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	courty.	delive.	electro	facade	kicker	meadow	office	pipes	playgr.	relief	relief.	terrace	terrai.

DeepPCF-MVS	79.04 1	85.30 1	88.93 1	81.06 6	88.77 5	90.48 2	85.46 11	73.08 3	90.97 1	73.77 8	84.81 1	85.95 2	77.43 3	88.22 2	87.73 2	87.85 21	94.34 2

CSCG		85.28 2	87.68 2	82.49 3	89.95 2	91.99 1	88.82 1	71.20 7	86.41 3	79.63 1	79.26 4	88.36 1	73.94 9	86.64 6	86.67 6	91.40 1	94.41 1

DeepC-MVS	78.47 2	84.81 3	86.03 3	83.37 1	89.29 3	90.38 3	88.61 2	76.50 1	86.25 4	77.22 3	75.12 7	80.28 5	77.59 2	88.39 1	88.17 1	91.02 2	93.66 3
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
DeepC-MVS_fast	78.24 3	84.27 4	85.50 5	82.85 2	90.46 1	89.24 6	87.83 4	74.24 2	84.88 6	76.23 4	75.26 6	81.05 4	77.62 1	88.02 3	87.62 3	90.69 4	92.41 4
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
ACMMP		83.42 5	85.27 6	81.26 5	88.47 6	88.49 7	88.31 3	72.09 4	83.42 8	72.77 10	82.65 2	78.22 8	75.18 6	86.24 8	85.76 7	90.74 3	92.13 5

3Dnovator+	75.73 4	82.40 6	82.76 7	81.97 4	88.02 7	89.67 4	86.60 5	71.48 6	81.28 10	78.18 2	64.78 24	77.96 9	77.13 4	87.32 4	86.83 4	90.41 5	91.48 7

PHI-MVS		82.36 7	85.89 4	78.24 13	86.40 10	89.52 5	85.52 10	69.52 11	82.38 9	65.67 22	81.35 3	82.36 3	73.07 14	87.31 5	86.76 5	89.24 11	91.56 6

OMC-MVS		80.26 8	82.59 8	77.54 15	83.04 18	85.54 14	83.25 16	65.05 20	87.32 2	72.42 11	72.04 12	78.97 6	73.30 12	83.86 14	81.60 18	88.15 15	88.83 13

MVS_111021_HR		80.13 9	81.46 9	78.58 11	85.77 11	85.17 15	83.45 15	69.28 12	74.08 21	70.31 17	74.31 9	75.26 13	73.13 13	86.46 7	85.15 9	89.53 9	89.81 10

ACMP	73.23 7	79.79 10	80.53 11	78.94 9	85.61 12	85.68 13	85.61 9	69.59 10	77.33 14	71.00 16	74.45 8	69.16 27	71.88 17	83.15 19	83.37 14	89.92 7	90.57 9
Qingshan Xu and Wenbing Tao: Planar Prior Assisted PatchMatch Multi-View Stereo. AAAI 2020
3Dnovator	73.76 5	79.75 11	80.52 12	78.84 10	84.94 16	87.35 10	84.43 13	65.54 18	78.29 13	73.97 7	63.00 28	75.62 12	74.07 8	85.00 11	85.34 8	90.11 6	89.04 11

AdaColmap		79.74 12	78.62 17	81.05 7	89.23 4	86.06 12	84.95 12	71.96 5	79.39 12	75.51 5	63.16 26	68.84 30	76.51 5	83.55 16	82.85 15	88.13 16	86.46 24

