Meshroom

Node-based photogrammetry suite that turns overlapping photos into textured 3D meshes on NVIDIA GPUs.

Open SourceSelf HostedOffline CapableGPU Required (4GB+ VRAM)
0.0 (0)

About

AliceVision's Meshroom started as a photogrammetry app and is now a general node-graph environment for image-processing pipelines, with 3D reconstruction still the headline use. Feed it a folder of overlapping photos and the bundled AliceVision plugin estimates camera poses, computes depth and outputs a textured mesh; other templates cover camera tracking, HDR merging, panorama stitching, photometric stereo, LiDAR E57 meshing and turntable object scanning. Each node caches its results, so changing one parameter recomputes only the steps downstream of it, and jobs can run locally or on a render farm. The 2025.1 release added an AI segmentation plugin driven by text prompts, and MeshroomHub hosts experimental add-ons for Gaussian splatting, monocular depth and dense matching that still need a developer setup. Prebuilt binaries cover Windows and Linux and target NVIDIA GPUs through CUDA 12 (compute capability 5.0 or later). The code is MPL-2.0, so commercial use is fine with file-level copyleft on modified sources. Suited to VFX artists, scanning hobbyists and researchers who want a free alternative to paid photogrammetry suites.

Should you use Meshroom?

Pick it when

Use Meshroom when you can photograph a real object or space from many angles and want a textured mesh through a GUI on a modest 4 GB GPU, with cached node steps for re-runs and the option to push jobs to a render farm.

Look elsewhere when

Skip it if you only have one image or a few unposed shots: photogrammetry needs broad coverage. VGGT or DUSt3R recover geometry from fewer views, and for real-time novel view rendering pick Gaussian Splatting.

Alternatives to Meshroom

  • Gaussian Splatting

    Photoreal real-time rendering from a multi-photo capture, but outputs splats rather than an editable mesh and is research-use only.

  • VGGT

    Feed-forward cameras, depth, and point maps from one to hundreds of views in seconds, but you get point maps, not a finished textured mesh.

  • Nerfstudio

    Radiance-field training with a web viewer under Apache-2.0, suited to view synthesis of a scene rather than mesh export.

  • DUSt3R

    Works from unposed image pairs with no calibration, but CC BY-NC-SA bars commercial use and you get point maps, not a textured mesh.

Reviews (0)

Leave a Review

No reviews yet. Be the first to review!

Details

Price
Free
Platform
Local/Desktop
Difficulty
Easy (2/5)
License
MPL-2.0
Minimum VRAM
4 GB
Added
Apr 3, 2026

Tags