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.

PHI-MVS		86.43 4	86.17 5	87.24 5	90.88 12	70.96 9	92.27 4	94.07 1	72.45 15	85.22 2	91.90 10	69.47 5	96.42 5	83.28 3	95.94 1	94.35 4

DeepPCF-MVS	80.84 1	88.10 1	88.56 1	86.73 7	92.24 6	69.03 12	89.57 15	93.39 5	77.53 3	89.79 1	94.12 2	78.98 1	96.58 4	85.66 1	95.72 2	94.58 3

DeepC-MVS	79.81 2	87.08 2	86.88 2	87.69 3	91.16 9	72.32 6	90.31 11	93.94 2	77.12 5	82.82 5	94.23 1	72.13 3	97.09 1	84.83 2	95.37 3	93.65 7
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	79.65 3	86.91 3	86.62 3	87.76 2	93.52 2	72.37 5	91.26 6	93.04 7	76.62 7	84.22 3	93.36 5	71.44 4	96.76 2	80.82 5	95.33 4	94.16 5
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
CSCG		86.41 5	86.19 4	87.07 6	92.91 4	72.48 4	90.81 8	93.56 4	73.95 14	83.16 4	91.07 14	75.94 2	95.19 8	79.94 6	94.38 5	93.55 8

3Dnovator+	77.84 4	85.48 7	84.47 7	88.51 1	91.08 10	73.49 1	93.18 1	93.78 3	80.79 1	76.66 17	93.37 4	60.40 20	96.75 3	77.20 8	93.73 6	95.29 1

3Dnovator	76.31 5	83.38 10	82.31 11	86.59 8	87.94 30	72.94 2	90.64 9	92.14 9	77.21 4	75.47 19	92.83 6	58.56 22	94.72 13	73.24 12	92.71 7	92.13 15

MAR-MVS		81.84 15	80.70 15	85.27 9	91.32 8	71.53 7	89.82 14	90.92 15	69.77 23	78.50 12	86.21 35	62.36 16	94.52 17	65.36 17	92.05 8	89.77 20
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
IS-MVSNet		83.15 11	82.81 10	84.18 16	89.94 17	63.30 32	91.59 5	88.46 28	79.04 2	79.49 9	92.16 8	65.10 11	94.28 19	67.71 16	91.86 9	94.95 2

Vis-MVSNet (Re-imp)		78.36 23	78.45 21	78.07 39	88.64 24	51.78 49	86.70 28	79.63 46	74.14 13	75.11 20	90.83 15	61.29 18	89.75 36	58.10 30	91.60 10	92.69 11

MG-MVS		83.41 8	83.45 8	83.28 23	92.74 5	62.28 34	88.17 24	89.50 20	75.22 9	81.49 6	92.74 7	66.75 7	95.11 10	72.85 13	91.58 11	92.45 13

API-MVS		81.99 14	81.23 14	84.26 15	90.94 11	70.18 10	91.10 7	89.32 22	71.51 16	78.66 11	88.28 26	65.26 10	95.10 11	64.74 19	91.23 12	87.51 31

OpenMVS	72.83 10	79.77 21	78.33 22	84.09 17	85.17 35	69.91 11	90.57 10	90.97 14	66.70 27	72.17 24	91.91 9	54.70 25	93.96 20	61.81 23	90.95 13	88.41 26

ACMMP		85.89 6	85.39 6	87.38 4	93.59 1	72.63 3	92.74 2	93.18 6	76.78 6	80.73 7	93.82 3	64.33 12	96.29 6	82.67 4	90.69 14	93.23 9

PAPR		80.80 17	79.93 17	83.38 20	90.02 16	63.60 31	84.81 33	90.50 18	67.73 26	74.57 22	87.61 29	66.55 8	93.87 22	69.11 15	88.51 15	92.67 12

AdaColmap		80.58 19	79.42 19	84.06 18	93.09 3	68.91 14	89.36 16	88.97 26	69.27 24	75.70 18	89.69 19	57.20 24	95.77 7	63.06 21	88.41 16	87.50 33

