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.

LTVRE_ROB	86.10 1	93.04 1	93.44 1	91.82 3	93.73 5	85.72 5	96.79 1	95.51 2	88.86 4	95.63 2	96.99 2	84.81 11	93.16 15	91.10 1	97.53 10	96.58 1
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
COLMAP_ROB	83.01 3	91.97 2	91.95 2	92.04 1	93.68 6	86.15 3	93.37 3	95.10 3	90.28 2	92.11 9	95.03 7	89.75 6	94.93 7	79.95 10	98.27 6	95.04 6
Johannes L. Schönberger, Enliang Zheng, Marc Pollefeys, Jan-Michael Frahm: Pixelwise View Selection for Unstructured Multi-View Stereo. ECCV 2016
ACMMwithACMP		91.91 3	91.87 3	92.03 2	95.53 2	85.91 4	93.35 4	94.16 6	82.52 9	92.39 8	94.14 12	89.15 7	95.62 3	87.35 2	98.24 7	94.56 8

ACMH+	77.89 11	90.73 4	91.50 4	88.44 13	93.00 8	76.26 22	89.65 7	95.55 1	87.72 6	93.89 5	94.94 8	91.62 1	93.44 11	78.35 13	98.76 2	95.61 3

ACMM	79.39 9	90.65 5	90.99 6	89.63 10	95.03 3	83.53 11	89.62 8	93.35 11	79.20 12	93.83 6	93.60 15	90.81 3	92.96 16	85.02 3	98.45 4	92.41 15
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
LS3D		90.60 6	90.34 9	91.38 4	89.03 22	84.23 9	93.58 2	94.68 4	90.65 1	90.33 11	93.95 13	84.50 12	95.37 4	80.87 9	95.50 20	94.53 10

ACMP	79.16 10	90.54 7	90.60 8	90.35 7	94.36 4	80.98 15	89.16 12	94.05 8	79.03 14	92.87 7	93.74 14	90.60 5	95.21 5	82.87 7	98.76 2	94.87 7
Qingshan Xu and Wenbing Tao: Planar Prior Assisted PatchMatch Multi-View Stereo. AAAI 2020
PMVS	80.48 6	90.08 8	90.66 7	88.34 15	96.71 1	92.97 2	90.31 6	89.57 24	88.51 5	90.11 12	95.12 6	90.98 2	88.92 26	77.55 16	97.07 12	83.13 36
Y. Furukawa, J. Ponce: Accurate, dense, and robust multiview stereopsis. PAMI (2010)
3Dnovator+	83.92 2	89.97 9	89.66 10	90.92 5	91.27 14	81.66 14	91.25 5	94.13 7	88.89 3	88.83 16	94.26 11	77.55 23	95.86 1	84.88 4	95.87 17	95.24 5

ACMH	76.49 14	89.34 10	91.14 5	83.96 25	92.50 9	70.36 29	89.55 9	93.84 9	81.89 10	94.70 4	95.44 4	90.69 4	88.31 28	83.33 6	98.30 5	93.20 14
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
DeepC-MVS	82.31 4	89.15 11	89.08 11	89.37 12	93.64 7	79.07 19	88.54 15	94.20 5	73.53 22	89.71 14	94.82 10	85.09 10	95.77 2	84.17 5	98.03 8	93.26 12
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
wuykxyi23d		88.46 12	88.80 12	87.44 18	90.96 15	93.03 1	85.85 22	81.96 37	74.58 21	98.58 1	97.29 1	87.73 8	87.31 29	82.84 8	99.41 1	81.99 38

OMC-MVS		88.19 13	87.52 13	90.19 8	91.94 11	81.68 13	87.49 17	93.17 12	76.02 18	88.64 17	91.22 21	84.24 13	93.37 12	77.97 15	97.03 13	95.52 4

DeepPCF-MVS	81.24 5	87.28 14	86.21 15	90.49 6	91.48 12	84.90 8	83.41 26	92.38 15	70.25 28	89.35 15	90.68 24	82.85 15	94.57 8	79.55 11	95.95 15	92.00 17

