mapillary/opensfm

OpenSfM is a Python library that takes a collection of ordinary photographs and reconstructs the 3D shape of a scene from them, along with figuring out exactly where each photo was taken and which direction the camera was pointing. This technique is called Structure from Motion, a method that works by finding matching visual features across multiple overlapping images and then computing the geometry that ties them all together. The README notes upfront that this repository is no longer under active development. The library handles the full processing pipeline, including feature detection, feature matching between image pairs, and the core math solvers that produce the final reconstruction. It can also incorporate data from GPS and accelerometers, which helps align the resulting 3D model to real-world geographic coordinates and makes the reconstruction more accurate when photos were taken across a large outdoor area. A built-in JavaScript viewer lets you inspect the 3D output directly in a browser window, which is useful for checking reconstruction quality or debugging a run that went wrong. The documentation covers building the library from source, running a reconstruction on your own images, and the full API reference. OpenSfM was created and maintained by Mapillary, a company that built one of the largest street-level photo platforms and was later acquired by Meta. The library served as the backbone for processing large-scale crowdsourced image collections into navigable 3D representations. Its focus on scalability and the ability to handle noisy real-world sensor data reflects that origin. The code is BSD-licensed, which permits broad reuse in other projects. Installation is supported via Conda and Docker, with build workflows targeting Ubuntu. Given that active development has stopped, users looking for a maintained alternative may want to check what the broader computer vision community has published more recently.