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.

DeepPCF-MVS	69.58 1	79.03 1	79.00 1	79.13 4	84.92 4	60.32 3	83.03 4	85.33 3	62.86 3	80.17 1	90.03 1	61.76 1	88.95 4	74.21 1	88.67 1	88.12 2

DeepC-MVS	68.55 2	78.02 2	77.46 2	79.68 1	84.43 5	61.68 1	84.51 3	86.17 2	62.66 4	74.68 2	89.18 3	57.25 2	89.71 1	70.47 2	87.80 3	87.66 3

DeepC-MVS_fast	68.22 3	77.40 3	76.78 3	79.27 3	86.46 1	61.11 2	85.41 2	84.91 4	63.46 2	72.86 3	88.92 4	53.24 3	89.19 3	67.88 3	87.88 2	87.52 4

3Dnovator+	66.72 4	75.84 4	74.57 4	79.66 2	82.40 6	59.92 4	85.83 1	86.32 1	66.92 1	67.80 8	89.24 2	42.03 9	89.38 2	64.07 6	86.50 4	89.69 1

MG-MVS		73.96 5	73.89 5	74.16 10	85.65 2	49.69 19	81.59 8	81.29 12	61.45 6	71.05 4	88.11 5	51.77 6	87.73 5	61.05 7	83.09 7	85.05 8

CLD-MVS		73.33 6	72.68 6	75.29 9	78.82 12	53.33 14	78.23 17	84.79 5	61.30 7	70.41 5	81.04 18	52.41 5	87.12 6	64.61 5	82.49 9	85.41 6

3Dnovator	64.47 5	72.49 7	71.39 7	75.79 5	77.70 15	58.99 6	80.66 11	83.15 6	62.24 5	65.46 10	86.59 6	42.38 8	85.52 9	59.59 9	84.72 5	82.85 10

ACMP	63.53 6	72.30 8	71.20 8	75.59 7	80.28 10	57.54 9	82.74 5	82.84 8	60.58 8	65.24 11	86.18 7	39.25 11	86.03 8	66.95 4	76.79 13	83.22 9
Qingshan Xu and Wenbing Tao: Planar Prior Assisted PatchMatch Multi-View Stereo. AAAI 2020
MAR-MVS		71.51 9	70.15 9	75.60 6	81.84 7	59.39 5	81.38 9	82.90 7	54.90 13	68.08 7	78.70 21	47.73 7	85.51 10	51.68 11	84.17 6	81.88 11
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
PCF-MVS	61.88 8	70.95 10	69.49 11	75.35 8	77.63 16	55.71 10	76.04 19	81.81 10	50.30 18	69.66 6	85.40 9	52.51 4	84.89 11	51.82 10	80.24 10	85.45 5
Andreas Kuhn, Shan Lin, Oliver Erdler: Plane Completion and Filtering for Multi-View Stereo Reconstruction. GCPR 2019
ACMM	61.98 7	70.80 11	69.73 10	74.02 11	80.59 9	58.59 7	82.68 6	82.02 9	55.46 12	67.18 9	84.39 12	38.51 14	83.17 14	60.65 8	76.10 14	80.30 12
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
AdaColmap		69.99 12	68.66 12	73.97 12	84.94 3	57.83 8	82.63 7	78.71 15	56.28 10	64.34 13	84.14 13	41.57 10	87.06 7	46.45 14	78.88 11	77.02 19

OpenMVS	61.03 9	68.85 13	67.56 13	72.70 13	74.26 20	53.99 13	81.21 10	81.34 11	52.70 15	62.75 14	85.55 8	38.86 13	84.14 13	48.41 13	83.01 8	79.97 13

TAPA-MVS	59.36 10	66.60 14	65.20 14	70.81 16	76.63 17	48.75 20	76.52 18	80.04 13	50.64 17	65.24 11	84.93 10	39.15 12	78.54 22	36.77 20	76.88 12	85.14 7
Andrea Romanoni, Matteo Matteucci: TAPA-MVS: Textureless-Aware PAtchMatch Multi-View Stereo. ICCV 2019
ACMH+	57.40 11	66.12 15	64.06 15	72.30 14	77.79 14	52.83 15	80.39 12	78.03 16	57.30 9	57.47 16	82.55 14	27.68 20	84.17 12	45.54 15	69.78 20	79.90 14

ACMH	55.70 13	65.20 16	63.57 16	70.07 18	78.07 13	52.01 18	79.48 14	79.69 14	55.75 11	56.59 18	80.98 19	27.12 21	80.94 18	42.90 16	71.58 17	77.25 18
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
PLC	56.13 12	65.09 17	63.21 17	70.72 17	81.04 8	54.87 12	78.57 15	77.47 17	48.51 20	55.71 20	81.89 16	33.71 15	79.71 19	41.66 18	70.37 19	77.58 16
Jie Liao, Yanping Fu, Qingan Yan, Chunxia xiao: Pyramid Multi-View Stereo with Local Consistency. Pacific Graphics 2019
LS3D		64.71 18	62.50 18	71.34 15	79.72 11	55.71 10	79.82 13	74.72 20	48.50 21	56.62 17	84.62 11	33.59 16	82.34 15	29.65 25	75.23 15	75.97 20

