3D Capture & Reality Modeling

Surfacing vs Gaussian Splatting

Both take a real-world capture and turn it into something you can spin around on screen — so both get called a “3D model” or “3D scan,” and that is exactly why they get confused. But they answer different questions. Surfacing builds explicit geometry — a mesh you can measure, model, and engineer. Gaussian splatting builds a photoreal appearance — a scene you can fly through, but not measure. Picking the wrong one wastes the capture, so here is the difference and when to use each.

Surfacing
Produces a mesh — explicit triangles and vertices. Measurable, editable, CAD/BIM-ready geometry.
Gaussian splatting
Produces a radiance field — millions of soft colored 3D blobs. Photoreal appearance, rendered in real time.
Shared input
Both can start from the same capture (a point cloud or overlapping photos).
The trap
Both look “3D” on screen — but one is geometry and the other is look.
Rule of thumb
Need to measure or model → surfacing. Need to show or experience → splatting.
Surfacing → meshexplicit geometryGaussian splatting → splatsappearance fieldvertices + faces · measurable · editablesoft 3D blobs · photoreal · not measurable
The same object, represented two ways. Surfacing produces a mesh — discrete vertices joined by hard triangular faces, the geometry of where the surface is (left). Gaussian splatting produces a cloud of soft, overlapping, colored blobs that reproduce how the scene looks from each angle, with no hard surface at all (right). Diagram: NDEVR.

The Short Answer

Surfacing and Gaussian splatting are not competing versions of the same thing — they are different kinds of output. Surfacing gives you geometry: a mesh that says where every surface is, which you can measure, cut, model, and manufacture from. Gaussian splatting gives you appearance: a radiance field that reproduces exactly how a place looks — reflections, shine, and fuzzy detail included — but that has no clean, measurable surface underneath.

★

The one question that decides it

Are you going to measure or build from the result — volumes, sections, CAD, BIM, simulation, 3D printing? Use surfacing. Are you going to look at or move through it — a client walkthrough, a marketing fly-through, a heritage record? Use Gaussian splatting. If you answered “both,” you want both deliverables — see Use them together.

What Each One Is

Surfacing → a mesh

Surfacing connects the measured points of a point cloud into a continuous surface: a mesh of triangles between discrete vertices. The mesh is explicit geometry — every face has a known position, so the model knows where each surface actually is. That is what lets you take measurements, compute volumes, cut cross-sections, detect clashes, run simulations, push it into CAD or BIM, and 3D-print it. The trade-off: a bare mesh looks plain. Realistic color and material have to be added on top, and even then a mesh struggles with reflections, glass, and fine fuzzy detail.

A subject, its textured 3D reconstruction, and the underlying shaded triangle mesh
Surfacing’s deliverable is the geometry on the right: a shaded triangle mesh — explicit, measurable surfaces — here shown beside the same subject and its textured reconstruction. Image: Cicero Moraes, CC BY-SA 4.0, via Wikimedia Commons.

Gaussian splatting → a radiance field

Gaussian splatting does not try to find surfaces at all. It represents the scene as millions of soft, colored, semi-transparent 3D blobs (“Gaussians”) that, blended together, reproduce what a camera would see from any viewpoint — a radiance field. The result is strikingly photoreal and renders in real time, capturing reflections, transparency, foliage, and view-dependent shine that meshes handle poorly. The trade-off: there is no crisp surface to measure. The blobs encode look, not location, so a splat is the wrong tool the moment you need a number off it.

A photorealistic Gaussian splatting reconstruction of a collapsed building from drone footage
Gaussian splatting’s deliverable is the look: a photoreal, walk-through scene — here a collapsed building reconstructed from drone footage — rendered from viewpoints the camera never actually shot. Image: Jurdein, CC BY-SA 4.0, via Wikimedia Commons.

The Core Difference: Geometry vs Appearance

Strip away the details and one distinction explains all the others. Surfacing answers “where is the surface?” Gaussian splatting answers “what would I see from here?”

Because a mesh stores actual geometry, anything that depends on geometry just works: a distance is the distance between two vertices, a volume is the volume enclosed by the faces, a section is a clean cut through them. Because a splat stores appearance, none of that has a well-defined answer — there is no single surface to measure, only a soft field of blobs tuned to look right. Conversely, because a splat captures how light actually behaves at each point and angle, it nails reflections and shine and fuzz; a mesh has to fake all of that with textures and materials, and still cannot truly reproduce a mirror or a pane of glass.

This is also why one is not a drop-in replacement for the other. A Gaussian splat will not give you a measurable CAD model no matter how real it looks, and a mesh will not give you a photoreal mirror-and-foliage walkthrough no matter how much you texture it. They are different deliverables built for different jobs.

Side by Side

Surfacing (mesh) versus Gaussian splatting, on the differences that drive the choice.
 Surfacing (mesh)Gaussian splatting
CapturesSurface geometry — where things areAppearance — how the scene looks from each angle
Built fromA point cloud (LiDAR or photos)Overlapping photos or video
Made ofTriangles & vertices (explicit geometry)Millions of soft 3D Gaussian blobs
Measure / volumes / sectionsYes — that is the pointNo — no defined surface to measure
CAD / BIM / simulation / 3D printYesNo
PhotorealismNeeds textures; weak on reflections, glass, fuzzExcellent — reflections, glass, foliage, shine
Real-time viewingYes — lightweight, runs anywhereYes — photoreal, needs a GPU + splat viewer
Editing / interopMature: edit, retopo, boolean, export everywhereEmerging tools; viewer-specific; large files
Best forMeasurement, engineering, modeling, fabricationVisualization, presentation, immersive walkthroughs

When to Use Which

Reach for surfacing

  • Measurements, volumes, sections
  • CAD / BIM / as-built models
  • Clash detection & simulation
  • 3D printing or fabrication
  • A clean, lightweight asset for any tool

Reach for splatting

  • Client & marketing fly-throughs
  • Immersive VR / AR walkthroughs
  • Heritage & “as-it-looked” records
  • Scenes with glass, water, foliage, shine
  • The most lifelike possible record

Use both

  • Survey & AEC deliverables
  • Digital twins
  • Heritage that must be measured
  • Inspection + presentation together
  • One site visit, two outputs

Put concretely: to compute how much fill a collapsed structure needs, surface the cloud and measure the mesh — a splat cannot give you the volume. To walk a client through that same structure exactly as it looked on site, build a Gaussian splat — a bare mesh will look lifeless next to it. The mistake that wastes a capture is using a beautiful splat where a measurable mesh was needed, or vice versa.

Use Them Together

For many real projects the answer is not either/or. The same drone flight or scan session can feed both pipelines: surface the data into a measurable mesh for the engineering, CAD, and volume work, and build a Gaussian splat from the imagery for the presentation and walkthrough. The mesh carries the numbers; the splat carries the experience. Increasingly, surveying, AEC, and heritage teams deliver exactly this pairing — accurate geometry to build from, and a photoreal scene to show.

To go deeper on either side, see Point Cloud Surfacing for how meshes are built and which surfacing method to use, and What is Gaussian Splatting? for how splats are trained and rendered.