PCF-MVS	73.52 7	80.38 20	78.84 20	85.01 10	87.71 31	68.99 13	83.65 35	91.46 13	63.00 36	77.77 14	90.28 18	66.10 9	95.09 12	61.40 24	88.22 17	90.94 17
Andreas Kuhn, Shan Lin, Oliver Erdler: Plane Completion and Filtering for Multi-View Stereo Reconstruction. GCPR 2019
CLD-MVS		82.31 13	81.65 13	84.29 14	88.47 28	67.73 18	85.81 30	92.35 8	75.78 8	78.33 13	86.58 34	64.01 14	94.35 18	76.05 9	87.48 18	90.79 18
Zhaoxin Li, Wangmeng Zuo, Zhaoqi Wang, Lei Zhang: Confidence-based Large-scale Dense Multi-view Stereo. IEEE Transaction on Image Processing, 2020
EPP-MVSNet		83.40 9	83.02 9	84.57 12	90.13 15	64.47 29	92.32 3	90.73 17	74.45 12	79.35 10	91.10 12	69.05 6	95.12 9	72.78 14	87.22 19	94.13 6

TAPA-MVS	73.13 9	79.15 22	77.94 23	82.79 25	89.59 18	62.99 33	88.16 25	91.51 11	65.77 29	77.14 16	91.09 13	60.91 19	93.21 26	50.26 43	87.05 20	92.17 14
Andrea Romanoni, Matteo Matteucci: TAPA-MVS: Textureless-Aware PAtchMatch Multi-View Stereo. ICCV 2019
PAPM		77.68 26	76.40 25	81.51 30	87.29 32	61.85 36	83.78 34	89.59 19	64.74 32	71.23 26	88.70 23	62.59 15	93.66 23	52.66 41	87.03 21	89.01 22

LS3D		76.95 27	74.82 28	83.37 21	90.45 13	67.36 21	89.15 17	86.94 30	61.87 41	69.52 32	90.61 16	51.71 30	94.53 16	46.38 47	86.71 22	88.21 27

OMC-MVS		82.69 12	81.97 12	84.85 11	88.75 22	67.42 19	87.98 26	90.87 16	74.92 10	79.72 8	91.65 11	62.19 17	93.96 20	75.26 10	86.42 23	93.16 10

PVSNet	64.34 17	72.08 40	70.87 38	75.69 45	86.21 34	56.44 46	74.37 48	80.73 44	62.06 40	70.17 28	82.23 43	42.86 42	83.31 46	54.77 36	84.45 24	87.32 34

CNLPA		78.08 24	76.79 24	81.97 28	90.40 14	71.07 8	87.59 27	84.55 34	66.03 28	72.38 23	89.64 21	57.56 23	86.04 42	59.61 26	83.35 25	88.79 24

ACMP	74.13 6	81.51 16	80.57 16	84.36 13	89.42 19	68.69 15	89.97 13	91.50 12	74.46 11	75.04 21	90.41 17	53.82 26	94.54 15	77.56 7	82.91 26	89.86 19
Qingshan Xu and Wenbing Tao: Planar Prior Assisted PatchMatch Multi-View Stereo. AAAI 2020
ACMM	73.20 8	80.78 18	79.84 18	83.58 19	89.31 20	68.37 16	89.99 12	91.60 10	70.28 18	77.25 15	89.66 20	53.37 28	93.53 25	74.24 11	82.85 27	88.85 23
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
PMMVS		69.34 44	68.67 41	71.35 48	75.67 48	62.03 35	75.17 47	73.46 50	50.00 50	68.68 35	79.05 48	52.07 29	78.13 47	61.16 25	82.77 28	73.90 50

PLC	70.83 11	78.05 25	76.37 26	83.08 24	91.88 7	67.80 17	88.19 23	89.46 21	64.33 33	69.87 30	88.38 25	53.66 27	93.58 24	58.86 27	82.73 29	87.86 28
Jie Liao, Yanping Fu, Qingan Yan, Chunxia xiao: Pyramid Multi-View Stereo with Local Consistency. Pacific Graphics 2019
CMPMVS	51.72 20	70.19 43	68.16 44	76.28 43	73.15 50	57.55 43	79.47 42	83.92 37	48.02 51	56.48 49	84.81 39	43.13 41	86.42 41	62.67 22	81.81 30	84.89 39
M. Jancosek, T. Pajdla: Multi-View Reconstruction Preserving Weakly-Supported Surfaces. CVPR 2011
PatchMatch-RL		72.38 39	70.90 37	76.80 42	88.60 25	67.38 20	79.53 41	76.17 49	62.75 39	69.36 34	82.00 44	45.51 37	84.89 44	53.62 38	80.58 31	78.12 49

