GLOSSARY
Photogrammetry
Photogrammetry reconstructs a 3D model from many overlapping photographs by finding matching features across images and triangulating their 3D positions. The classical, non-AI way to digitize objects.
Definition
A typical photogrammetry pipeline: take 50–500 photos of an object from every angle, run feature matching (SIFT, SuperPoint) across them, solve for camera positions and a sparse 3D point cloud (Structure-from-Motion), then densify the point cloud and fit a mesh (Multi-View Stereo plus Poisson surface reconstruction).
Tools: RealityCapture (paid, fast), Metashape (paid, gold standard), Meshroom (free, open-source AliceVision), Polycam (mobile-first, mixes photogrammetry and Gaussian splatting).
Why it matters
Photogrammetry is the right tool when geometric accuracy matters and the object exists. Heritage scanning, surveying, digital archives, e-commerce product capture: photogrammetry gives you a metric reconstruction tied to real-world dimensions.
The cost is photo count, processing time, and surface requirements. Glossy, transparent, or featureless surfaces (clear glass, shiny metal, plain white walls) confuse feature matching and reconstruct poorly. AI-based image-to-3D tolerates much sparser input but produces non-metric output.
Common confusion
Photogrammetry and image-to-3D both turn images into 3D, but they are different things. Photogrammetry triangulates only what was photographed; the back side comes out wrong if you did not photograph it. Image-to-3D hallucinates the missing views with prior knowledge from a generative model.