DeepC-MVS_fast	80.27 8	86.23 15	85.65 17	87.96 16	91.30 13	76.92 21	87.19 18	91.99 16	70.56 27	84.96 23	90.69 23	80.01 20	95.14 6	78.37 12	95.78 18	91.82 18
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
TAPA-MVS	77.73 12	85.71 16	84.83 18	88.37 14	88.78 23	79.72 18	87.15 19	93.50 10	69.17 29	85.80 20	89.56 26	80.76 18	92.13 20	73.21 20	95.51 19	93.25 13
Andrea Romanoni, Matteo Matteucci: TAPA-MVS: Textureless-Aware PAtchMatch Multi-View Stereo. ICCV 2019
F-COLMAP		84.97 17	83.42 22	89.63 10	92.39 10	83.40 12	88.83 13	91.92 17	73.19 23	80.18 33	89.15 27	77.04 25	93.28 14	65.82 27	92.28 29	92.21 16

3Dnovator	80.37 7	84.80 18	84.71 19	85.06 21	86.36 29	74.71 25	88.77 14	90.00 22	75.65 19	84.96 23	93.17 16	74.06 29	91.19 22	78.28 14	91.09 31	89.29 22

Gipuma		84.44 19	86.33 14	78.78 31	84.20 33	73.57 26	89.55 9	90.44 20	84.24 7	84.38 26	94.89 9	76.35 27	80.40 37	76.14 17	96.80 14	82.36 37
S. Galliani, K. Lasinger, K. Schindler: Massively Parallel Multiview Stereopsis by Surface Normal Diffusion. ICCV 2015
PHI-MVS		84.08 20	84.14 20	83.90 26	87.97 26	75.63 23	89.44 11	89.40 26	78.16 16	84.70 25	86.59 34	78.40 21	91.24 21	69.61 23	92.98 27	84.82 32

AdaColmap		83.66 21	83.69 21	83.57 27	90.05 18	72.26 28	86.29 20	90.00 22	78.19 15	81.65 32	87.16 33	83.40 14	94.24 9	61.69 32	94.76 23	84.21 34

CNLPA		83.55 22	83.10 23	84.90 22	89.34 20	83.87 10	84.54 25	88.77 28	79.09 13	83.54 28	88.66 28	74.87 28	81.73 36	66.84 26	92.29 28	89.11 23

CLD-MVS		83.18 23	82.64 24	84.79 23	89.05 21	67.82 32	77.93 36	92.52 14	68.33 30	85.07 22	81.54 39	82.06 17	92.96 16	69.35 24	97.91 9	93.57 11
Zhaoxin Li, Wangmeng Zuo, Zhaoqi Wang, Lei Zhang: Confidence-based Large-scale Dense Multi-view Stereo. IEEE Transaction on Image Processing, 2020
ANet_high		83.17 24	85.68 16	75.65 37	81.24 38	45.26 43	79.94 34	92.91 13	83.83 8	91.33 10	96.88 3	80.25 19	85.92 34	68.89 25	95.89 16	95.76 2

LF4IMVS		82.75 25	81.93 27	85.19 20	82.08 36	80.15 17	85.53 24	88.76 29	68.01 32	85.58 21	87.75 29	71.80 33	86.85 31	74.02 18	93.87 26	88.58 25

PCF-MVS	74.62 15	82.15 26	80.92 29	85.84 19	89.43 19	72.30 27	80.53 32	91.82 18	57.36 38	87.81 18	89.92 25	77.67 22	93.63 10	58.69 36	95.08 22	91.58 19
Andreas Kuhn, Shan Lin, Oliver Erdler: Plane Completion and Filtering for Multi-View Stereo Reconstruction. GCPR 2019
PLC	73.85 16	82.09 27	80.31 30	87.45 17	90.86 16	80.29 16	85.88 21	90.65 19	68.17 31	76.32 38	86.33 35	73.12 31	92.61 18	61.40 34	90.02 34	89.44 21
Jie Liao, Yanping Fu, Qingan Yan, Chunxia xiao: Pyramid Multi-View Stereo with Local Consistency. Pacific Graphics 2019
OpenMVS	76.72 13	81.98 28	82.00 26	81.93 28	84.42 31	68.22 30	88.50 16	89.48 25	66.92 34	81.80 31	91.86 19	72.59 32	90.16 25	71.19 21	91.25 30	87.40 26

TinyColmap		81.25 29	82.34 25	77.99 34	85.33 30	60.68 35	82.32 27	88.33 30	71.26 26	86.97 19	92.22 18	77.10 24	86.98 30	62.37 30	95.17 21	86.31 29

