This table lists the benchmark results for the high-res multi-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 high-res indoor outdoor botani. boulde. bridge door exhibi. lectur. living. lounge observ. old co. statue terrac.
3Dnovator+ 77.84 1 85.48 1 84.47 1 88.51 1 91.08 3 73.49 1 93.18 1 93.78 1 80.79 1 76.66 6 93.37 1 60.40 5 96.75 1 77.20 2 93.73 1 95.29 1 3Dnovator 76.31 2 83.38 2 82.31 2 86.59 2 87.94 12 72.94 2 90.64 2 92.14 3 77.21 2 75.47 7 92.83 2 58.56 6 94.72 3 73.24 5 92.71 2 92.13 3 CLD-MVS 82.31 3 81.65 3 84.29 6 88.47 11 67.73 9 85.81 13 92.35 2 75.78 3 78.33 2 86.58 15 64.01 2 94.35 7 76.05 3 87.48 6 90.79 5 MAR-MVS 81.84 4 80.70 4 85.27 3 91.32 2 71.53 3 89.82 6 90.92 9 69.77 9 78.50 1 86.21 16 62.36 3 94.52 6 65.36 6 92.05 3 89.77 7 ACMP+ 74.13 3 81.51 5 80.57 5 84.36 5 89.42 6 68.69 6 89.97 5 91.50 6 74.46 4 75.04 8 90.41 6 53.82 8 94.54 4 77.56 1 82.91 9 89.86 6 ACMM 73.20 5 80.78 6 79.84 6 83.58 8 89.31 7 68.37 7 89.99 4 91.60 4 70.28 6 77.25 4 89.66 8 53.37 10 93.53 10 74.24 4 82.85 10 88.85 8 Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo . CVPR 2019 PCF-MVS 73.52 4 80.38 7 78.84 7 85.01 4 87.71 13 68.99 5 83.65 15 91.46 7 63.00 15 77.77 3 90.28 7 66.10 1 95.09 2 61.40 9 88.22 5 90.94 4 OpenMVS 72.83 7 79.77 8 78.33 8 84.09 7 85.17 15 69.91 4 90.57 3 90.97 8 66.70 11 72.17 9 91.91 3 54.70 7 93.96 8 61.81 8 90.95 4 88.41 9 TAPA-MVS 73.13 6 79.15 9 77.94 9 82.79 11 89.59 5 62.99 16 88.16 12 91.51 5 65.77 12 77.14 5 91.09 4 60.91 4 93.21 11 50.26 17 87.05 7 92.17 2 : TAPA-MVS: Textureless-Aware PAtchMatch Multi-View Stereo . PLC 70.83 8 78.05 10 76.37 10 83.08 10 91.88 1 67.80 8 88.19 11 89.46 10 64.33 14 69.87 10 88.38 9 53.66 9 93.58 9 58.86 10 82.73 11 87.86 11 LS3D 76.95 11 74.82 11 83.37 9 90.45 4 67.36 10 89.15 7 86.94 14 61.87 17 69.52 12 90.61 5 51.71 11 94.53 5 46.38 19 86.71 8 88.21 10 LTVRE_ROB 69.57 9 76.25 12 74.54 12 81.41 14 88.60 9 64.38 15 79.24 18 89.12 11 70.76 5 69.79 11 87.86 10 49.09 12 93.20 12 56.21 12 80.16 13 86.65 15 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 ACMP 68.96 10 76.01 13 74.01 13 82.03 12 88.60 9 65.31 14 88.86 8 87.55 13 70.25 7 67.75 13 87.47 12 41.27 16 93.19 13 58.37 11 75.94 17 87.60 12 ACMH 67.68 11 75.89 14 73.93 14 81.77 13 88.71 8 66.61 11 88.62 9 89.01 12 69.81 8 66.78 15 86.70 14 41.95 15 91.51 15 55.64 13 78.14 14 87.17 13 Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo . CVPR 2019 MSDG 73.36 15 70.99 15 80.49 16 84.51 16 65.80 12 80.71 16 86.13 15 65.70 13 65.46 16 83.74 18 44.60 13 90.91 17 51.13 16 76.89 15 84.74 17 COLMAP_ROB 66.92 12 73.01 16 70.41 16 80.81 15 87.13 14 65.63 13 88.30 10 84.19 16 62.96 16 63.80 18 87.69 11 38.04 17 92.56 14 46.66 18 74.91 18 84.24 18 Johannes L. Schönberger, Enliang Zheng, Marc Pollefeys, Jan-Michael Frahm: Pixelwise View Selection for Unstructured Multi-View Stereo . ECCV 2016 OpenMVS_ROB 64.09 13 70.56 17 68.19 17 77.65 18 80.26 18 59.41 18 85.01 14 82.96 19 58.76 18 65.43 17 82.33 19 37.63 18 91.23 16 45.34 20 76.03 16 82.32 19 CMPMVS 51.72 14 70.19 18 68.16 18 76.28 19 73.15 20 57.55 19 79.47 17 83.92 17 48.02 20 56.48 20 84.81 17 43.13 14 86.42 19 62.67 7 81.81 12 84.89 16 M. Jancosek, T. Pajdla: Multi-View Reconstruction Preserving Weakly-Supported Surfaces . CVPR 2011 LF4IMVS 64.02 19 62.19 19 69.50 20 70.90 21 53.29 20 76.13 19 77.18 21 52.65 19 58.59 19 80.98 20 23.55 20 76.52 20 53.06 15 66.66 20 78.68 20 Gipuma 45.18 20 41.86 20 55.16 23 77.03 19 51.52 21 32.50 21 80.52 20 32.46 23 27.12 22 35.02 25 9.52 23 75.50 22 22.31 22 60.21 22 38.45 24 S. Galliani, K. Lasinger, K. Schindler: Massively Parallel Multiview Stereopsis by Surface Normal Diffusion . ICCV 2015 PMVS 37.38 15 44.16 21 40.28 21 55.82 22 40.82 24 42.54 23 65.12 20 63.99 23 34.43 22 24.48 23 57.12 22 3.92 24 76.17 21 17.10 24 55.52 23 48.75 22 Y. Furukawa, J. Ponce: Accurate, dense, and robust multiview stereopsis . PAMI (2010) MVE 26.22 16 30.37 22 25.89 22 43.81 24 44.55 23 35.46 24 28.87 22 39.07 24 18.20 24 18.58 25 40.18 24 2.68 25 47.37 24 17.07 25 23.78 25 48.60 23 Simon Fuhrmann, Fabian Langguth, Michael Goesele: MVE - A Multi-View Reconstruction Environment . EUROGRAPHICS Workshops on Graphics and Cultural Heritage (2014) FPMVS 59.81 21 65.08 22 51.03 22 69.58 22 41.46 21 40.67 21 72.32 21 16.46 21 70.00 23 24.24 21 65.42 21 58.40 21 DeepMVS_CX 27.40 25 40.17 25 26.90 25 24.59 25 17.44 25 23.95 24 48.61 23 9.77 22 26.48 25 18.06 23 24.47 24 28.83 25 ITE_SJBPF 78.22 17 81.77 17 60.57 17 83.30 18 69.25 10 67.54 14 87.20 13 36.33 19 87.28 18 54.34 14 74.62 19 86.80 14
