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	81.17 1	89.72 1	91.38 1	84.72 12	93.00 6	58.16 22	96.72 1	94.41 2	86.50 1	90.25 1	97.83 1	75.46 1	98.67 3	92.78 1	95.49 1	97.32 1

DeepC-MVS_fast	79.48 2	87.95 2	88.00 2	87.79 3	95.86 5	68.32 7	95.74 2	94.11 5	83.82 3	83.49 3	96.19 2	64.53 3	98.44 5	83.42 4	94.88 2	96.61 2
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
MG-MVS		87.11 3	86.27 3	89.62 1	97.79 1	76.27 1	94.96 5	94.49 1	78.74 8	83.87 2	92.94 7	64.34 4	96.94 11	75.19 7	94.09 4	95.66 3

PAPM		85.89 4	85.46 4	87.18 5	88.20 17	72.42 2	92.41 12	92.77 8	82.11 4	80.34 4	93.07 6	68.27 2	95.02 15	78.39 6	93.59 5	94.09 9

DeepC-MVS	77.85 3	85.52 5	85.24 5	86.37 9	88.80 16	66.64 10	92.15 13	93.68 6	81.07 5	76.91 6	93.64 4	62.59 5	98.44 5	85.50 2	92.84 7	94.03 10
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
MAR-MVS		84.18 6	83.43 6	86.44 7	96.25 4	65.93 12	94.28 6	94.27 4	74.41 10	79.16 5	95.61 3	53.99 9	98.88 2	69.62 12	93.26 6	94.50 7
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
CLD-MVS		82.73 7	82.35 7	83.86 16	87.90 18	67.65 8	95.45 3	92.18 11	85.06 2	72.58 8	92.27 13	52.46 10	95.78 14	84.18 3	79.06 18	88.16 20
Zhaoxin Li, Wangmeng Zuo, Zhaoqi Wang, Lei Zhang: Confidence-based Large-scale Dense Multi-view Stereo. IEEE Transaction on Image Processing, 2020
3Dnovator	73.91 4	82.69 8	80.82 9	88.31 2	89.57 14	71.26 4	92.60 10	94.39 3	78.84 7	67.89 16	92.48 11	48.42 15	98.52 4	68.80 13	94.40 3	95.15 5

PHI-MVS		82.50 9	80.83 8	87.49 4	96.99 2	69.34 6	94.18 7	91.54 12	71.10 16	72.81 7	92.93 8	55.10 8	99.04 1	65.74 16	87.11 10	94.10 8

3Dnovator+	73.60 5	82.10 10	80.60 11	86.60 6	90.89 12	66.80 9	95.20 4	93.44 7	74.05 11	67.42 17	92.49 10	49.46 12	97.65 8	70.80 10	91.68 8	95.33 4

ACMMwithACMP		81.49 11	80.67 10	83.93 15	91.71 9	62.90 17	92.13 14	92.22 10	71.79 15	71.68 11	93.49 5	50.32 11	96.96 10	78.47 5	84.22 12	91.93 12

PVSNet	73.49 6	80.05 12	78.63 12	84.31 13	90.92 11	64.97 14	92.47 11	91.05 14	79.18 6	72.43 9	90.51 19	37.05 24	94.06 22	68.06 14	86.00 11	93.90 11

PCF-MVS	73.15 7	79.29 13	77.63 14	84.29 14	86.06 20	65.96 11	87.03 26	91.10 13	69.86 18	69.79 15	90.64 16	57.54 6	96.59 12	64.37 18	82.29 14	90.32 16
Andreas Kuhn, Shan Lin, Oliver Erdler: Plane Completion and Filtering for Multi-View Stereo Reconstruction. GCPR 2019
AdaColmap		78.94 14	77.00 15	84.76 11	96.34 3	65.86 13	92.66 9	87.97 22	62.18 22	70.56 13	92.37 12	43.53 18	97.35 9	64.50 17	82.86 13	91.05 14

OMC-MVS		78.67 15	77.91 13	80.95 19	85.76 22	57.40 25	88.49 22	88.67 20	73.85 12	72.43 9	92.10 14	49.29 13	94.55 18	72.73 9	77.89 19	90.91 15