PCF-MVS	73.28 6	79.42 13	80.41 13	78.26 12	84.88 17	88.17 8	86.08 6	69.85 9	75.23 18	68.43 18	68.03 19	78.38 7	71.76 19	81.26 27	80.65 25	88.56 13	91.18 8
Andreas Kuhn, Shan Lin, Oliver Erdler: Plane Completion and Filtering for Multi-View Stereo Reconstruction. GCPR 2019
CLD-MVS		79.35 14	81.23 10	77.16 16	85.01 14	86.92 11	85.87 8	60.89 41	80.07 11	75.35 6	72.96 11	73.21 14	68.43 25	85.41 10	84.63 11	87.41 23	85.44 26
Zhaoxin Li, Wangmeng Zuo, Zhaoqi Wang, Lei Zhang: Confidence-based Large-scale Dense Multi-view Stereo. IEEE Transaction on Image Processing, 2020
MAR-MVS		79.21 15	80.32 14	77.92 14	87.46 8	88.15 9	83.95 14	67.48 16	74.28 19	68.25 19	64.70 25	77.04 10	72.17 16	85.42 9	85.00 10	88.22 14	87.62 19
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
ACMM	72.26 8	78.86 16	78.13 18	79.71 8	86.89 9	83.40 24	86.02 7	70.50 8	75.28 17	71.49 14	63.01 27	69.26 26	73.57 11	84.11 13	83.98 12	89.76 8	87.84 17
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
MVS_111021_LR		78.13 17	79.85 16	76.13 18	81.12 25	81.50 29	80.28 18	65.25 19	76.09 16	71.32 15	76.49 5	72.87 15	72.21 15	82.79 20	81.29 20	86.59 30	87.91 16

TAPA-MVS	71.42 9	77.69 18	80.05 15	74.94 21	80.68 26	84.52 17	81.36 17	63.14 28	84.77 7	64.82 23	68.72 15	75.91 11	71.86 18	81.62 23	79.55 32	87.80 22	85.24 27
Andrea Romanoni, Matteo Matteucci: TAPA-MVS: Textureless-Aware PAtchMatch Multi-View Stereo. ICCV 2019
CNLPA		77.20 19	77.54 19	76.80 17	82.63 20	84.31 18	79.77 20	64.64 21	85.17 5	73.18 9	56.37 42	69.81 23	74.53 7	81.12 28	78.69 37	86.04 34	87.29 21

PVSNet_Blended		76.21 20	77.52 21	74.69 23	79.46 29	83.79 21	77.50 26	64.34 24	69.88 24	71.88 12	68.54 17	70.42 21	67.05 26	83.48 17	79.63 29	87.89 19	86.87 22

PVSNet_BlendedMVS		76.21 20	77.52 21	74.69 23	79.46 29	83.79 21	77.50 26	64.34 24	69.88 24	71.88 12	68.54 17	70.42 21	67.05 26	83.48 17	79.63 29	87.89 19	86.87 22

OpenMVS	70.44 10	76.15 22	76.82 26	75.37 20	85.01 14	84.79 16	78.99 23	62.07 37	71.27 23	67.88 20	57.91 38	72.36 16	70.15 21	82.23 22	81.41 19	88.12 17	87.78 18

PLC	68.99 11	75.68 23	75.31 27	76.12 19	82.94 19	81.26 30	79.94 19	66.10 17	77.15 15	66.86 21	59.13 35	68.53 31	73.73 10	80.38 31	79.04 35	87.13 24	81.68 36
Jie Liao, Yanping Fu, Qingan Yan, Chunxia xiao: Pyramid Multi-View Stereo with Local Consistency. Pacific Graphics 2019
LS3D		74.08 24	73.39 29	74.88 22	85.05 13	82.62 27	79.71 21	68.66 13	72.82 22	58.80 28	57.61 39	61.31 37	71.07 20	80.32 32	78.87 36	86.00 35	80.18 37

EPP-MVSNet		74.00 25	77.41 23	70.02 30	80.53 27	83.91 20	74.99 34	62.68 34	65.06 31	49.77 45	68.68 16	72.09 17	63.06 33	82.49 21	80.73 21	89.12 12	88.91 12

IS_MVSNet		73.33 26	77.34 24	68.65 33	81.29 24	83.47 23	74.45 37	63.58 27	65.75 28	48.49 48	67.11 22	70.61 20	54.63 49	84.51 12	83.58 13	89.48 10	86.34 25

