Skip to main content
Back

OpenMVS v2.4.0 Rewrites Its Viewer and Automates 3D Scene Cropping

OpenMVS ships v2.4.0 with automatic region-of-interest detection, a rewritten viewer, VGGSFM support, and improved dense reconstruction from sparse SfM output.

  • OpenMVS v2.4.0 ships a full viewer rewrite on OpenGL 3 with ImGui and in-app workflows
  • Automatic Region-Of-Interest detection now focuses dense reconstruction compute on the target object
  • New VGGSFM importer and COLMAP SIMPLE_PINHOLE model widen SfM front-end compatibility
  • Depth-map confidences are normalized to [0,1] with a neighbor-agreement tuner for stability
  • Python utilities added for depth-to-TSDF meshing and NPY/EXR depth import via the repo
  • CUDA support extended to A100 and newer architectures; AGPL-3.0 license still applies

Photogrammetry pipelines have a well-known handoff problem. Structure-from-Motion tools recover camera poses and a sparse cloud of 3D points, but turning that skeleton into a watertight textured mesh is a separate, messier stage. OpenMVS is the open-source library that has quietly occupied that back half of the pipeline for years, and its new v2.4.0 release brings a viewer rewrite, automatic region-of-interest detection, and tighter integration with modern SfM tools like VGGSFM.

Where OpenMVS sits in the pipeline

OpenMVS is a C++ library aimed at the multi-view stereo community. It provides a complete set of algorithms to recover the full surface of a scene, taking camera poses plus a sparse point cloud as input and producing a textured mesh as output. The library pairs naturally with SfM front-ends like OpenMVG or COLMAP, which handle camera calibration and then hand the sparse reconstruction over for densification and meshing.

The pipeline breaks into four stages that map directly onto the CLI tools shipped in the repo:

  • Dense point-cloud reconstruction, where completeness and speed matter more than absolute accuracy because a later refinement pass cleans things up
  • Mesh reconstruction, which fits a surface to the dense cloud and needs to be robust to outliers