A team from the Vision and Imaging Technologies (VIT) department at Fraunhofer Heinrich-Hertz-Institut (HHI) has been honored with the CVMP Research Impact Award for its groundbreaking research in the field of compact 3D scene representations. The award ceremony took place during the 22nd ACM SIGGRAPH European Conference on Visual Media Production, one of Europe's leading conferences for visual media production.
The research team, consisting of Wieland Morgenstern, Florian Barthel, Anna Hilsmann, and Peter Eisert, developed the award-winning method “Compact 3D Scene Representation via Self-Organizing Gaussian Grids” in collaboration with Humboldt University in Berlin. The work was first presented at ECCV last year and has since received a great response from the international research community.
New efficiency for 3D Gaussian splatting scenes
This outstanding work presents a particularly compact representation for 3D Gaussian splatting (3DGS). By organizing scene parameters into smooth, two-dimensional grids, it is possible to compress data significantly while still enabling high-quality representations. Highly complex 3D scenes can therefore be stored efficiently using standard image compression methods – without changing the basic structure of 3DGS. Depending on the scene, the compression achieves reduction rates between 20× and 40× while maintaining high visual quality.
Even higher factors are possible for certain configurations. At the same time, the decompressed scenes remain compatible with established rendering workflows, allowing the method to be integrated directly into common 3DGS pipelines.
The award-winning technology thus offers a solution to key challenges facing modern visual media: large 3D data sets can be significantly reduced in size and transferred more quickly without losing quality. This opens new possibilities for XR applications, for example, where low-memory devices need to load high-quality 3D content quickly.
The award at CVMP underscores the importance of the work for current and future production workflows in visual media processing. Read the full paper here.
