This table lists the benchmark results for the low-res many-view scenario.
The following metrics are evaluated:
Accuracy [%]: The fraction of the reconstruction which is closer to the ground truth than the evaluation threshold distance (*). Larger is better.Completeness [%]: The fraction of the ground truth which is closer to the reconstruction than the evaluation threshold distance (*). Larger is better.F1 score [%]: The harmonic mean of accuracy and completeness, used to rank methods based on both metrics. Larger is better.Time [s]: The runtime of the method. Smaller is better.
(*) For exact definitions, detailing how potentially incomplete ground truth is taken into account, see our paper .
The datasets are grouped into different categories, and result averages are computed for a category and method if results of the method are available for all datasets within the category.
Note that the category "all" includes both the high-res multi-view and the low-res many-view scenarios.
Methods with suffix _ROB may participate in the Robust Vision Challenge .
training datasets , ground truth and accuracy / completeness visualizations are also available. The visualizations may not work with mobile browsers.
Set : Test
Metric : F1 score [%]
Tolerance : 2cm
Method
Info all low-res indoor outdoor lakeside sand box storage room storage room 2 tunnel
DeepPCF-MVS 80.84 1 63.41 1 53.95 1 69.71 1 69.72 3 71.26 1 48.73 1 59.18 1 68.16 1 DeepC-MVS_fast 79.62 2 62.35 2 51.75 4 69.41 2 69.38 4 70.78 2 46.85 2 56.66 4 68.08 2 DeepC-MVS 79.50 3 61.99 3 52.12 3 68.57 3 67.79 5 70.58 3 45.96 3 58.27 3 67.34 3 3Dnovator+ 77.84 4 59.53 4 47.18 7 67.76 4 70.63 1 67.68 4 43.34 7 51.01 9 64.98 5 3Dnovator 76.31 5 59.36 5 48.19 5 66.81 5 70.04 2 67.59 5 44.27 5 52.10 7 62.80 7 TAPA-MVS 73.13 9 58.67 6 52.34 2 62.89 9 63.68 8 63.46 9 45.85 4 58.83 2 61.53 8 Andrea Romanoni, Matteo Matteucci: TAPA-MVS: Textureless-Aware PAtchMatch Multi-View Stereo . ICCV 2019 PCF-MVS 73.52 7 57.06 7 48.10 6 63.03 8 66.85 6 62.36 10 43.32 8 52.89 5 59.86 10 Andreas Kuhn, Shan Lin, Oliver Erdler: Plane Completion and Filtering for Multi-View Stereo Reconstruction . GCPR 2019 ACMP 74.13 6 56.41 8 44.60 11 64.28 6 61.46 10 65.81 7 37.07 13 52.13 6 65.56 4 Qingshan Xu and Wenbing Tao: Planar Prior Assisted PatchMatch Multi-View Stereo . AAAI 2020 OpenMVS 72.83 10 56.18 9 45.66 9 63.19 7 64.67 7 65.13 8 39.60 11 51.71 8 59.76 11 ACMM 73.20 8 55.01 10 43.19 13 62.89 9 59.60 12 66.07 6 35.89 15 50.48 10 63.01 6 Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo . CVPR 2019 LTVRE_ROB 69.57 12 53.52 11 45.46 10 58.89 11 58.76 13 60.60 13 43.91 6 47.01 13 57.32 14 Andreas Kuhn, Heiko Hirschmüller, Daniel Scharstein, Helmut Mayer: A TV Prior for High-Quality Scalable Multi-View Stereo Reconstruction . International Journal of Computer Vision 2016 PLC 70.83 11 53.50 12 45.79 8 58.64 13 56.31 15 60.76 12 41.70 9 49.88 11 58.86 13 Jie Liao, Yanping Fu, Qingan Yan, Chunxia xiao: Pyramid Multi-View Stereo with Local Consistency . Pacific Graphics 2019 COLMAP_ROB 66.92 15 52.32 13 42.45 14 58.89 11 56.18 16 61.09 11 38.61 12 46.28 14 59.41 12 Johannes L. Schönberger, Enliang Zheng, Marc Pollefeys, Jan-Michael Frahm: Pixelwise View Selection for Unstructured Multi-View Stereo . ECCV 2016 ACMH+ 68.96 13 52.02 14 43.83 12 57.49 14 59.85 11 51.27 20 40.04 10 47.62 12 61.35 9 OpenMVS_ROB 64.09 16 48.56 15 38.68 16 55.15 15 61.90 9 49.65 22 36.02 14 41.33 18 53.89 16 ACMH 67.68 14 47.97 16 38.24 18 54.45 16 56.88 14 50.08 21 34.49 17 41.99 17 56.40 15 Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo . CVPR 2019 CIDER 47.21 17 38.36 17 53.11 17 55.37 17 55.10 18 33.25 19 43.46 16 48.87 19 Qingshan Xu and Wenbing Tao: Learning Inverse Depth Regression for Multi-View Stereo with Correlation Cost Volume . AAAI 2020 AttMVS 45.85 18 39.26 15 50.25 20 49.36 20 51.75 19 34.83 16 43.70 15 49.63 17 P-MVSNet 44.46 19 37.09 19 49.37 22 49.27 21 49.30 23 34.35 18 39.83 19 49.54 18 A-TVSNet + Gipuma 42.12 20 29.09 21 50.80 18 48.28 22 56.53 15 29.14 20 29.04 22 47.59 20 ANet 39.53 21 24.60 23 49.48 21 49.42 19 55.75 16 22.08 22 27.13 24 43.25 22 ANet-0.75 39.47 22 22.61 25 50.71 19 49.43 18 55.75 16 18.21 25 27.01 25 46.96 21 MVSNet 38.33 23 26.99 22 45.89 23 40.49 24 57.57 14 20.52 23 33.47 21 39.63 23 R-MVSNet 36.87 24 29.78 20 41.61 24 42.00 23 46.99 24 24.73 21 34.83 20 35.83 24 MVSCRF 28.32 25 18.18 26 35.09 25 32.16 27 44.37 26 14.66 27 21.69 28 28.73 26 Pnet_fast 23.85 26 17.41 27 28.15 26 32.72 25 24.90 31 12.22 34 22.60 26 26.83 27 Snet 23.85 26 17.41 27 28.15 26 32.72 25 24.90 31 12.22 34 22.60 26 26.83 27 A1Net 23.74 28 23.95 24 23.61 29 21.43 32 19.62 34 19.93 24 27.96 23 29.78 25 PMVS 37.38 18 21.09 29 11.49 37 27.48 28 24.09 28 44.44 25 15.98 26 7.01 38 13.92 33 Y. Furukawa, J. Ponce: Accurate, dense, and robust multiview stereopsis . PAMI (2010) F/T MVSNet+Gipuma 17.31 30 14.39 31 19.26 31 16.83 34 19.67 33 14.03 29 14.75 31 21.27 29 MVSNet + Gipuma 16.91 31 14.12 32 18.77 34 16.39 36 19.13 36 13.98 30 14.27 32 20.78 30 hgnet 16.45 32 12.31 34 19.21 32 21.51 30 30.31 29 12.80 31 11.82 35 5.81 35 DPSNet 16.45 32 12.31 34 19.21 32 21.51 30 30.31 29 12.80 31 11.82 35 5.81 35 MVE 26.22 19 16.26 34 16.97 29 15.79 36 11.75 38 19.40 35 14.45 28 19.48 29 16.21 32 Simon Fuhrmann, Fabian Langguth, Michael Goesele: MVE - A Multi-View Reconstruction Environment . EUROGRAPHICS Workshops on Graphics and Cultural Heritage (2014) firsttry 15.88 35 12.57 33 18.08 35 17.80 33 17.14 37 11.54 37 13.61 33 19.31 31 example 15.69 36 10.14 38 19.39 30 23.73 29 30.32 28 11.10 38 9.17 37 4.13 38 unMVSv1 13.05 37 12.03 36 13.73 37 14.18 37 16.18 38 12.13 36 11.93 34 10.84 34 RMVSNet 11.09 38 15.49 30 8.17 39 9.16 39 10.05 39 12.57 33 18.40 30 5.29 37 CMPMVS 51.72 17 7.38 39 0.03 39 12.27 38 2.37 40 34.46 27 0.06 39 0.00 39 0.00 39 M. Jancosek, T. Pajdla: Multi-View Reconstruction Preserving Weakly-Supported Surfaces . CVPR 2011 dnet 0.00 40 0.00 40 0.00 40 0.00 41 0.00 40 0.00 40 0.00 39 0.00 39 UnsupFinetunedMVSNet 16.83 34
