Alignment Tool

COLMAP

COLMAP is the standard open-source Structure-from-Motion and Multi-View Stereo toolkit used to recover camera poses and sparse points before training many 3D Gaussian Splatting models.

Stage: AlignmentLocal reconstructionResearch workflowsCLI automation

What It Does

COLMAP estimates the position and orientation of every camera from overlapping images. In a Gaussian Splatting pipeline, this alignment stage is often more important than the trainer choice because poor poses create floaters, blur, and fractured geometry later.

It is useful for both casual captures and controlled production shoots. The GUI is friendly for first reconstructions, while the command line is better when you need repeatable processing for many datasets.

How To Use It In 3DGS

  • Capture an image set with strong overlap, stable exposure, and minimal motion blur.
  • Run feature extraction, feature matching, and sparse reconstruction in COLMAP.
  • Inspect camera coverage and remove failed or outlier images before training.
  • Export or keep the COLMAP sparse folder for trainers such as the INRIA implementation, gsplat, Nerfstudio, or OpenSplat.

Things To Watch

  • Shiny, transparent, moving, or textureless surfaces can reduce matching quality.
  • Wide-angle phone footage may need correct camera model settings or careful frame extraction.
  • A reconstruction can look plausible while still containing bad camera poses, so inspect the sparse cloud and camera distribution.

2026 Status Notes

  • COLMAP 4.0 added GLOMAP as the first-class global mapper through `global_mapper` and `automatic_reconstructor --mapper GLOBAL`.
  • LightGlue and ALIKED support make current COLMAP more attractive for difficult matching sets than older 3.x-era notes imply.
  • For 3DGS, keep both sparse reconstruction and original image folder together; many trainers assume this COLMAP directory convention.