OpenMVS	70.45 9	78.54 16	75.92 16	86.41 8	85.93 21	71.68 3	92.74 8	92.51 9	66.49 19	64.56 19	91.96 15	43.88 17	98.10 7	54.61 21	90.65 9	89.44 18

PatchmatchNet		77.46 17	74.63 17	85.96 10	89.55 15	70.35 5	79.97 30	89.55 18	72.23 14	70.94 12	76.91 34	57.03 7	92.79 28	54.27 22	81.17 15	94.74 6

ACMP	71.68 8	75.58 18	74.23 18	79.62 23	84.97 25	59.64 20	90.80 16	89.07 19	70.39 17	62.95 22	87.30 22	38.28 20	93.87 24	72.89 8	71.45 24	85.36 23
Qingshan Xu and Wenbing Tao: Planar Prior Assisted PatchMatch Multi-View Stereo. AAAI 2020
PLC	68.80 12	75.23 19	73.68 19	79.86 22	92.93 7	58.68 21	90.64 17	88.30 21	60.90 23	64.43 20	90.53 18	42.38 19	94.57 17	56.52 20	76.54 20	86.33 21
Jie Liao, Yanping Fu, Qingan Yan, Chunxia xiao: Pyramid Multi-View Stereo with Local Consistency. Pacific Graphics 2019
TAPA-MVS	70.22 10	74.94 20	73.53 20	79.17 24	90.40 13	52.07 32	89.19 20	89.61 17	62.69 21	70.07 14	92.67 9	48.89 14	94.32 19	38.26 32	79.97 16	91.12 13
Andrea Romanoni, Matteo Matteucci: TAPA-MVS: Textureless-Aware PAtchMatch Multi-View Stereo. ICCV 2019
ACMM	69.62 11	74.34 21	72.73 21	79.17 24	84.25 28	57.87 23	90.36 18	89.93 16	63.17 20	65.64 18	86.04 23	37.79 22	94.10 21	65.89 15	71.52 23	85.55 22
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
CNLPA		74.31 22	72.30 22	80.32 20	91.49 10	61.66 19	90.85 15	80.72 31	56.67 30	63.85 21	90.64 16	46.75 16	90.84 31	53.79 23	75.99 21	88.47 19

PVSNet_0	68.08 13	71.81 23	68.32 24	82.27 17	84.68 26	62.31 18	88.68 21	90.31 15	75.84 9	57.93 23	80.65 29	37.85 21	94.19 20	69.94 11	29.05 39	90.31 17

F-COLMAP		70.66 24	68.44 23	77.32 27	86.37 19	55.91 28	88.00 24	86.32 23	56.94 29	57.28 24	88.07 20	33.58 27	92.49 29	51.02 25	68.37 27	83.55 24

MSDG		69.54 25	65.73 26	80.96 18	85.11 24	63.71 15	84.19 28	83.28 29	56.95 28	54.50 27	84.03 24	31.50 28	96.03 13	42.87 28	69.13 25	83.14 25

LS3D		69.17 26	66.40 25	77.50 26	91.92 8	56.12 27	85.12 27	80.37 32	46.96 35	56.50 25	87.51 21	37.25 23	93.71 25	32.52 36	79.40 17	82.68 26

ACMH	63.93 15	68.62 27	64.81 27	80.03 21	85.22 23	63.25 16	87.72 25	84.66 26	60.83 24	51.57 31	79.43 32	27.29 29	94.96 16	41.76 29	64.84 29	81.88 27
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
ACMH+	65.35 14	67.65 28	64.55 28	76.96 28	84.59 27	57.10 26	88.08 23	80.79 30	58.59 27	53.00 30	81.09 28	26.63 30	92.95 27	46.51 26	61.69 31	80.82 29

USDC		67.43 29	64.51 29	76.19 30	77.94 32	55.29 29	78.38 31	85.00 25	73.17 13	48.36 33	80.37 30	21.23 34	92.48 30	52.15 24	64.02 30	80.81 30

CMPMVS	48.56 19	66.77 30	64.41 30	73.84 32	70.65 36	50.31 34	77.79 33	85.73 24	45.54 36	44.76 34	82.14 26	35.40 25	90.14 33	63.18 19	74.54 22	81.07 28
M. Jancosek, T. Pajdla: Multi-View Reconstruction Preserving Weakly-Supported Surfaces. CVPR 2011
OpenMVS_ROB	61.12 16	66.39 31	62.92 32	76.80 29	76.51 33	57.77 24	89.22 19	83.41 28	55.48 31	53.86 29	77.84 33	26.28 31	93.95 23	34.90 34	68.76 26	78.68 32