LTVRE_ROB	55.42 14	63.15 19	61.23 19	68.92 20	76.57 18	47.80 21	59.92 23	76.39 18	54.35 14	58.67 15	82.46 15	29.44 19	81.49 16	42.12 17	71.14 18	77.46 17
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
MSDG		61.81 20	59.23 20	69.55 19	72.64 21	52.63 16	70.45 21	75.81 19	51.38 16	53.70 22	76.11 23	29.52 17	81.08 17	37.70 19	65.79 23	74.93 21

COLMAP_ROB	52.97 15	61.27 21	58.81 21	68.64 21	74.63 19	52.51 17	78.42 16	73.30 21	49.92 19	50.96 23	81.51 17	23.06 23	79.40 21	31.63 23	65.85 22	74.01 22
Johannes L. Schönberger, Enliang Zheng, Marc Pollefeys, Jan-Michael Frahm: Pixelwise View Selection for Unstructured Multi-View Stereo. ECCV 2016
OpenMVS_ROB	52.78 16	60.03 22	58.14 22	65.69 22	70.47 22	44.82 23	75.33 20	70.86 22	45.04 24	56.06 19	76.00 24	26.89 22	79.65 20	35.36 22	67.29 21	72.60 23

CMPMVS	42.80 17	57.81 23	55.97 23	63.32 23	60.98 25	47.38 22	64.66 22	69.50 24	32.06 25	46.83 24	77.80 22	29.50 18	71.36 23	48.68 12	73.75 16	71.21 25
M. Jancosek, T. Pajdla: Multi-View Reconstruction Preserving Weakly-Supported Surfaces. CVPR 2011
LF4IMVS		42.95 24	42.26 24	45.04 27	48.30 27	32.50 28	54.80 25	48.49 29	28.03 27	40.51 25	70.16 25	9.24 27	43.89 30	31.63 23	49.18 27	58.72 26

ANet_high		41.38 25	37.47 25	53.11 25	39.73 30	24.45 33	56.94 24	69.69 23	47.65 22	26.04 27	52.32 27	12.44 25	62.38 26	21.80 27	10.61 33	72.49 24

PMVS	28.69 18	36.22 26	33.29 26	45.02 28	36.82 31	35.98 27	54.68 26	48.74 28	26.31 28	21.02 28	51.61 28	2.88 30	60.10 27	9.99 32	47.58 28	38.99 30
Y. Furukawa, J. Ponce: Accurate, dense, and robust multiview stereopsis. PAMI (2010)
Gipuma		34.77 27	31.91 27	43.33 30	62.05 24	37.87 26	20.39 30	67.03 25	23.23 29	18.41 29	25.84 32	4.24 28	62.73 24	14.71 29	51.32 26	29.38 32
S. Galliani, K. Lasinger, K. Schindler: Massively Parallel Multiview Stereopsis by Surface Normal Diffusion. ICCV 2015
wuykxyi23d		28.12 28	22.54 29	44.87 29	34.97 32	32.11 29	37.96 28	47.31 30	13.32 31	9.29 33	23.72 33	0.45 33	56.58 28	21.85 26	13.98 30	45.93 27

PNet_i23d		27.88 29	25.99 28	33.55 31	47.54 28	25.89 31	47.24 27	32.91 31	21.44 30	15.90 30	38.09 29	0.85 32	42.76 31	16.90 28	13.03 31	32.00 31

MVE	17.77 19	21.41 30	17.77 30	32.34 32	34.34 33	25.44 32	16.11 31	24.11 32	11.19 32	13.22 31	31.92 30	1.58 31	30.95 32	10.47 31	17.03 29	40.62 29
Simon Fuhrmann, Fabian Langguth, Michael Goesele: MVE - A Multi-View Reconstruction Environment. EUROGRAPHICS Workshops on Graphics and Cultural Heritage (2014)
wuyk23d		13.32 31	12.52 31	15.71 33	47.54 28	26.27 30	31.06 29	1.98 34	4.93 34	5.18 34	1.94 34	0.45 33	18.54 33	6.81 34	12.83 32	2.33 34

FPMVS				45.39 26	58.03 26	41.01 25		53.81 27	30.07 26	33.71 26	64.03 26	11.69 26	52.08 29	14.01 30	55.11 25	43.09 28

DeepMVS_CX				12.03 34	17.97 34	10.91 34		10.60 33	7.46 33	11.07 32	28.36 31	3.28 29	11.29 34	8.01 33	9.74 34	13.89 33

ITE_SJBPF				62.09 24	66.16 23	44.55 24		64.32 26	47.36 23	55.31 21	80.34 20	19.27 24	62.68 25	36.29 21	62.39 24	79.04 15

High-resolution multi-view benchmark