LTVRE_ROB	69.57 12	76.25 29	74.54 29	81.41 31	88.60 25	64.38 30	79.24 43	89.12 23	70.76 17	69.79 31	87.86 27	49.09 32	93.20 27	56.21 34	80.16 32	86.65 38
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
F-COLMAP		76.38 28	74.33 30	82.50 26	89.28 21	66.95 22	88.41 21	89.03 24	64.05 35	66.83 40	88.61 24	46.78 34	92.89 29	57.48 31	78.55 33	87.67 29

ACMH	67.68 15	75.89 32	73.93 32	81.77 29	88.71 23	66.61 23	88.62 20	89.01 25	69.81 22	66.78 41	86.70 33	41.95 43	91.51 32	55.64 35	78.14 34	87.17 35
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
MSDG		73.36 34	70.99 36	80.49 33	84.51 37	65.80 26	80.71 38	86.13 32	65.70 30	65.46 42	83.74 40	44.60 39	90.91 35	51.13 42	76.89 35	84.74 40

PatchmatchNet		73.12 36	71.33 35	78.49 37	83.18 38	60.85 38	79.63 40	78.57 47	64.13 34	71.73 25	79.81 47	51.20 31	85.97 43	57.40 32	76.36 36	88.66 25
Fangjinhua Wang, Silvano Galliani, Christoph Vogel, Pablo Speciale, Marc Pollefeys: PatchmatchNet: Learned Multi-View Patchmatch Stereo.
USDC		70.33 42	68.37 42	76.21 44	80.60 43	56.23 47	79.19 44	86.49 31	60.89 43	61.29 45	85.47 37	31.78 48	89.47 37	53.37 39	76.21 37	82.94 42

OpenMVS_ROB	64.09 18	70.56 41	68.19 43	77.65 40	80.26 44	59.41 41	85.01 31	82.96 40	58.76 47	65.43 43	82.33 42	37.63 46	91.23 34	45.34 49	76.03 38	82.32 43

ACMH+	68.96 13	76.01 30	74.01 31	82.03 27	88.60 25	65.31 28	88.86 18	87.55 29	70.25 19	67.75 37	87.47 30	41.27 44	93.19 28	58.37 29	75.94 39	87.60 30

IB-MVS	68.01 14	75.85 33	73.36 33	83.31 22	84.76 36	66.03 24	83.38 36	85.06 33	70.21 20	69.40 33	81.05 45	45.76 36	94.66 14	65.10 18	75.49 40	89.25 21

EPMVS		69.02 45	68.16 44	71.59 47	79.61 45	49.80 52	77.40 45	66.93 54	62.82 38	70.01 29	79.05 48	45.79 35	77.86 48	56.58 33	75.26 41	87.13 36

COLMAP_ROB	66.92 16	73.01 37	70.41 40	80.81 32	87.13 33	65.63 27	88.30 22	84.19 36	62.96 37	63.80 44	87.69 28	38.04 45	92.56 31	46.66 46	74.91 42	84.24 41
Johannes L. Schönberger, Enliang Zheng, Marc Pollefeys, Jan-Michael Frahm: Pixelwise View Selection for Unstructured Multi-View Stereo. ECCV 2016
ITE_SJBPF				78.22 38	81.77 42	60.57 39		83.30 38	69.25 25	67.54 38	87.20 31	36.33 47	87.28 40	54.34 37	74.62 43	86.80 37

test_0402		72.79 38	70.44 39	79.84 34	88.13 29	65.99 25	85.93 29	84.29 35	65.57 31	67.40 39	85.49 36	46.92 33	92.61 30	35.88 50	74.38 44	80.94 45

