Duplicate Structure Correction in SfM
This work is a post-processing step to structure from motion (SfM) in which our method analyzes a 3D reconstruction, identifies errors caused by duplicate structures within the scene, and then attempts to correct those errors.
Correcting for Duplicate Scene Structure in Sparse 3D Reconstruction
ECCV 2014 PDF
ECCV 2014 Poster
ECCV 2014 BibTeX
Datasets
Each dataset contains the images, match data, and 3D reconstruction used as input for our method.
Alexander Nevsky Cathedral (560 MB)
Arc de Triomphe (692 MB)
Berliner Dom (3.04 GB)
Big Ben (750 MB)
Brandenburg Gate (253 MB)
Cereal [1] (108 MB)
Church on Spilled Blood (403 MB)
Indoor [2] (234 MB)
Radcliffe Camera (416 MB)
Street [1] (44 MB)
References
[1] Roberts, R., Sinha, S.N., Szeliski, R., Steedly, D.: Structure from motion for scenes with large duplicate structures. CVPR (2011)
[2] Jiang, N., Tan, P., Cheong, L.F.: Seeing Double Without Confusion: Structure-from-Motion in Highly Ambiguous Scenes. CVPR (2012)
Recovering Correct Reconstructions from Indistinguishable Geometry
3DV 2014 PDF
3DV 2014 Poster
3DV 2014 BibTeX