High-res multi-view results

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.

Click a dataset result cell to show a visualization of the reconstruction. For training datasets, ground truth and accuracy / completeness visualizations are also available. The visualizations may not work with mobile browsers.

Method	all	high-res multi-view	indoor	outdoor	botani.	boulde.	bridge	door	exhibi.	lectur.	living.	lounge	observ.	old co.	statue	terrac.

DeepC-MVS_fast	98.59 1	99.49 1	99.37 1	99.83 1	99.82 3	99.71 2	99.37 12	99.48 16	99.23 2	99.81 3	99.86 8	98.10 4	99.94 5	98.81 5	99.89 6	99.85 1

DeepC-MVS	98.44 2	99.42 2	99.32 3	99.74 4	99.83 2	99.58 6	99.61 5	99.54 9	98.90 6	99.86 1	99.88 5	97.55 8	99.96 4	98.73 7	99.96 1	99.67 6

LS3D		99.27 3	99.12 4	99.74 4	99.18 20	99.75 1	99.56 7	99.57 6	98.45 8	99.49 5	99.85 9	97.77 7	99.94 5	98.33 9	99.84 10	99.52 11

3Dnovator	97.25 8	99.24 4	99.05 5	99.81 3	99.12 22	99.66 4	99.84 1	99.74 1	99.09 4	98.92 13	99.90 2	95.94 11	99.98 1	98.95 2	99.92 2	99.79 5

3Dnovator+	97.12 12	99.18 5	98.97 7	99.82 2	99.17 21	99.68 3	99.81 3	99.51 13	99.20 3	98.72 15	99.89 3	95.68 13	99.97 3	98.86 4	99.86 8	99.81 3

MG-MVS		99.13 6	99.02 6	99.45 12	99.57 9	98.63 12	99.07 18	99.34 22	98.99 5	99.61 4	99.82 12	97.98 5	99.87 11	97.00 19	99.80 11	99.85 1

PLC	97.94 4	99.02 7	98.85 8	99.53 8	99.66 6	99.01 11	99.24 16	99.52 12	96.85 19	99.27 9	99.48 23	98.25 3	99.91 7	97.76 14	99.62 14	99.65 7
Jie Liao, Yanping Fu, Qingan Yan, Chunxia xiao: Pyramid Multi-View Stereo with Local Consistency. Pacific Graphics 2019
AdaColmap		99.01 8	98.80 9	99.66 6	99.56 10	99.54 7	99.18 17	99.70 3	98.18 11	99.35 7	99.63 20	96.32 10	99.90 9	97.48 16	99.77 12	99.55 9

MSDG		98.98 9	98.80 9	99.53 8	99.76 4	99.19 9	98.75 20	99.55 8	97.25 17	99.47 6	99.77 14	97.82 6	99.87 11	96.93 20	99.90 4	99.54 10

MAR-MVS		98.86 10	98.63 12	99.54 7	99.37 14	99.66 4	99.45 10	99.54 9	96.61 20	99.01 12	99.40 24	97.09 9	99.86 13	97.68 15	99.53 15	99.10 13
Zhenyu Xu, Yiguang Liu, Xuelei Shi, Ying Wang, Yunan Zheng: MARMVS: Matching Ambiguity Reduced Multiple View Stereo for Efficient Large Scale Scene Reconstruction. CVPR 2020
COLMAP_ROB	97.56 6	98.86 10	98.75 11	99.17 13	99.88 1	98.53 13	99.34 13	99.59 5	97.55 15	98.70 16	99.89 3	95.83 12	99.90 9	98.10 11	99.90 4	99.08 14
Johannes L. Schönberger, Enliang Zheng, Marc Pollefeys, Jan-Michael Frahm: Pixelwise View Selection for Unstructured Multi-View Stereo. ECCV 2016
DeepPCF-MVS	98.18 3	98.81 12	99.37 1	97.12 23	99.60 8	91.75 24	98.61 22	99.44 20	99.35 1	99.83 2	99.85 9	98.70 2	99.81 14	99.02 1	99.91 3	99.81 3

OpenMVS	96.50 15	98.47 13	98.12 14	99.52 10	99.04 23	99.53 8	99.82 2	99.72 2	94.56 21	98.08 22	99.88 5	94.73 14	99.98 1	97.47 17	99.76 13	99.06 15

ACMM	97.58 5	98.37 14	98.34 13	98.48 15	99.41 12	97.10 15	99.56 7	99.45 18	98.53 7	99.04 11	99.85 9	93.00 16	99.71 18	98.74 6	97.45 20	98.64 18
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
CLD-MVS		98.16 15	98.10 15	98.33 19	99.29 16	96.82 18	98.75 20	99.44 20	97.83 14	99.13 10	99.55 22	92.92 17	99.67 20	98.32 10	97.69 18	98.48 21

