VisualSFM : A Visual Structure from Motion System

Changchang Wu


VisualSFM is a GUI application for 3D reconstruction using structure from motion (SFM). The reconstruction system integrates several of my previous projects: SIFT on GPU(SiftGPU), Multicore Bundle Adjustment, and Towards Linear-time Incremental Structure from Motion. In particular, the GUI SfM system and SiftGPU both started as course projects at UNC. VisualSFM runs very fast by exploiting multicore parallelism in feature detection, feature matching, and bundle adjustment.

This program provides interfaces to run Yasutaka Furukawa's PMVS/CMVS tool and to prepare data for Michal Jancosek's CMP-MVS. In additional to SiftGPU, VisualSFM also supports external SIFT binaries including the original version by Lowe and the VLFeat version.

Research binary of this program is available upon request, which includes the following projects: Viewpoint Invariant Patch, Single View Repetition Detection, and Schematic Surface Reconstruction.  

Structure from Motion - A Visual Approach

Reconstruct 3D with a few button clicks, and watch the dynamic reconstruction process!
You still have the option to run from command line without a GUI!


Download V0.5.24 (changelog with new feature documentation; latest news at Google group)

Windows* (x64, win32,   install instructions, * for nVidia CUDA or CUDA Simulation)
Windows  (x64, win32,   install instructions)
Linux    (x86_64, i386, install instructions)
Mac OSX  (x86_64, i386, install instructions, check out the package contributed by Iván)

VisualSFM is free for personal, non-profit or academic use. See README for more details. 
* Please cite VisualSFM according to README in your publication.

* Post questions and see discussions at the Google Group, or email me with VisualSFM in the title.

Basic usage, image size, customized matching, controls, parameters, nvm file, demo, ...
The changelist page offers limited documentations for recently added features.

Introductory videos(1, 2) and a tutorial by Eugene Liscio. Tutorial in French by Mathis Fantin.

Live Reconstruction Visualization! (more videos)

Below is a 3 minute live run of reconstruction of 130 images(using pre-computed matches).