How can new rendering technologies improve perception and immersion in mixed reality applications? 3D Gaussian Splatting (3DGS) is a promising and currently rapidly growing method in computer graphics. It enables highly realistic, real-time reconstructions of real scenes based on image or video recordings and thus opens up new perspectives for visual quality and interaction in XR applications. The VWDS department has already published numerous papers in the field of computer graphics, particularly focusing on the special requirements of mixed and mediated reality applications.

 

However, beyond traditional issues, the question of how new rendering approaches such as 3DGS compare to established methods and what potential they offer for immersive human-computer interaction is becoming increasingly important.

Our current study addresses this question and examines the use of 3D Gaussian splatting in comparison to conventional video pass-through. As part of an experimental use case, we analyzed distortion-reduced representations of static scenes. For this purpose, a realistic table scene with various objects was prepared, which was viewed by test subjects using a head-mounted display (Meta Quest 3). The scene was presented either using video pass-through or as a virtual reconstruction based on 3DGS.

The results show that 3D Gaussian splatting has great potential for use in mixed reality applications. Promising prospects are opening up, particularly at the level of display accuracy. With further improved rendering methods and optimized training times, 3DGS could represent a powerful alternative for the representation of real scenes in computer graphics in the future.

The study was presented at the IEEE International Symposium on Mixed and Augmented Reality(ISMAR). The corresponding publication can be found under the following link: h ttps:// doi.org/10.1109/ISMAR67309.2025.00131

In addition, two further poster papers were presented at the same conference. On the one hand, these deal with a preventive data set and a preliminary analysis of head movements in static 3DGS scenes. On the other hand, the internally developed ARC framework was extended to support 3DGS representations of real recorded scenes. The approach enables a quick pre-evaluation without an additional structure-from-motion process.

The poster work was also presented at ISMAR and is available at the following links:

3DGS head motion pre-analysis dataset: https://doi.org/10.1109/ISMAR-Adjunct68609.2025.00235

ARC – 3DGS-Erweiterung: https://doi.org/10.1109/ISMAR-Adjunct68609.2025.00229