Work Tech art and tools
Auto-Photogrammetry
A Houdini tool that turns a 24-million-polygon scan into a game-ready asset.
- Project
- Auto-Photogrammetry (Houdini Digital Asset)
- Studio
- FutureGames (personal R&D)
- Years
- 2021
- Engine
- Houdini, Unreal Engine 4
- Role
- Tool author
- Team
- Solo
- Status
- Study
Problem
A photogrammetry scan captures a real object in enormous detail, and a game cannot use it as it comes: millions of polygons, floating noise, no usable UVs, and the light of the day baked into the colour. Making each scan game-ready by hand means cleanup, retopology, unwrapping, baking and LODs, every time.
Constraints
- The test scan: a tree trunk at 24 million polygons, from about 100 iPhone 8 photos processed in Agisoft Metashape.
- Houdini has to stay workable with data that heavy.
- Any scan, at any scale, without manual sculpting or painting.
What I built
A Houdini Digital Asset that takes a scan through three steps.
Clean and reduce. A small-pieces threshold deletes the floating scan noise, the mesh is realigned from its surface normals, and decimation brings the density down. The decimated result can be cached into the .hip file. The file grows, but Houdini stops recooking the early nodes, and everything downstream gets much faster.
Clip, scale and unwrap. A visual clipping box cuts away the ground and anything else around the object. Scans are scaled to one unit automatically, you set the target polycount for the low-poly mesh, and the UVs unwrap at the press of a button, with checkboxes to inspect the seams and the layout.
Bake, delight and LOD. Textures bake from the scan to the low-poly mesh at the resolution and path you choose. A delighting toggle removes the highlights and shadows the scan captured, a visual guide shows the ray-trace distance so the detail transfers cleanly, and a full LOD chain comes out ready for the engine.
Result
One tool, three steps, from a raw scan to a game-ready asset with its LODs, with no sculpting or painting by hand. Tested in Unreal Engine 4, five years before EQLabs.