A collection of tools and containers to enable simpler structure from motion reconstruction for end users.
The container can reconstruct 3D scene from a series of photos of a scene
by utilizing Structure from Motion (SfM).
Main script is at /sfm/sfm and it is a shell script that does the following:
- VisualSFM (ccwu.me/vsfm) is used for feature detection and matching as well
- SIFT - by using vlfeat (www.vlfeat.org)
- Yasutaka Furukawa's PMVS/CMVS (grail.cs.washington.edu/software/cmvs/)
- Screened Poisson Surface Reconstruction (http://www.cs.jhu.edu/~misha/Code/PoissonRecon)
- texrecon (https://github.com/nmoehrle/mvs-texturing)
- meshlab (www.meshlab.net)
- Nexus by Visual Computing Laboratory - ISTI - CNR (http://vcg.isti.cnr.it/nexus)
The container expects images in /input directory and will save output to
/output directory. It will produce sparse and dense reconstruction as well
as textured models in OBJ and PLY formats as well as NXS format suitable for streaming.
You can use binding mount or volumes, etc. for /input and /output. For example:
docker run --mount type=bind,source=/path/to/images,target=/input --mount type=bind,source=/path/to/output,target=/output sfm:latest /sfm/sfm
docker run -ti -v $(pwd)/input:/input -v $(pwd)/output:/output sfm:latest /sfm/sfm
This container was produced in EU H2020 project VI-SEEM (www.vi-seem.eu)