Vis-MVSNet		72.77 27	77.20 25	67.59 37	74.19 35	84.01 19	76.61 28	61.69 40	60.62 42	50.61 41	70.25 13	71.31 18	55.57 48	83.85 15	82.28 16	86.90 25	88.08 15
Jingyang Zhang, Yao Yao, Shiwei Li, Zixin Luo, Tian Fang: Visibility-aware Multiview Stereo Network. BMVC 2020
IterMVS-LS		71.69 28	72.82 32	70.37 29	77.54 33	76.34 42	75.13 33	60.46 42	61.53 41	57.57 29	64.89 23	67.33 33	66.04 28	77.09 38	77.37 39	85.48 38	85.18 28

MSDG		71.52 29	69.87 39	73.44 25	82.21 23	79.35 32	79.52 22	64.59 22	66.15 27	61.87 25	53.21 47	56.09 43	65.85 29	78.94 36	78.50 38	86.60 29	76.85 42

ACMH+	66.54 13	71.36 30	70.09 37	72.85 26	82.59 21	81.13 31	78.56 24	68.04 15	61.55 40	52.52 33	51.50 49	54.14 46	68.56 24	78.85 37	79.50 33	86.82 26	83.94 30

UGNet		71.34 31	77.54 19	64.11 43	71.92 41	82.63 26	71.66 42	62.36 35	65.59 29	51.91 36	73.52 10	70.93 19	41.16 58	81.33 24	80.50 26	85.65 37	88.25 14

IB-MVS	66.94 12	71.21 32	71.66 34	70.68 28	79.18 31	82.83 25	72.61 40	61.77 39	59.66 46	63.44 24	53.26 46	59.65 39	59.16 39	76.78 40	82.11 17	87.90 18	87.33 20
Christian Sormann, Mattia Rossi, Andreas Kuhn and Friedrich Fraundorfer: IB-MVS: An Iterative Algorithm for Deep Multi-View Stereo based on Binary Decisions. BMVC 2021
GBi-Net		70.78 33	73.37 30	67.76 34	72.95 37	78.00 33	75.15 31	62.72 31	64.13 32	51.44 37	58.37 36	69.02 28	57.59 42	81.33 24	80.72 22	86.70 27	82.02 32

ACMH	65.37 14	70.71 34	70.00 38	71.54 27	82.51 22	82.47 28	77.78 25	68.13 14	56.19 50	46.06 51	54.30 45	51.20 50	68.68 23	80.66 30	80.72 22	86.07 33	84.45 29
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
FMVSNet3		70.49 35	72.90 31	67.67 36	72.88 39	77.98 35	74.96 35	62.72 31	64.13 32	51.44 37	58.37 36	69.02 28	57.43 44	79.43 34	79.57 31	86.59 30	81.81 35

FMVSNet2		70.39 36	72.67 33	67.72 35	72.95 37	78.00 33	75.15 31	62.69 33	63.29 36	51.25 39	55.64 43	68.49 32	57.59 42	80.91 29	80.35 27	86.70 27	82.02 32

FMVSNet1		68.84 37	70.47 36	66.94 39	71.35 43	77.68 37	74.71 36	62.35 36	56.93 48	49.94 44	50.01 50	64.59 35	57.07 46	81.33 24	80.72 22	86.25 32	82.00 34

Vis-MVSNet (Re-imp)		67.83 38	73.52 28	61.19 49	78.37 32	76.72 40	66.80 51	62.96 29	65.50 30	34.17 60	67.19 21	69.68 24	39.20 60	79.39 35	79.44 34	85.68 36	76.73 43

PatchMatch-RL		67.78 39	66.65 45	69.10 31	73.01 36	72.69 45	68.49 45	61.85 38	62.93 38	60.20 26	56.83 41	50.42 52	69.52 22	75.62 43	74.46 44	81.51 45	73.62 48

