When you purchase through links on our articles, Future and its syndication partners may earn a commission.

We’ve written a lot recently about Gaussian splashes. This high-fidelity alternative to photometry is a relatively new technique for transforming 2D photos or videos into 3D scenes. Adobe’s new free app 3D Substance Viewer supports it, and we have seen impressive results for 3D visualizations in Unreal Engine 5 (see also our V-Ray 7 review).

Gaussian splashes can be used to capture precise lighting and reflections as well as geometry. The researchers just described that new technology could mean you won’t even need to capture the images or videos of the Gaussian splashes yourself. ActiveSplat is an autonomous tool that moves around an indoor space and produces a realistic and accurate 3D representation with incredible efficiency.

Developed by researchers at the Beijing Institute of Technology and the Institute of Artificial Intelligence at Tsinghua University, ActiveSplat is “a high-fidelity autonomous reconstruction system exploiting Gaussian splash.” THE research paper claims the system can “build a 3D map on the fly” using a Gaussian map and Voronoi graph for “efficient and comprehensive exploration with high-fidelity reconstruction results.”

The hybrid map representation uses both dense information about the environment and a sparse abstraction of space, using a sparse topology for efficient viewpoint sampling and path planning while also using dependent dense prediction of view to select viewpoints, “facilitating efficient decision-making with promising accuracy and completeness.” A hierarchical planning strategy based on the topology map helps mitigate repetitive trajectories and the system can adjust its tilt and tilt to handle unseen or high-loss areas.

What is 3D Gaussian Splash?

Gaussian splashing is a relatively new technique that has taken the 3D industry by storm over the past couple of years. This is a method of scanning real-world objects via video or photo and reconstructing them in 3D. Blending rasterization and optimization, it can model the real world with incredible detail and fidelity, accurately capturing geometry, lighting and reflections.

While triangle-based meshes are traditionally the way to represent 3D objects via photogrammetry, splats are volume-based. 3D data is defined using Gaussians which are each defined by a position, covariance, color and alpha channel, and Gaussian splashes can be created automatically from a few smartphone photos or videos using the machine learning in applications like Polycam and Luma AI (see also our selection of best 3D scanners and the best 3d modeling software).

Gaussian splashes allow for real-time photo realism, direct editing, and the ability to relight. Since the technology is so new, many potential applications have yet to be developed, including in the integration of 3D with virtual reality and augmented reality, and GUI-based production pipelines, such as Postshot in version beta, are just beginning to emerge.