ACMH	97.28 7	98.10 16	97.99 16	98.44 17	99.41 12	96.96 17	99.60 6	99.56 7	98.09 12	98.15 21	99.91 1	90.87 22	99.70 19	98.88 3	97.45 20	98.67 17
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
ACMP	97.20 10	98.06 17	97.94 17	98.45 16	99.37 14	97.01 16	99.44 11	99.49 15	97.54 16	98.45 19	99.79 13	91.95 18	99.72 16	97.91 13	97.49 19	98.62 19
Qingshan Xu and Wenbing Tao: Planar Prior Assisted PatchMatch Multi-View Stereo. AAAI 2020
LTVRE_ROB	97.16 11	98.02 18	97.90 18	98.40 18	99.23 19	96.80 19	99.70 4	99.60 4	97.12 18	98.18 20	99.70 16	91.73 19	99.72 16	98.39 8	97.45 20	98.68 16
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
ACMH+	97.24 9	97.92 19	97.78 19	98.32 20	99.46 11	96.68 20	99.56 7	99.54 9	98.41 9	97.79 23	99.87 7	90.18 23	99.66 21	98.05 12	97.18 24	98.62 19

TAPA-MVS	97.07 14	97.74 20	97.34 21	98.94 14	99.70 5	97.53 14	99.25 15	99.51 13	91.90 22	99.30 8	99.63 20	98.78 1	99.64 22	88.09 26	99.87 7	99.65 7
Andrea Romanoni, Matteo Matteucci: TAPA-MVS: Textureless-Aware PAtchMatch Multi-View Stereo. ICCV 2019
PCF-MVS	97.08 13	97.66 21	97.06 22	99.47 11	99.61 7	99.09 10	98.04 24	99.25 23	91.24 23	98.51 18	99.70 16	94.55 15	99.91 7	92.76 24	99.85 9	99.42 12
Andreas Kuhn, Shan Lin, Oliver Erdler: Plane Completion and Filtering for Multi-View Stereo Reconstruction. GCPR 2019
LF4IMVS		97.52 22	97.46 20	97.70 22	98.98 24	95.55 22	99.29 14	98.82 27	98.07 13	98.66 17	99.64 19	89.97 24	99.61 24	97.01 18	96.68 26	97.94 24

OpenMVS_ROB	92.34 16	94.38 23	93.70 24	96.41 24	97.38 26	93.17 23	99.06 19	98.75 28	86.58 26	94.84 25	98.26 27	81.53 26	99.32 25	89.01 25	97.87 16	96.76 25

CMPMVS	69.68 19	94.13 24	94.90 23	91.84 27	97.24 27	80.01 27	98.52 23	99.48 16	89.01 24	91.99 27	99.67 18	85.67 25	99.13 26	95.44 22	97.03 25	96.39 26
M. Jancosek, T. Pajdla: Multi-View Reconstruction Preserving Weakly-Supported Surfaces. CVPR 2011
Gipuma		90.99 25	90.15 25	93.51 25	98.73 25	90.12 25	93.98 29	99.45 18	79.32 28	92.28 26	94.91 28	69.61 27	97.98 27	87.42 27	95.67 27	92.45 28
S. Galliani, K. Lasinger, K. Schindler: Massively Parallel Multiview Stereopsis by Surface Normal Diffusion. ICCV 2015
PNet_i23d		79.43 26	77.68 26	84.67 29	86.18 29	71.69 31	96.50 27	93.68 30	75.17 29	71.33 29	91.18 30	32.18 30	90.62 30	78.57 30	74.34 30	91.71 30

ANet_high		77.30 27	74.86 28	84.62 30	75.88 33	77.61 28	97.63 25	93.15 32	88.81 25	64.27 31	89.29 31	36.51 29	83.93 33	75.89 31	52.31 33	92.33 29

MVE	76.82 17	76.91 28	74.31 29	84.70 28	85.38 31	76.05 29	96.88 26	93.17 31	67.39 32	71.28 30	89.01 32	21.66 34	87.69 32	71.74 33	72.29 31	90.35 31
Simon Fuhrmann, Fabian Langguth, Michael Goesele: MVE - A Multi-View Reconstruction Environment. EUROGRAPHICS Workshops on Graphics and Cultural Heritage (2014)
PMVS	70.75 18	75.98 29	74.97 27	79.01 33	70.98 34	55.18 33	93.37 30	98.21 29	65.08 33	61.78 33	93.83 29	21.74 33	92.53 28	78.59 29	91.12 29	89.34 32
Y. Furukawa, J. Ponce: Accurate, dense, and robust multiview stereopsis. PAMI (2010)
wuykxyi23d		74.42 30	71.19 30	84.14 31	76.16 32	74.29 30	96.00 28	92.57 33	69.57 31	63.84 32	87.49 33	21.98 31	88.86 31	75.56 32	57.50 32	89.26 33

wuyk23d		40.18 31	41.29 31	36.84 34	86.18 29	49.12 34	79.73 31	22.81 34	27.64 34	25.46 34	28.45 34	21.98 31	48.89 34	55.80 34	23.56 34	12.51 34

FPMVS				82.75 32	86.77 28	63.39 32		98.92 25	74.11 30	83.39 28	98.98 26	50.85 28	92.40 29	84.54 28	94.97 28	92.46 27

DeepMVS_CX				93.34 26	99.29 16	82.27 26		99.22 24	85.15 27	96.33 24	99.05 25	90.97 21	99.73 15	93.57 23	97.77 17	98.01 23

ITE_SJBPF				98.08 21	99.29 16	96.37 21		98.92 25	98.34 10	98.83 14	99.75 15	91.09 20	99.62 23	95.82 21	97.40 23	98.25 22

High-resolution multi-view benchmark