COLMAP_ROB	62.73 15	67.66 40	66.76 44	68.70 32	80.49 28	77.98 35	75.29 30	62.95 30	63.62 34	49.96 43	47.32 53	50.72 51	58.57 40	76.87 39	75.50 42	84.94 39	75.33 46
Johannes L. Schönberger, Enliang Zheng, Marc Pollefeys, Jan-Michael Frahm: Pixelwise View Selection for Unstructured Multi-View Stereo. ECCV 2016
anonymous		67.65 41	69.83 40	65.09 41	75.39 34	76.55 41	74.42 38	63.75 26	53.55 53	49.37 47	59.41 34	62.45 36	44.44 54	79.71 33	79.82 28	83.17 42	77.36 40

RPSCF		67.64 42	71.25 35	63.43 46	61.86 56	70.73 48	67.26 48	50.86 51	74.20 20	58.91 27	67.49 20	69.33 25	64.10 30	71.41 52	68.45 53	77.61 47	77.17 41

USDC		67.36 43	67.90 43	66.74 40	71.72 42	75.23 43	71.58 43	60.28 43	67.45 26	50.54 42	60.93 29	45.20 55	62.08 35	76.56 41	74.50 43	84.25 40	75.38 45

IterMVS		66.36 44	68.30 42	64.10 44	69.48 49	74.61 44	73.41 39	50.79 52	57.30 47	48.28 49	60.64 30	59.92 38	60.85 36	74.14 47	72.66 46	81.80 44	78.82 38

TDRefinement		66.09 45	65.03 47	67.31 38	69.73 47	76.75 39	75.33 29	64.55 23	60.28 44	49.72 46	45.63 54	42.83 56	60.46 37	75.75 42	75.95 41	84.08 41	78.04 39

PMMVS		65.06 46	69.17 41	60.26 51	55.25 61	63.43 54	66.71 52	43.01 63	62.41 39	50.64 40	69.44 14	67.04 34	63.29 32	74.36 46	73.54 45	82.68 43	73.99 47

MDTV_nov1_ep13		64.37 47	65.24 46	63.37 47	68.94 50	70.81 47	72.40 41	50.29 53	60.10 45	53.91 31	60.07 32	59.15 40	57.21 45	69.43 53	67.30 54	77.47 48	69.78 50

PatchmatchNet		64.21 48	64.65 48	63.69 45	71.29 44	68.66 51	69.63 44	51.70 49	63.04 37	53.77 32	59.83 33	58.34 42	60.23 38	68.54 55	66.06 56	75.56 52	68.08 52
Fangjinhua Wang, Silvano Galliani, Christoph Vogel, Pablo Speciale, Marc Pollefeys: PatchmatchNet: Learned Multi-View Patchmatch Stereo.
TinyColmap		62.84 49	61.03 54	64.96 42	69.61 48	71.69 46	68.48 46	59.76 44	55.41 51	47.69 50	47.33 52	34.20 61	62.76 34	74.52 44	72.59 47	81.44 46	71.47 49

CMPMVS	47.78 17	62.49 50	62.52 51	62.46 48	70.01 46	70.66 49	62.97 57	51.84 48	51.98 55	56.71 30	42.87 55	53.62 47	57.80 41	72.23 51	70.37 51	75.45 53	75.91 44
M. Jancosek, T. Pajdla: Multi-View Reconstruction Preserving Weakly-Supported Surfaces. CVPR 2011
LTVRE_ROB	59.44 16	61.82 51	62.64 50	60.87 50	72.83 40	77.19 38	64.37 55	58.97 45	33.56 64	28.00 61	52.59 48	42.21 57	63.93 31	74.52 44	76.28 40	77.15 50	82.13 31
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
MDTV_nov1_ep13_2view		60.16 52	60.51 55	59.75 52	65.39 54	69.05 50	68.00 47	48.29 58	51.99 54	45.95 52	48.01 51	49.64 53	53.39 50	68.83 54	66.52 55	77.47 48	69.55 51

EPMVS		60.00 53	61.97 52	57.71 54	68.46 51	63.17 55	64.54 54	48.23 59	63.30 35	44.72 53	60.19 31	56.05 44	50.85 52	65.27 57	62.02 57	69.44 60	63.81 56