MG-MVS		80.32 30	80.94 28	78.47 33	88.18 25	52.62 40	82.29 28	85.01 34	72.01 24	79.24 34	92.54 17	69.36 34	93.36 13	70.65 22	89.19 35	89.45 20

MAR-MVS		80.24 31	78.74 33	84.73 24	86.87 27	78.18 20	85.75 23	87.81 32	65.67 35	77.84 35	78.50 40	73.79 30	90.53 24	61.59 33	90.87 32	85.49 30
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
MSDG		80.06 32	79.99 31	80.25 29	83.91 34	68.04 31	77.51 37	89.19 27	77.65 17	81.94 30	83.45 37	76.37 26	86.31 33	63.31 29	86.59 36	86.41 27

OpenMVS_ROB	70.19 17	77.77 33	77.46 34	78.71 32	84.39 32	61.15 34	81.18 30	82.52 35	62.45 36	83.34 29	87.37 32	66.20 35	88.66 27	64.69 28	85.02 37	86.32 28

USDC		76.63 34	76.73 35	76.34 36	83.46 35	57.20 36	80.02 33	88.04 31	52.14 40	83.65 27	91.25 20	63.24 36	86.65 32	54.66 37	94.11 25	85.17 31

wuyk23d		75.13 35	79.30 32	62.63 41	75.56 41	75.18 24	80.89 31	73.10 38	75.06 20	94.76 3	95.32 5	87.73 8	52.85 43	34.16 43	97.11 11	59.85 41

CMPMVS	59.41 18	75.12 36	73.57 36	79.77 30	75.84 40	67.22 33	81.21 29	82.18 36	50.78 42	76.50 37	87.66 30	55.20 39	82.99 35	62.17 31	90.64 33	89.09 24
M. Jancosek, T. Pajdla: Multi-View Reconstruction Preserving Weakly-Supported Surfaces. CVPR 2011
FPMVS		72.29 37	72.00 37	73.14 38	88.63 24	85.00 6	74.65 39	67.39 41	71.94 25	77.80 36	87.66 30	50.48 42	75.83 39	49.95 38	79.51 40	58.58 43

PAPM		71.77 38	70.06 38	76.92 35	86.39 28	53.97 37	76.62 38	86.62 33	53.44 39	63.97 41	84.73 36	57.79 37	92.34 19	39.65 40	81.33 39	84.45 33

PatchmatchNet		69.71 39	68.83 39	72.33 39	77.66 39	53.60 38	79.29 35	69.99 40	57.66 37	72.53 39	82.93 38	46.45 43	80.08 38	60.91 35	72.09 41	83.31 35

PVSNet	58.17 19	66.41 40	65.63 40	68.75 40	81.96 37	49.88 42	62.19 40	72.51 39	51.03 41	68.04 40	75.34 41	50.84 41	74.77 40	45.82 39	82.96 38	81.60 39

PVSNet_0	51.08 20	56.10 41	54.97 41	59.48 42	75.12 43	53.28 39	55.16 41	61.89 42	44.30 43	59.16 42	62.48 44	54.22 40	65.91 41	35.40 42	47.01 43	59.25 42

PNet_i23d		52.13 42	51.24 42	54.79 44	75.56 41	45.26 43	54.54 42	52.55 43	66.95 33	57.19 43	65.82 43	13.15 45	63.40 42	36.39 41	39.04 44	55.71 44

MVE	40.22 21	51.82 43	50.47 43	55.87 43	62.66 44	51.91 41	31.61 43	39.28 44	40.65 44	50.76 44	74.98 42	56.24 38	44.67 44	33.94 44	64.11 42	71.04 40
Simon Fuhrmann, Fabian Langguth, Michael Goesele: MVE - A Multi-View Reconstruction Environment. EUROGRAPHICS Workshops on Graphics and Cultural Heritage (2014)
ITE_SJBPF				90.11 9	90.72 17	84.97 7		90.30 21	81.56 11	90.02 13	91.20 22	82.40 16	90.81 23	73.58 19	94.66 24	94.56 8

DeepMVS_CX				24.13 45	32.95 45	29.49 45		21.63 45	12.07 45	37.95 45	45.07 45	30.84 44	19.21 45	17.94 45	33.06 45	23.69 45

High-resolution multi-view benchmark