← All articles

3DGS vs. photogrammetry for cars: which 3D scan method should you use?

If you want a 3D model of a real car, you will quickly run into two terms: photogrammetry and Gaussian splatting (3DGS). Both start from photos or video. Both give you something you can rotate on screen. But they work in completely different ways, and for cars that difference matters a lot. This article compares 3DGS and photogrammetry for cars, so you can choose the right method for your project.

The short answer

  • Choose Gaussian splatting if you want a photorealistic visual copy of a car that you can view, share and use in real-time scenes, captured quickly from one video.
  • Choose photogrammetry if you need a mesh with clean surfaces that you can edit, measure or 3D print, and you are willing to spend more effort on capture and clean-up.

The rest of this article explains why.

How photogrammetry works

Photogrammetry has been around for decades. You take many overlapping photos, software finds matching points between them, calculates where each photo was taken, builds a dense point cloud and then turns that into a mesh: a surface made of triangles. Finally the photos are projected onto the mesh as a texture.

The output is a classic 3D model, usually OBJ, FBX or glTF. Every 3D program understands it. You can remesh it, measure it, edit it, and bring it into almost any tool.

How Gaussian splatting works

Gaussian splatting starts the same way: overlapping images and camera reconstruction. After that, it goes a different direction. Instead of building a surface, it fills the scene with millions of small, soft, coloured Gaussians and trains them until rendered views look like the original images. There is no single surface; there is a cloud of blobs that together *look* like the car from every angle.

The output is usually a .ply file with Gaussian parameters. You view it in a splat viewer such as SuperSplat, in a browser viewer, or in a game engine with a 3DGS plugin. For a full introduction, read what is Gaussian splatting?

Cars are a hard case for photogrammetry

Photogrammetry relies on finding the same physical point in different photos. That works well on stone, wood, fabric or matte objects. Cars are almost the opposite:

  • Glossy paint reflects the surroundings, so the "same" point looks different in every photo.
  • Glass is transparent and reflective at the same time.
  • Chrome and dark paint have little visible texture to match.
  • Large smooth panels offer few distinct features.

The result in photogrammetry is often a bumpy, melted-looking body with holes in the windows. Professionals work around this by spraying cars with matte scanning spray, using polarising filters or controlled studio light. That is impractical for most people.

Why 3DGS handles cars better visually

Gaussian splatting does not need to understand the exact surface to look right. It can represent reflections as view-dependent colour: the paint looks different from different angles, just like in reality. Thin details, soft edges and partly transparent areas are also easier to represent with blobs than with triangles.

That is why 3DGS car scans often look strikingly realistic even when captured with a phone in a normal environment. Shiny surfaces remain a challenge (we cover that in why shiny paint is hard for Gaussian splatting), but the failure modes are softer than the holes and blobs of photogrammetry.

Side-by-side comparison

Gaussian splatting (3DGS)Photogrammetry
OutputCloud of Gaussians (.ply)Textured mesh (OBJ, FBX, glTF)
Visual realismVery high, including reflectionsGood on matte surfaces, weaker on gloss
Glossy car paintHandled reasonably wellOften bumpy or broken
Editing geometryLimitedFull mesh editing
Measuring / CADNot suitablePossible with care
Capture effortOne steady videoMany photos, often controlled light
Real-time viewingFast with splat renderersFast with any 3D engine
Tool supportGrowing quicklyUniversal

When to choose photogrammetry

Photogrammetry still wins when you need:

  • Geometry to measure, inspect or reverse-engineer
  • A model to 3D print
  • Clean topology for animation or physics
  • Compatibility with tools that do not support splats yet

When to choose Gaussian splatting

3DGS is the better choice when you care most about how the car looks:

  • An interactive 3D view to share with others
  • A realistic car in a real-time scene, AR or VR prototype
  • Reference material for CGI and game artists
  • Research and experiments with neural rendering
  • A quick visual copy of your own car

For working with splats in an engine, see using car splats in Unreal Engine.

Try a car splat without building a pipeline

Building your own 3DGS pipeline means picking frames, running camera reconstruction, tuning training and cleaning the ground and background. CarTwin does that for you, with a pipeline built specifically for cars. You film two calm rounds around the car with your phone, upload the video, and after roughly 60 to 75 minutes you get a model to view in the browser, share via a link and download as .ply.

CarTwin is still fully in development and in an early pilot phase, so results are not always a perfect one-to-one copy of the car. If a scan fails because of an error in our software, you get your credit back.

Conclusion

Photogrammetry and Gaussian splatting are not rivals so much as tools for different jobs. For measurable geometry, photogrammetry. For realistic, shareable visuals of a car captured from a single video, 3DGS is hard to beat. The fastest way to judge is to see it: check the examples, read how to film a car for 3DGS and start your first scan. Bundles start at one credit; see pricing.