ground truth rendering
error distribution
percentage of pixels with an error of n*sigma
errors larger than 10*sigma are collected
in the last bin (11*sigma)
cumulative error distribution
percentage of pixels with
an error < n*sigma
FUR ST6 ST4 ZAH VU TYL TYL09 ECCV234 FUR ST6 ST4 ZAH VU TYL TYL09 ECCV234

FUR Y. Furukawa and J.Ponce Accurate, Dense, and Robust Multi-View Stereopsis CVPR 2007 [website]

ST4 C. Strecha, R. Fransens, L. Van Gool Wide-baseline Stereo from Multiple Views: a Probabilistic Account CVPR 2004 vol.2 pages 552-559 [pdf] [demo]

ST6 C. Strecha, R. Fransens, L. Van Gool Combined Depth and Outlier Estimation in Multi-View Stereo CVPR 2006 [pdf][demo]

ZAH Zaharescu, A. and Boyer, E. and Horaud, R.P. TransforMesh: a topology-adaptive mesh-based approach to surface evolution ACCV 2007 [website]

VU: Vu, H. and Keriven, R. and Labatut, P. and Pons, J.-P Towards high-resolution large-scale multi-view stereo CVPR 2009

TYL:R. Tylecek and R.Sara Depth Map Fusion with Camera Position Refinement CVWW 2009 [pdf]
TYL09:R. Tylecek and R.Sara Refinement of Surface Mesh for Accurate Multi-View Reconstruction, Modeling3D ACCV 2009 [pdf]

ECCV234:submission


  • Acknowledgements Many thanks to all who provided their results on these datasets.
  • The original results are published in: C. Strecha, W. von Hansen, L. Van Gool, P. Fua, U. Thoennessen On Benchmarking Camera Calibration and Multi-View Stereo for High Resolution Imagery CVPR 2008 [pdf]
  • Note that the results in the CVPR paper are obtained by evaluating only the 6th image (0005.ppm). Furthermore, the error histograms had been are plotted against 3sigma.
  • When using the datasets in any publication, please refer to them with the exact naming convention (e.g. fountain-P11, Herz-Jesu-K7, castle-R20...) and cite the above paper.