RPSCF		73.23 35	71.46 34	78.54 36	82.50 39	59.85 40	82.18 37	82.84 41	58.96 46	71.15 27	89.41 22	45.48 38	84.77 45	58.82 28	71.83 45	91.02 16

TinyColmap		67.30 47	64.81 46	74.76 46	81.92 41	56.68 45	80.29 39	81.49 42	60.33 44	56.27 50	83.22 41	24.77 50	87.66 39	45.52 48	69.47 46	79.95 46

TAMVS		76.00 31	74.94 27	79.20 35	82.02 40	58.65 42	88.83 19	88.75 27	70.11 21	67.87 36	86.96 32	58.69 21	91.44 33	64.20 20	66.99 47	87.51 31

LF4IMVS		64.02 48	62.19 48	69.50 49	70.90 51	53.29 48	76.13 46	77.18 48	52.65 49	58.59 46	80.98 46	23.55 51	76.52 50	53.06 40	66.66 48	78.68 48

FPMVS		53.68 50	51.64 50	59.81 52	65.08 52	51.03 51	69.48 50	69.58 52	41.46 52	40.67 51	72.32 50	16.46 53	70.00 55	24.24 54	65.42 49	58.40 53

TDRefinement		67.49 46	64.34 47	76.92 41	73.47 49	61.07 37	84.86 32	82.98 39	59.77 45	58.30 48	85.13 38	26.06 49	87.89 38	47.92 44	60.59 50	81.81 44

Gipuma		45.18 52	41.86 52	55.16 55	77.03 47	51.52 50	32.50 56	80.52 45	32.46 54	27.12 53	35.02 57	9.52 55	75.50 52	22.31 55	60.21 51	38.45 57
S. Galliani, K. Lasinger, K. Schindler: Massively Parallel Multiview Stereopsis by Surface Normal Diffusion. ICCV 2015
PMVS	37.38 21	44.16 53	40.28 53	55.82 54	40.82 58	42.54 54	65.12 52	63.99 55	34.43 53	24.48 54	57.12 53	3.92 56	76.17 51	17.10 57	55.52 52	48.75 54
Y. Furukawa, J. Ponce: Accurate, dense, and robust multiview stereopsis. PAMI (2010)
PVSNet_0	57.27 19	61.67 49	59.27 49	68.85 50	79.61 45	57.44 44	68.01 51	73.44 51	55.93 48	58.54 47	70.41 51	44.58 40	77.55 49	47.01 45	35.91 53	71.55 51

DeepMVS_CX				27.40 58	40.17 59	26.90 59		24.59 58	17.44 58	23.95 55	48.61 54	9.77 54	26.48 58	18.06 56	24.47 54	28.83 58

MVE	26.22 22	30.37 56	25.89 56	43.81 57	44.55 57	35.46 57	28.87 57	39.07 57	18.20 57	18.58 57	40.18 56	2.68 57	47.37 57	17.07 58	23.78 55	48.60 55
Simon Fuhrmann, Fabian Langguth, Michael Goesele: MVE - A Multi-View Reconstruction Environment. EUROGRAPHICS Workshops on Graphics and Cultural Heritage (2014)
wuykxyi23d		39.76 54	33.18 55	59.51 53	46.98 56	44.01 53	57.70 54	67.74 53	24.13 56	13.98 58	34.33 58	1.27 59	71.33 54	34.23 51	18.23 56	63.18 52

PNet_i23d		38.26 55	35.42 54	46.79 56	58.74 53	35.48 56	59.65 53	51.25 56	32.45 55	23.44 56	47.53 55	2.04 58	58.96 56	25.60 53	18.09 57	45.92 56

ANet_high		50.57 51	46.10 51	63.99 51	48.67 55	39.13 55	70.99 49	80.85 43	61.39 42	31.18 52	57.70 52	17.02 52	73.65 53	31.22 52	15.89 58	79.18 47

wuyk23d		16.82 57	15.94 57	19.46 59	58.74 53	31.45 58	39.22 55	3.74 59	6.84 59	6.04 59	2.70 59	1.27 59	24.29 59	10.54 59	14.40 59	2.63 59

Test By Simon												64.33 12

High-resolution multi-view benchmark