|
Nicolas von Lützow I'm a PhD candidate in the TUM Visual Computing and AI Lab under the supervision of Prof. Matthias Nießner. Before starting my PhD I studied Informatics (M.Sc.) at TUM and Computer Science (B.Sc.) at TUHH. My research interests lie in 3D generation, scene reconstruction, and novel view synthesis. |
Publications |
|
GaussianGPT: Towards Autoregressive 3D Gaussian Scene Generation
Nicolas von Lützow, Barbara Rössle, Katharina Schmid, Matthias Nießner arXiv, 2026 project page / arXiv We introduce GaussianGPT, a transformer-based model that directly generates 3D Gaussians via next-token prediction, enabling full 3D scene generation, completion, and outpainting through autoregressive sampling. |
|
LinPrim: Linear Primitives for Differentiable Volumetric Rendering
Nicolas von Lützow, Matthias Nießner NeurIPS, 2025 project page / arXiv We develop a differentiable volumetric rendering framework that uses transparent polyhedral primitives to represent scenes and render them in real-time. The primitives offer a bounded and compact representation that improves performance on specular areas and achieves better depth estimates while remaining compatible with advances developed for 3D Gaussians. |
|
This website is based on Jon Barron's source code. |