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 : 50cm
Method
Info all low-res indoor outdoor lakeside sand box storage room storage room 2 tunnel
DeepPCF-MVS 98.18 3 96.68 1 95.10 1 97.74 1 98.30 1 97.78 4 92.41 2 97.79 1 97.12 1 DeepC-MVS_fast 98.59 1 96.44 2 94.71 3 97.59 2 98.20 2 97.75 5 92.08 3 97.35 5 96.83 2 DeepC-MVS 98.44 2 96.08 3 94.14 10 97.37 3 97.83 5 97.46 10 90.75 13 97.52 3 96.81 3 CMAPSRf 96.07 4 94.39 7 97.19 5 96.86 8 97.92 2 91.54 10 97.24 7 96.79 4 TAPA-SR 95.82 5 94.80 2 96.49 9 96.19 11 97.58 9 92.04 4 97.56 2 95.71 12 ACMM 97.58 5 95.68 6 94.49 6 96.48 10 96.12 12 97.20 13 92.04 4 96.94 10 96.10 9 Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo . CVPR 2019 PCF-MVS 97.08 13 95.68 6 93.36 14 97.23 4 98.16 3 96.93 14 91.43 11 95.29 18 96.61 5 Andreas Kuhn, Shan Lin, Oliver Erdler: Plane Completion and Filtering for Multi-View Stereo Reconstruction . GCPR 2019 ACMH+ 97.24 9 95.61 8 93.77 13 96.83 6 96.69 9 97.71 6 90.70 14 96.85 11 96.09 10 TAPA-MVS 97.07 14 95.48 9 94.55 4 96.10 12 95.68 15 97.30 11 91.73 8 97.37 4 95.32 14 Andrea Romanoni, Matteo Matteucci: TAPA-MVS: Textureless-Aware PAtchMatch Multi-View Stereo . ICCV 2019 CMAPbl 95.42 10 94.03 12 96.35 11 95.96 13 96.92 15 91.10 12 96.97 8 96.16 8 PLC 97.94 4 95.35 11 94.28 9 96.06 13 95.26 18 96.59 16 91.61 9 96.96 9 96.34 7 Jie Liao, Yanping Fu, Qingan Yan, Chunxia xiao: Pyramid Multi-View Stereo with Local Consistency . Pacific Graphics 2019 ACMH 97.28 7 95.29 12 93.25 15 96.65 7 96.32 10 97.93 1 89.82 16 96.68 13 95.69 13 Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo . CVPR 2019 ACMP 97.20 10 95.11 13 93.01 17 96.52 8 95.68 15 97.26 12 89.74 17 96.27 15 96.61 5 Qingshan Xu and Wenbing Tao: Planar Prior Assisted PatchMatch Multi-View Stereo . AAAI 2020 LTVRE_ROB 97.16 11 95.10 14 94.13 11 95.75 14 95.69 14 95.73 20 92.87 1 95.39 17 95.83 11 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 GSE 95.00 15 94.33 8 95.45 16 94.81 19 96.53 18 91.98 6 96.68 13 95.02 16 A-TVSNet + Gipuma 94.96 16 94.54 5 95.24 18 95.31 17 96.58 17 91.82 7 97.27 6 93.83 17 COLMAP_ROB 97.56 6 94.43 17 93.09 16 95.33 17 94.55 20 96.37 19 89.45 18 96.72 12 95.07 15 Johannes L. Schönberger, Enliang Zheng, Marc Pollefeys, Jan-Michael Frahm: Pixelwise View Selection for Unstructured Multi-View Stereo . ECCV 2016 3Dnovator 97.25 8 92.47 18 87.97 24 95.47 15 97.93 4 97.85 3 86.02 22 89.92 24 90.64 20 3Dnovator+ 97.12 12 92.17 19 87.57 25 95.24 18 97.49 6 97.68 7 85.24 23 89.91 25 90.54 21 LPCS 92.07 20 92.93 18 91.49 23 90.32 24 93.29 25 90.66 15 95.20 19 90.87 19 OpenMVS 96.50 15 91.77 21 86.65 27 95.19 20 97.38 7 97.66 8 84.60 24 88.69 28 90.53 22 R-MVSNet 90.93 22 89.80 20 91.68 22 90.81 23 94.15 24 87.67 19 91.93 23 90.07 23 CPR_FA 89.35 23 88.37 23 90.00 26 89.38 26 87.69 33 87.29 21 89.46 27 92.93 18 CIDER 88.55 24 83.13 31 92.16 21 93.95 21 95.61 21 83.79 27 82.47 31 86.92 25 Qingshan Xu and