borglab/gtsfm

End-to-end SFM framework based on GTSAM

What it solves

Georgia Tech Structure-from-Motion (GTSfM) is an end-to-end pipeline designed to reconstruct 3D scenes from 2D images. It addresses the computational intensity of traditional Structure-from-Motion (SfM) by providing a modular, parallelized architecture that leverages distributed computing to handle large-scale reconstruction tasks efficiently.

How it works

GTSfM uses a modular design built on top of the GTSAM library. It processes images through a front-end that detects and matches features (using methods like SIFT, SuperPoint, or SuperGlue) and a back-end that performs rotation averaging, translation averaging, and bundle adjustment to estimate camera poses and 3D geometry. The entire pipeline is parallelized using Dask, allowing it to scale across multiple CPU/GPU workers. It includes a browser-based studio interface for managing datasets, configuring models, and visualizing 3D results, including support for Gaussian Splatting.

Who it’s for

This project is for researchers, computer vision engineers, and developers working on 3D reconstruction, photogrammetry, or robotics who need a scalable, extensible, and high-performance SfM pipeline that can be deployed on local machines or distributed clusters.

Highlights

  • Distributed Architecture: Native support for parallel computation using Dask to accelerate large-scale reconstruction.
  • Modular Design: Highly swappable components for feature detection, description, matching, and bundle adjustment.
  • GTSFM Studio: A built-in browser-based workspace for managing runs, monitoring logs, and visualizing 3D reconstructions.
  • Versatile Output: Supports standard formats like COLMAP and provides integration scripts for NeRF-based pipelines like Nerfstudio.
  • Flexible Deployment: Supports local execution, remote GPU cluster submission, and Google Colab integration.

Related

  • Project
    cre185/InstantSfM

    A GPU-native Structure from Motion (SfM) pipeline that accelerates the process of reconstructing 3D structures from 2D images, including integrated support for 3D Gaussian Splatting.

  • Project
    mapillary/OpenSfM

    A Structure from Motion library in Python that reconstructs 3D scenes and camera poses from multiple images, integrating sensor data for geographical alignment.

  • Project
    cvg/vidmap

    VidMap is an offline Structure-from-Motion system that reconstructs 3D maps and camera trajectories from video by exploiting temporal data and global optimization.

  • Project
    colmap/colmap

    COLMAP is an open‑source Structure‑from‑Motion and Multi‑View Stereo pipeline that automatically turns collections of photos into 3‑D models. It offers a GUI, command‑line tools, Python bindings, GPU‑accelerated feature extraction, and cross‑platform binaries, making it a go‑to foundation for research and applications in computer vision, robotics, AR/VR, and any workflow that needs camera poses and dense scene geometry.

  • Project
    cvg/limap

    A toolbox for holistic 3D mapping and structure from motion that jointly optimizes points, lines, planes, and parametric primitives for more accurate 3D reconstructions.