Scaniverse has released more information about their SPZ file format and has open-sourced it, making it accessible to developers worldwide.
Gaussian Splatting, a Radiance Field method for creating highly realistic 3D scans, is reshaping industries that leverage 2D imaging. However, the large file sizes of these scans, often beginning at several hundred megabytes to multiple gigabytes, make storage and sharing challenging, especially on mobile devices. To address this, Scaniverse developed the SPZ format to significantly compress these files, making uploads, downloads, and storage more efficient.
Since Gaussian Splatting launched in Scaniverse earlier this year, SPZ has been in use internally. In August, SPZ export support was added to Scaniverse, and import functionality was introduced in Niantic Studio—a free, web-based tool for 3D and AR content creation. Now, Niantic is open-sourcing SPZ under an MIT license, inviting developers to experiment and contribute via GitHub. In fact, this topic was discussed in tomorrow's episode of the View Dependent Podcast, where the main focus was coincidentally Gaussian Splatting compression.
The SPZ format uses a column-based structure, grouping similar values together (e.g., positions with positions, rotations with rotations), which are then gzip-compressed. This organization means that the order in which Gaussians are stored affects file size, particularly when similar values are grouped. Future algorithms, such as Self-Organizing Gaussians, which is used in Nerfstudio, could leverage this feature to further optimize storage.
SPZ focuses on effective representation and compression of data without specifying parameters like initialization or densification, which impact file size independently of the format. It prioritizes compatibility and simplicity over achieving maximum compression, making it more accessible for broader adoption—essentially serving as a "ply v2" with sensible defaults.
One limitation is that SPZ currently only has a C++ API, making it less accessible for developers who prefer Python or JavaScript. However, expanding language support appears to be on the roadmap. Additionally, while SPZ has not yet been evaluated against state-of-the-art (SOTA) compression methods, its open-source nature invites community exploration and benchmarking. To learn more about the current state of compression, I recommend checking out 3DGS.ZIP.
The SPZ format includes a versioning system, allowing for future updates. It not only stores information about spherical harmonics and other splat-specific data but also uses advanced compression techniques like fixed-point quantization and logarithmic encoding to shrink data sizes while preserving visual fidelity. This results in the massive reduction in file size—around 90% compared to unmodified PLY files.
Open-sourcing the format under the MIT license encourages community involvement, and Niantic has established a Developer section on their Community site for sharing creations, asking questions, and providing feedback.
The goal with SPZ is clear: make working with splats easy, fast, and accessible. With the Scaniverse app, users can process splats on their phone in about the same time it takes to develop an instant Polaroid picture—hence the term, "Polaroid for 3D." From the app, users can export splats in SPZ format for use elsewhere and even upload them to a global map of public places.
With Radiance Field methods such as Gaussian Splatting poised to transform the XR landscape, SPZ is set to play a key role in accelerating this shift. By making splats more efficient to store, share, and use, Niantic's Scaniverse hopes that SPZ will help shape the future of spatial computing. Scaniverse can be downloaded on both iOS and Android devices.
New to Gaussian Splatting? Start here
