Announcement Text
.png)
Gaussian splatting is one of the most significant advances in 3D reconstruction in the last decade. It's also the technology behind Orbit Measure — Manifold's any-phone 3D scanning feature. This post explains what gaussian splatting is, why it matters for field documentation, and how it works in practice on real job sites.
Gaussian splatting (formally, 3D Gaussian Splatting, or 3DGS) is a real-time novel view synthesis technique developed by researchers at Inria and the Max Planck Institute for Informatics, first published in 2023. It represents a 3D scene as a collection of 3D Gaussian primitives — think of them as small, semi-transparent ellipsoids — each with its own position, shape, opacity, and colour. When rendered from any viewpoint, these overlapping Gaussians combine to produce a photorealistic reconstruction of the original scene.
The key difference from earlier techniques: gaussian splatting can reconstruct scenes faster and render them in real time. NeRF (Neural Radiance Fields), the previous state of the art, required significant GPU compute to render even a single frame. 3DGS renders at interactive frame rates, which makes it practical for viewing 3D reconstructions in a browser on ordinary hardware.
LiDAR-based scanning (like Apple's RoomPlan API, used in Manifold's Floor Plan Scan feature) works by emitting infrared pulses and measuring the time they take to return. It produces a point cloud — a dense map of geometric positions — that can be used to generate floor plans and simple 3D meshes. LiDAR is fast and accurate, but it requires specific hardware (iPhone 12 Pro or newer) and produces geometric data without photorealistic appearance.
Gaussian splatting works from ordinary video or photographs. Instead of measuring distances with hardware sensors, it uses computer vision to infer depth and structure from multiple frames of standard camera footage. The result is a photorealistic 3D model — not just a geometric mesh — reconstructed from the light and colour information in the images themselves. This is why Orbit Measure works on any phone: it uses the camera you already have, not a specialist sensor you need to buy.
Manifold's founder Erik Peterson presented the Orbit Measure technology and its underlying approach at the Reality Capture Conference, explaining how the system processes video footage into measurable 3D models using gaussian splatting reconstruction.
The workflow in practice: a contractor walks a space for 60 seconds, recording standard video with any iOS or Android phone. That footage is uploaded to Manifold's servers, which run the gaussian splatting reconstruction pipeline in the cloud. The output is a fully measurable 3D model that can be viewed in any browser and measured from any angle.
The advantage of server-side processing is that the computational demands of gaussian splatting reconstruction happen off the device. The phone's job is to capture good video footage. The cloud does the rest. This is why the any-phone requirement works — you don't need an iPhone Pro's depth sensor because the reconstruction is deriving depth from the visual information in the video itself.
Orbit Measure produces models accurate to approximately half an inch under normal conditions. This is more than sufficient for field documentation use cases: recording existing conditions, capturing dimensions for estimates, as-built documentation, and dispute protection.
There are genuine limitations to the approach. Gaussian splatting reconstruction performs best in well-lit spaces with clear visual texture. Very dark spaces, spaces with highly reflective surfaces (like polished floors or mirrors), and spaces with repetitive visual patterns (like identical white tiles) can produce noisier reconstructions. Scanning in those conditions still produces usable models in most cases, but the accuracy may degrade.
For very large outdoor spaces or environments where millimetre-scale accuracy is required for structural or civil engineering purposes, a terrestrial laser scanner or photogrammetry survey with ground control points is still the appropriate tool. Orbit Measure is built for the field documentation needs of trade contractors — not survey-grade geospatial work.
Photogrammetry — reconstructing 3D models from photographs — has been available for years. Tools like Polycam's photo mode, KIRI Engine, and RealityCapture use photogrammetric reconstruction. For field contractors, the practical problem with traditional photogrammetry is capture requirements: you typically need dozens or hundreds of overlapping photographs from carefully planned angles, often requiring a specific capture protocol.
Gaussian splatting reconstruction from video is more practical for field use: walk through a space naturally with your phone recording, and the reconstruction pipeline handles the rest. The resulting models are also more visually accurate — gaussian splatting preserves lighting and appearance information in a way that geometric photogrammetry meshes typically don't.
The practical implication is straightforward: you now have access to technology that, five years ago, required a $50,000 terrestrial laser scanner and a specialist operator. Orbit Measure puts gaussian splatting reconstruction into the hands of any contractor with a phone, at $24/user/month, with no specialist training required.
The output — a photorealistic, measurable 3D model of any interior space — is genuinely useful for field documentation. You can take measurements between any two points in the model from your desk. You can share it via link with clients, adjusters, or engineers who view it in a browser without installing anything. You can use it as a timestamped record of existing conditions before or after work.
For contractors who have tried photogrammetry tools and found them impractical for field use, or who have been told that 3D scanning requires LiDAR hardware, Orbit Measure's gaussian splatting approach changes the equation.
Gaussian splatting is moving fast. The original 2023 paper has already spawned dozens of variants — compressed representations, dynamic scene reconstruction, real-time editing of splat scenes. The field documentation applications are still early. What exists today is already useful enough to change how contractors document their jobs. What's coming in the next two to three years is likely to be significantly more capable.
Manifold is building Orbit Measure on this foundation — iterating on the reconstruction pipeline as the underlying research advances. The goal is to make spatial documentation from a phone fast enough, accurate enough, and cheap enough that it becomes a standard part of every contractor's workflow on every job.
Learn more about Orbit Measure or start a free trial — no credit card required. If you want to see a live demonstration of Orbit Measure on a real space, book a 15-minute demo with the team.
Compare Manifold
Real pricing, honest comparisons. No spin.