LTVRE_ROB	59.60 17	66.27 32	63.54 31	74.45 31	84.00 29	51.55 33	67.08 35	83.53 27	58.78 26	54.94 26	80.31 31	34.54 26	93.23 26	40.64 31	68.03 28	78.58 33
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	57.96 18	62.98 33	59.65 33	72.98 33	81.44 30	53.00 31	83.75 29	75.53 35	48.34 34	48.81 32	81.40 27	24.14 32	90.30 32	32.95 35	60.52 33	75.65 34
Johannes L. Schönberger, Enliang Zheng, Marc Pollefeys, Jan-Michael Frahm: Pixelwise View Selection for Unstructured Multi-View Stereo. ECCV 2016
TinyColmap		60.32 34	56.42 34	72.00 34	78.78 31	53.18 30	78.36 32	75.64 34	52.30 32	41.59 36	75.82 35	14.76 35	88.35 34	35.84 33	54.71 35	74.46 35

LF4IMVS		54.01 35	52.12 35	59.69 36	62.41 37	39.91 36	68.59 34	68.28 39	42.96 37	44.55 35	75.18 36	14.09 36	68.39 37	41.36 30	51.68 36	70.78 36

FPMVS		45.64 36	43.10 36	53.23 38	51.42 39	36.46 38	64.97 37	71.91 37	29.13 39	27.53 37	61.55 37	9.83 37	65.01 39	16.00 41	55.58 34	58.22 38

ANet_high		40.27 37	35.20 37	55.47 37	34.74 42	34.47 39	63.84 38	71.56 38	48.42 33	18.80 38	41.08 39	9.52 38	64.45 40	20.18 38	8.66 44	67.49 37

Gipuma		34.91 38	31.44 38	45.30 40	70.99 35	39.64 37	19.85 43	72.56 36	20.10 42	16.16 40	21.47 43	5.08 40	71.16 36	13.07 42	43.70 37	25.08 44
S. Galliani, K. Lasinger, K. Schindler: Massively Parallel Multiview Stereopsis by Surface Normal Diffusion. ICCV 2015
PNet_i23d		32.77 39	29.98 39	41.11 42	48.05 40	29.17 40	65.82 36	50.02 41	21.42 40	14.74 41	37.19 41	1.11 43	55.11 42	19.75 39	11.77 42	39.06 40

PMVS	26.43 20	31.84 40	28.16 40	42.89 41	25.87 45	27.58 42	50.92 39	49.78 42	21.37 41	14.17 42	40.81 40	2.01 41	66.62 38	9.61 44	38.88 38	34.49 43
Y. Furukawa, J. Ponce: Accurate, dense, and robust multiview stereopsis. PAMI (2010)
wuykxyi23d		29.03 41	23.09 41	46.84 39	31.67 44	28.82 41	43.46 40	57.72 40	14.39 43	7.52 44	20.84 44	0.64 44	60.29 41	21.57 37	10.04 43	51.40 39

MVE	24.84 21	24.35 42	19.77 42	38.09 43	34.56 43	26.92 43	26.57 41	38.87 43	11.73 44	11.37 43	27.44 42	1.37 42	50.42 43	11.41 43	14.60 41	36.93 41
Simon Fuhrmann, Fabian Langguth, Michael Goesele: MVE - A Multi-View Reconstruction Environment. EUROGRAPHICS Workshops on Graphics and Cultural Heritage (2014)
wuyk23d		11.30 43	10.95 43	12.33 45	48.05 40	19.89 45	25.89 42	1.92 45	3.58 45	3.12 45	1.37 45	0.64 44	15.77 45	6.23 45	7.77 45	1.35 45

DeepMVS_CX				34.71 44	51.45 38	24.73 44		28.48 44	31.46 38	17.49 39	52.75 38	5.80 39	42.60 44	18.18 40	19.42 40	36.81 42

ITE_SJBPF				70.43 35	74.44 34	47.06 35		77.32 33	60.16 25	54.04 28	83.53 25	23.30 33	84.01 35	43.07 27	61.58 32	80.21 31

High-resolution multi-view benchmark