Wenbing Tao: Learning Inverse Depth Regression for Multi-View Stereo with Correlation Cost Volume . AAAI 2020 ANet-0.75 87.89 25 89.58 21 86.76 27 84.37 36 90.29 29 84.51 25 94.65 20 85.61 28 OpenMVS_ROB 92.34 16 87.46 26 81.75 32 91.27 24 92.55 22 95.21 22 81.21 31 82.29 32 86.05 26 P-MVSNet 85.35 27 91.57 19 81.21 34 78.76 43 82.87 40 87.56 20 95.58 16 82.00 31 ANet 85.16 28 85.20 29 85.14 30 84.37 36 90.29 29 82.00 29 88.39 29 80.78 33 unMVSv1 83.89 29 84.37 30 83.58 32 84.00 40 85.63 35 83.17 28 85.56 30 81.11 32 A1Net 82.90 30 89.39 22 78.57 42 85.19 30 63.18 51 84.36 26 94.42 21 87.34 24 RMVSNet 80.62 31 86.71 26 76.57 43 84.14 38 78.09 43 80.22 32 93.20 22 67.48 49 Pnet_fast 80.38 32 65.52 48 90.29 25 90.03 25 94.91 23 60.54 48 70.51 47 85.93 27 AttMVS 80.32 33 78.27 33 81.68 33 78.98 42 86.16 34 79.76 33 76.79 44 79.89 34 Pnet-new- 79.46 34 69.54 45 86.08 28 84.76 31 90.31 28 65.13 42 73.95 46 83.16 29 cscdsrtA 79.00 35 69.56 44 85.29 29 85.95 28 92.82 26 62.08 45 77.05 40 77.11 35 hgnet 78.82 36 76.20 35 80.57 38 84.50 32 82.32 41 75.56 35 76.84 41 74.88 38 DPSNet 78.82 36 76.20 35 80.57 38 84.50 32 82.32 41 75.56 35 76.84 41 74.88 38 Pnet-blend++ 78.13 38 73.62 39 81.13 35 84.46 34 88.20 31 66.50 39 80.74 34 70.74 47 Pnet-blend 78.13 38 73.62 39 81.13 35 84.46 34 88.20 31 66.50 39 80.74 34 70.74 47 cscd 78.10 40 68.51 46 84.50 31 85.47 29 91.56 27 59.79 50 77.22 39 76.47 36 Pnet-eth 77.75 41 85.52 28 72.57 48 75.57 45 65.89 50 81.44 30 89.61 26 76.26 37 MVSCRF 77.21 42 74.57 38 78.98 41 79.29 41 83.87 38 75.01 37 74.12 45 73.78 42 MVE 76.82 17 76.63 43 77.52 34 76.03 45 70.40 50 85.49 36 75.68 34 79.36 36 72.20 44 Simon Fuhrmann, Fabian Langguth, Michael Goesele: MVE - A Multi-View Reconstruction Environment . EUROGRAPHICS Workshops on Graphics and Cultural Heritage (2014) example 74.48 44 65.59 47 80.41 40 84.10 39 83.16 39 62.82 44 68.36 49 73.98 41 Snet 74.39 45 64.28 50 81.13 35 88.04 27 73.18 48 60.15 49 68.41 48 82.18 30 MVSNet 73.65 46 70.01 43 76.07 44 76.39 44 77.59 44 61.93 46 78.09 38 74.23 40 MVSNet + Gipuma 72.61 47 71.20 41 73.54 46 75.39 46 73.46 47 65.60 41 76.81 43 71.78 45 firsttry 72.49 48 75.56 37 70.44 49 68.80 52 70.25 49 69.31 38 81.82 33 72.28 43 cscdsrtB 70.07 49 71.18 42 69.33 50 69.73 51 77.06 45 64.06 43 78.31 37 61.20 50 F/T MVSNet+Gipuma 69.80 50 64.46 49 73.36 47 75.05 47 73.57 46 60.74 47 68.18 50 71.46 46 PMVS 70.75 18 58.19 51 43.22 51 68.18 51 73.06 49 84.09 37 54.23 51 32.21 51 47.37 51 Y. Furukawa, J. Ponce: Accurate, dense, and robust multiview stereopsis . PAMI (2010) FADENet 18.58 52 15.51 52 20.63 52 21.45 53 32.74 53 15.09 52 15.93 52 7.68 52 CMPMVS 69.68 19 11.01 53 0.97 53 17.70 53 10.09 54 43.01 52 1.93 53 0.00 53 0.00 53 M. Jancosek, T. Pajdla: Multi-View Reconstruction Preserving Weakly-Supported Surfaces . CVPR 2011 dnet 0.00 54 0.00 54 0.00 54 0.00 55 0.00 54 0.00 54 0.00 53 0.00 53 UnsupFinetunedMVSNet 75.05 47
