Experimenting with Gaussian Splats

Beth Mutch
Video Producer + Project Manager

At its core, 3D Gaussian splatting is about translating volumetric data into a visual format that can be integrated seamlessly into film scenes. It does so by employing a Gaussian function to create a smooth transition of data points in a three-dimensional space, much like a painter blending hues seamlessly on a canvas.

We decided to explore the fascinating world of 3D Gaussian splatting through two different lenses: one using 25 frames and the other with 250 frames. We will unpack the insights from our video and discuss the impact of frame count on the quality and utility of 3D Gaussian splats in visualization and animation.

Our experiment was straightforward: create 3D Gaussian splats using two distinct frame counts to observe the differences in the final visualization. Why the focus on frame count? Because in the art of 3D visualization, the number of frames can drastically alter the smoothness, details, and realism of the sequence.

Starting with 25 frames, we aimed for speed and efficiency. Fewer frames mean less data to process and a faster rendering time, which can be crucial when working under tight deadlines or when quick prototyping is needed.

The result? A visualization that gives a broad idea of the data distribution and a rough concept of the subject, but clearly does not work properly.

Upping the ante to 250 frames, we expected to see a more refined and smooth visualization. With ten times the data, the splatting effect benefitted from the increased information, allowing for a more continuous and detailed representation of the volumetric data. This level of detail is particularly important for high-quality productions where precision and realism are paramount.

The video illustrates a side-by-side comparison, showcasing how the increased frame count leads to a more fluid and life-like animation. The 250-frame version demonstrates how subtle changes and transitions in the data become visible, capturing nuances that the 25-frame version simply cannot provide.

While the 250-frame splats offer a superior visual experience, it’s not always the practical choice. There definitely are trade-offs between the two, considering factors like computational resources, project timelines, and the intended use of the splats. For instance, in real-time applications or when working with massive datasets, fewer frames might be the pragmatic approach, but more than 25 is needed.

Related reads