TAMVS		59.58 54	62.81 49	55.81 56	66.03 53	65.64 53	63.86 56	48.74 56	49.95 57	37.07 57	54.77 44	58.54 41	44.44 54	72.29 50	71.79 48	74.70 55	66.66 53

dps		58.64 55	58.61 57	58.68 53	70.34 45	59.09 58	67.11 49	44.24 62	60.34 43	52.03 35	57.20 40	53.25 49	49.99 53	61.49 59	58.26 60	68.37 62	60.64 58

MIMVSNet		58.52 56	61.34 53	55.22 57	60.76 57	67.01 52	66.81 50	49.02 55	56.43 49	38.90 56	40.59 58	54.54 45	40.57 59	73.16 48	71.65 49	75.30 54	66.00 54

FMVSNet5		57.24 57	60.02 56	53.99 59	56.45 60	62.74 56	65.27 53	47.03 60	55.14 52	39.55 55	40.88 56	53.42 48	41.83 56	72.35 49	71.10 50	73.79 56	64.50 55

MVS-HIRNet		54.41 58	52.10 59	57.11 55	58.99 59	56.10 60	49.68 61	49.10 54	46.18 59	52.15 34	33.18 60	46.11 54	56.10 47	63.19 58	59.70 59	76.64 51	60.25 59

FPMVS		51.87 59	50.00 61	54.07 58	66.83 52	57.25 59	60.25 58	50.91 50	50.25 56	34.36 59	36.04 59	32.02 62	41.49 57	58.98 62	56.07 61	70.56 58	59.36 60

MIMVSNet1		49.27 60	53.25 58	44.62 62	44.61 64	61.52 57	53.61 59	52.18 47	41.62 60	18.68 63	28.14 64	41.58 58	25.50 64	68.46 56	69.04 52	73.15 57	62.37 57

HyFIR lowres test		47.69 61	47.28 62	48.16 61	47.71 62	49.23 62	38.98 63	38.64 64	39.70 63	42.14 54	29.64 62	40.45 59	51.68 51	58.56 64	55.86 62	69.80 59	57.54 62

N_pmnet		47.35 62	50.13 60	44.11 63	59.98 58	51.64 61	51.86 60	44.80 61	49.58 58	20.76 62	40.65 57	40.05 60	29.64 62	59.84 60	55.15 63	57.63 63	54.00 63

PMVS	39.38 18	46.06 63	43.30 63	49.28 60	62.93 55	38.75 64	41.88 62	53.50 46	33.33 65	35.46 58	28.90 63	31.01 63	33.04 61	58.61 63	54.63 64	68.86 61	57.88 61
Y. Furukawa, J. Ponce: Accurate, dense, and robust multiview stereopsis. PAMI (2010)
new_pmnet		38.40 64	42.64 64	33.44 65	37.54 65	45.00 63	36.60 64	32.72 66	40.27 61	12.72 65	29.89 61	28.90 64	24.78 66	53.17 65	52.90 65	56.31 64	48.34 64

Gipuma		36.38 65	35.80 65	37.07 64	45.76 63	33.90 65	29.81 65	48.47 57	39.91 62	18.02 64	8.00 67	8.14 66	25.14 65	59.29 61	61.02 58	55.19 65	40.31 65
S. Galliani, K. Lasinger, K. Schindler: Massively Parallel Multiview Stereopsis by Surface Normal Diffusion. ICCV 2015
MVE	19.12 19	20.47 66	23.27 66	17.20 66	12.66 66	25.41 66	10.52 67	34.14 65	14.79 67	6.53 67	8.79 66	4.68 67	16.64 67	29.49 66	41.63 66	22.73 66	38.11 66
Simon Fuhrmann, Fabian Langguth, Michael Goesele: MVE - A Multi-View Reconstruction Environment. EUROGRAPHICS Workshops on Graphics and Cultural Heritage (2014)
DeepMVS_CX						18.74 67	18.55 66	8.02 67	26.96 66	7.33 66	23.81 65	13.05 65	25.99 63	25.17 67		22.45 67	36.25 67

High-resolution multi-view benchmark