Spline’s Revolutionary Update: Integrating Gaussian Splatting for Enhanced 3D Web Experience

Gaussian Splats Have Arrived!

Spline is revolutionizing 3D design by incorporating Gaussian Splatting. This innovative method breathes life into real-life data, rendering them in 3D space in real-time. It’s akin to Radiance Field methods (NeRFs) but with notable advantages: it’s quicker to set up, renders faster, and achieves equal or superior quality.

Seamlessly Import .ply Format Splats

Begin by creating Gaussian splats using apps like Polycam or Luma. These can be crafted from recorded videos on your phone or camera. Once created, simply drag and drop the .ply file into Spline’s editor. The process is streamlined, with a maximum file size of 480MB, making it accessible and efficient.

Customize and Crop with Ease

Spline allows you to crop and adjust splats for both performance and aesthetic purposes. You can select splats, create crop areas, and adjust properties like type, center, size, and rotation. This not only enhances the visual appeal but also optimizes performance by focusing on the essential elements of your splats.

Blend Splats with 3D Objects

Mixing splats with 3D objects offers a realm of creativity but requires attention to compatibility. Currently, some features like transitions and materials’ transparency are either partially supported or not compatible with splats. However, the potential for creating unique and immersive 3D experiences is immense.

Export and Embed with Flexibility

The splats you create and refine in Spline can be exported and embedded into websites, offering a new dimension to web-based 3D experiences. While still an experimental feature, especially for mobile support, it’s a huge leap forward in rendering hyperrealistic 3D representations on the web.

Understanding Gaussian Splatting

Gaussian Splatting, originating in the scientific field in the 90s, has recently seen a resurgence, particularly in real-time scene visualization. It represents a 3D scene as millions of particles (Gaussians) with distinct properties like position, orientation, scale, opacity, and color. These particles are converted to 2D space for efficient rendering, leveraging machine learning to train the Gaussian parameters.

This technique is distinct from traditional computer graphics, which rely on meshes of polygons, and from Neural Radiance Fields (NeRF), which use AI to render 3D scenes from multiple viewpoints but are limited by their computational demands and inability to render scenes into traditional graphics meshes. Gaussian Splatting offers a more efficient alternative, requiring less computing power for real-time rendering.

One notable aspect of 3D Gaussian Splatting is its use of “anisotropic” Gaussians, which are non-spherical and directionally stretched. This characteristic makes 3D Gaussians differentiable, allowing them to be trained using deep learning techniques. The entire rendering pipeline is made differentiable, which is essential for the system’s operation.