Illustration of foveated rendering in of a Gaussian Splatting scene using VR-HTGS, showing inhomogeneous tile layouts that dynamically adjust to the user's gaze position.

Fast Eye-Tracked 3D Gaussian Rendering for Virtual Reality

Oral Presentation Gaussian SplattingVirtual realityFoveated renderingEye trackingImage-based renderingRasterization

Proc. Vision, Modeling and Visualization (VMV), 2026
Computer Graphics Lab,
TU Braunschweig Timon Scholz, Computer Graphics Lab,
TU Braunschweig Florian Hahlbohm, Computer Graphics Lab,
TU Braunschweig Martin Eisemann, Computer Graphics Lab,
TU Braunschweig
Cluster of Excellence PhoenixD,
Leibniz University Hannover Susana Castillo, Computer Graphics Lab,
TU Braunschweig
University of New Mexico,
USA Marcus Magnor

VR-HTGS enables artifact-free, eye-tracked foveated rendering of 3D Gaussian Splatting in VR at 120+ Hz by dynamically adapting the tile-based rasterizer to a heterogeneous tile layout following the gaze position, without sacrificing perceptual quality or model fidelity.

Project Page PaperVideo (Coming Soon)Code (Coming Soon) BibTeX
Reconstruction of two dynamic scenes from monocular video; dynamic scenes show camera frustums indicating recording trajectory of the input video

SPaGS: Fast and Accurate 3D Gaussian Splatting for Spherical Panoramas

Oral Presentation Image-based renderingGaussian SplattingReconstructionRasterizationPanoramic ImagingUAV

EGSR 2025 (CGF Journal Track), 2025
Computer Graphics Lab,
TU Braunschweig Junbo Li, Computer Graphics Lab,
TU Braunschweig Florian Hahlbohm, Computer Graphics Lab,
TU Braunschweig Timon Scholz, Computer Graphics Lab,
TU Braunschweig Martin Eisemann, Computer Graphics Lab,
TU Braunschweig Jan-Philipp Tauscher, Computer Graphics Lab,
TU Braunschweig
University of New Mexico,
USA Marcus Magnor

SPaGS enables high-quality, real-time free-viewpoint rendering from 360-degree panoramic images by introducing a new method for computing axis-aligned bounding boxes for spherical images that significantly accelerates omnidirectional ray casting of 3D Gaussians

Project Page PaperSupplementalVideoCodeDataset BibTeX
Reconstruction of two dynamic scenes from monocular video; dynamic scenes show camera frustums indicating recording trajectory of the input video

D-NPC: Dynamic Neural Point Clouds for Non-Rigid View Synthesis from Monocular Video

Oral Presentation Image-based renderingPoint-based modelsReconstructionRasterization

Eurographics (Computer Graphics Forum), 2025
Computer Graphics Lab,
TU Braunschweig Moritz Kappel, Computer Graphics Lab,
TU Braunschweig Florian Hahlbohm, Computer Graphics Lab,
TU Braunschweig Timon Scholz, Computer Graphics Lab,
TU Braunschweig Susana Castillo, Max Planck Institute for Informatics,
Saarland Informatics Campus Christian Theobalt, Computer Graphics Lab,
TU Braunschweig Martin Eisemann, Max Planck Institute for Informatics,
Saarland Informatics Campus Vladislav Golyanik, Computer Graphics Lab,
TU Braunschweig Marcus Magnor

D-NPC enables fast, non-rigid view synthesis from monocular video by expanding extending Implicit Neural Point Clouds (INPC) to the temporal domain.

Project Page VideoPaperCode BibTeX
© Website by Timon Scholz, 2026 · commit 8b9bc89