Lukas Radl

Passionate about digitally representing the physical world

Currently in Zurich, CH

I'm a University Assistant/PhD Student at the Institute of Visual Computing, Graz University of Technology, where I mostly work on 3D Scene Representations and GenAI, supervised by Markus Steinberger.

Previously, I received my Master's degree in Computer Science (with distinction) and my Bachelor's degree in Software Engineering both at Graz University of Technology between 2018 and 2023. During this time, I worked as a study assistant, supporting various visual computing classes. In 2019 & 2020, I was featured on the Dean's List of my faculty (top 5% of students).

News

• 05/25: I'm joining Meta Reality Labs as a Research Scientist Intern!
• 02/25: VRSplat has been accepted to I3D'25.
• 02/25: Invited Presentation at CVWW'25!
• 01/25: I'm attending Wintergraph'25 in Kaprun, Austria!
• 07/24: I'll be in Denver for HPG'24 and SIGGRAPH'24 to present StopThePop!
• 06/24: I'm attending CVPR'24 in Seattle to present NeRF Internals and LAENeRF!
• 05/24: StopThePop has been accepted to SIGGRAPH'24 (Journal Track) and will appear in ACM TOG 43(4).
• 04/24: Our NeRF Internals paper has been accepted to the CVPR'24 Workshop on Neural Rendering Intelligence!
• 02/24: LAENeRF has been accepted to CVPR'24!
• 02/24: Starting my PhD at Graz University of Technology!

Research

I'm interested in computer graphics, computer vision, machine learning and parallel processing. Most of my work focuses on editable, view-consistent 3D scene representations.

We combine StopThePop, Optimal Projection and Mini-Splatting with single-pass foveated rendering for view-consistent real-time rendering of Gaussian Splats on HMDs.

We leverage the architecture of Instant-NGP to cache view-independent latent codes in a frustum voxel grid for high-quality real-time and offline rendering.

Sorting 3D Gaussians along view rays leads to increased view-consistency. With our hierarchical, GPU-friendly resorting scheme, our renderer is only 4% slower than 3DGS on average.

Locally Stylized Neural Radiance Fields via point-based 3D style transfer with geometry-aware losses - reduced background artefacts, more detail retention and view-consistency.

Using density estimates derived from activations for inverse transform sampling in NeRFs allows for faster inference and comparable visual quality.

Method and apparatus for aorta segmentation in Slicer3D, as well as an open-source CTA dataset with ground-truth annotations, to be used in research.

Reviewing

Since 2024, I have been reviewing for several journals and conferences!

• Computers & Graphics
• Computer Graphics Forum (CGF)
• IEEE VR
• Eurographics
• ICCV

Teaching

I have been involved in teaching for various visual computing courses: ☀️ denotes a course in the summer term, ❄️ denotes a course in the winter term.

Tutoring

Study Assistant work: Supporting students during the exercise, grading and interviews.

2022/23

• Real-Time Graphics: Exercise ❄️
• Mathematical Principles in Visual Computing: Exercise ☀️
• Computer Graphics: Exercise ☀️
• Computer Vision: Exercise ☀️

2021/22

• Real-Time Graphics: Exercise ❄️
• Computer Graphics: Exercise ☀️
• Computer Vision: Exercise ☀️

2020/21

• Computer Graphics: Exercise ☀️
• Computer Vision: Exercise ☀️

Exercise Coordination

Since starting my PhD, I'm now coordinating exercises for visual computing courses.

2024/25

• Real-Time Graphics: Exercise ❄️
• Object-Oriented Programming 1: Exercise ❄️
• Writing Scientific Papers: Seminar ❄️
• Computer Graphics: Exercise ☀️

2023/24

• Real-Time Graphics: Exercise ❄️
• Computer Graphics: Exercise ☀️

Open Student Projects!

We are interested in real-time rendering in general, with a specific focus on Radiance Fields.

Template adapted from Jon Barron.