HubertKompanowski
Building generative systems for 3D scenes, perception, and real-world autonomy.
About Me
I’m a researcher and engineer bridging Generative AI and Physical Intelligence. As a PhD candidate at Trinity College Dublin, I build Generative 3D pipelines that can power creative media and autonomous system simulators.
My background is grounded in shipping code for critical systems. Previously, I engineered computer vision stacks for Mars rovers and highway surveillance, gaining deep insight into real-world AV perception challenges. I've also deployed medical diagnostic models and hold 8 U.S. patents.
Whether solving long-tail generation for synthetic data or designing neural assets for 3D environments, my goal is to make generative models reliable for real-world application.
Research Interests
Generative World Models
Simulation for AVs & Robotics
3D & Video Generative AI
Assets for Games, VR & Media
Robotics & AVs Vision
End-to-End Vision & Control
Medical AI Vision
Imaging & Diagnostics
Selected Research
FROMAT: Tri-Stream Feature Routing for Material Appearance Transfer in Generative Novel View Synthesis
FROMAT is a lightweight appearance transfer technique for multiview diffusion models. We use self-attention adaptation for high-fidelity material transfer from single reference image.
Dream-in-Style: Text-to-3D Generation using Stylized Score Distillation
Dream-in-Style is a method for stylizing text-to-3D generation. We introduce stylized score distillation to guide a text-to-3D optimization process to output visually plausible geometry and appearance.