini.

Publications

Scene Generation at Absolute Scale: Utilizing Semantic and Geometric Guidance From Text for Accurate and Interpretable 3D Indoor Scene Generation

Stefan Ainetter, Thomas Deixelberger, Edoardo A. Dominici, Philipp Drescher, Konstantinos Vardis, Markus Steinberger

ArXiv 2026

TL;DR: We use a metric-scale geometric proxy to drive scene generation through multi-view diffusion models. We introduce a multimodal panorama generation model to generate starting frames from which we efficiently sample trajectories for world exploration and for 3DGS optimization.

Arxiv

PixelSync: Enhancing Texture Generation via Explicit Feature Synchronization

Philipp Drescher, Edoardo A. Dominici, Giacomo Nazzaro, Konstantinos Vardis, Stefan Hauswiesner, Markus Steinberger

I3D 2026

TL;DR: Training-free multi-view consistent texture synthesis via attention and latent synchronization in diffusion models.

(Coming soon)

DreamAnywhere: Object-Centric Panoramic 3D Scene Generation

Edoardo A. Dominici, Jozef Hladky, Floor Verhoeven, Lukas Radl, Thomas Deixelberger, Stefan Ainetter, Philipp Drescher, Stefan Hauswiesner, Arno Coomans, Giacomo Nazzaro, Konstantinos Vardis, Markus Steinberger

WACV 2026 Oral

TL;DR: Modular 3D scene generation from text using 360° panoramas, object-centric decomposition, and hybrid inpainting for immersive navigation and editing.

Project Arxiv Cite

Real-time Rendering with a Neural Irradiance Volume

Arno Coomans, Giacomo Nazzaro, Edoardo A. Dominici, Christian Döring, Floor Verhoeven, Konstantinos Vardis, Markus Steinberger

Eurographics 2026, Computer Graphics Forum

TL;DR: We introduce an irradiance formulation for neural network-based radiance caching which improves quality and compression compared to existing probe based methods.

Arxiv Cite

Real-time Neural Rendering of Dynamic Light Fields

Arno Coomans, Edoardo A. Dominici, Christian Döring, Joerg H. Mueller, Jozef Hladky, Markus Steinberger

Eurographics 2024, Computer Graphics Forum

TL;DR: Introduce a temporal extension to neural radiosity and show how dynamic light fields can be cached into compact neural networks and encodings for real-time rendering.

Paper Cite

Focus Range: Production Ray Tracing of Depth of Field

Edoardo A. Dominici, Emanuel Schrade, Basile Fraboni, Luke Emrose, Curtis Black

SIGGRAPH Asia 2023 Technical Communications

TL;DR: Novel piece-wise linear ray tracing formulation for defocus blur. We integrated it into a production path tracer supporting glossy materials and reflections so that the Leo character could be focused through the looking glass <3

Project Paper Cite

PROVIDE: A Probabilistic Framework for Unsupervised Video Decomposition

Polina Zablotskaia, Edoardo A. Dominici, Leonid Sigal, Andreas Lehrmann

UAI 2021, Proceedings of Machine Learning Research

TL;DR: Formulate a video as a generative image formation model from multiple objects, then introduce spatio-temporal cues in a variational optimization framework through a 2D-LSTM.

Project Cite

PolyFit: Perception-Aligned Vectorization of Raster Clip-Art via Intermediate Polygonal Fitting

Edoardo A. Dominici, Nico Schertler, Jonathan Griffin, Shayan Hoshyari, Leonid Sigal, Alla Sheffer

SIGGRAPH 2020, ACM Transactions on Graphics 39(4)

TL;DR: We introduce an image vectorization algorithm that leverages an intermediate polygonal approximation for regularity detection and to guide the curve fitting.

Project Code Cite

Perception-Driven Semi-Structured Boundary Vectorization

Shayan Hoshyari, Edoardo A. Dominici, Alla Sheffer, Nathan Carr, Zhaowen Wang, Duygu Ceylan, I-Chao Shen

SIGGRAPH 2018, ACM Transactions on Graphics 37(4)

TL;DR: We introduce a vectorization algorithm that alternates corner detection and curve fitting guided by perceptual principles to generate clean, artist-friendly boundary curves.

Project Demo Cite

Contact

Email: edoaramis [at] gmail.com

Location: Huawei Research Center, Zürich, Switzerland