Low-res many-view results

This table lists the benchmark results for the low-res many-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	low-res many-view	indoor	outdoor	lakeside	sand box	storage room	storage room 2	tunnel

DeepPCF-MVS	93.97 1	87.63 1	82.10 1	91.31 1	91.80 1	91.34 1	77.64 1	86.56 1	90.79 1

DeepC-MVS_fast	93.89 2	86.59 2	80.39 2	90.73 2	91.00 3	90.77 2	75.90 3	84.89 3	90.41 2
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
DeepC-MVS	93.07 3	85.88 3	79.73 4	89.98 3	89.82 6	90.09 5	74.93 5	84.53 4	90.03 3
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
TAPA-MVS	90.10 7	84.80 4	80.13 3	87.91 6	88.22 8	88.54 9	75.35 4	84.91 2	86.96 14
Andrea Romanoni, Matteo Matteucci: TAPA-MVS: Textureless-Aware PAtchMatch Multi-View Stereo. ICCV 2019
PCF-MVS	89.48 11	84.51 5	76.96 10	89.54 4	91.30 2	88.76 7	73.35 8	80.56 13	88.57 4
Andreas Kuhn, Shan Lin, Oliver Erdler: Plane Completion and Filtering for Multi-View Stereo Reconstruction. GCPR 2019
TAPA-MVS(SR)		83.88 6	78.08 7	87.74 8	88.99 7	86.49 16	73.17 10	82.99 6	87.76 10

COLMAP(base)		83.59 7	77.50 9	87.65 9	86.74 13	87.68 11	73.30 9	81.70 10	88.52 5

PLC	91.00 6	83.54 8	78.38 5	86.98 13	85.38 17	87.33 13	73.45 6	83.31 5	88.23 9
Jie Liao, Yanping Fu, Qingan Yan, Chunxia xiao: Pyramid Multi-View Stereo with Local Consistency. Pacific Graphics 2019
GSE		83.39 9	77.82 8	87.10 12	85.70 16	87.27 14	73.41 7	82.23 8	88.34 8

COLMAP(SR)		83.28 10	76.24 13	87.97 5	87.61 9	87.90 10	72.76 11	79.71 15	88.39 7

ACMH+	87.92 14	82.46 11	76.63 12	86.35 16	86.95 11	84.52 21	71.68 12	81.58 11	87.56 12

ACMP	89.59 10	82.31 12	75.00 15	87.18 11	84.49 22	88.57 8	68.43 19	81.58 11	88.49 6
Qingshan Xu and Wenbing Tao: Planar Prior Assisted PatchMatch Multi-View Stereo. AAAI 2020
LTVRE_ROB	88.41 13	82.23 13	78.11 6	84.97 19	86.24 14	82.83 27	76.93 2	79.30 16	85.84 16
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
ACMM	89.79 8	82.20 14	75.34 14	86.76 14	84.94 19	87.68 11	68.85 18	81.84 9	87.67 11
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
3Dnovator	91.36 5	82.19 15	74.09 17	87.59 10	89.83 5	90.10 4	71.54 13	76.63 20	82.84 21

3Dnovator+	91.43 4	81.91 16	73.11 20	87.78 7	90.06 4	90.17 3	70.40 16	75.82 23	83.12 18

COLMAP_ROB	87.81 15	81.57 17	74.74 16	86.13 17	83.81 23	87.08 15	69.46 17	80.01 14	87.49 13
Johannes L. Schönberger, Enliang Zheng, Marc Pollefeys, Jan-Michael Frahm: Pixelwise View Selection for Unstructured Multi-View Stereo. ECCV 2016
IB-MVS	87.33 17	81.13 18	76.89 11	83.96 21	86.75 12	84.70 20	71.43 14	82.36 7	80.43 24
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
ACMH	87.59 16	80.55 19	73.69 18	85.11 18	86.17 15	82.81 28	68.19 20	79.20 17	86.36 15
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
OpenMVS	89.19 12	80.38 20	71.40 22	86.37 15	87.42 10	89.29 6	67.70 21	75.10 24	82.40 23

HY-MVS	89.66 9	79.55 21	71.86 21	84.69 20	84.86 20	86.32 17	66.45 22	77.26 19	82.88 20

LPCS		78.54 22	73.62 19	81.82 24	79.59 29	83.19 25	71.25 15	75.99 22	82.69 22

A-TVSNet + Gipuma		77.14 23	67.96 25	83.26 22	80.53 26	86.26 18	65.25 23	70.67 25	82.99 19

BP-MVSNet		76.76 24	68.58 24	82.22 23	84.51 21	77.66 38	60.98 28	76.18 21	84.48 17
Christian Sormann, Patrick Knöbelreiter, Andreas Kuhn, Mattia Rossi, Thomas Pock, Friedrich Fraundorfer: BP-MVSNet: Belief-Propagation-Layers for Multi-View-Stereo. 3DV 2020
PVSNet_0	82.17 19	74.27 25	64.50 27	80.78 25	81.77 25	80.64 31	61.11 26	67.89 28	79.92 25

PVSNet	86.66 18	73.29 26	65.21 26	78.67 29	80.40 27	77.98 36	64.04 25	66.38 31	77.62 26

OpenMVS_ROB	81.14 20	73.28 27	63.11 29	80.06 27	82.70 24	81.62 30	61.11 26	65.12 33	75.87 27

CIDER		72.92 28	62.80 30	79.66 28	80.16 28	83.22 24	60.69 29	64.90 34	75.61 29
Qingshan Xu and Wenbing Tao: Learning Inverse Depth Regression for Multi-View Stereo with Correlation Cost Volume. AAAI 2020
test_1126		71.43 29	57.57 35	80.67 26	85.16 18	84.45 22	55.16 33	59.98 38	72.40 31

test_1205		70.75 30	62.35 31	76.35 32	78.70 30	85.03 19	55.63 32	69.08 26	65.34 42

R-MVSNet		70.54 31	63.98 28	74.91 34	73.85 36	77.87 37	59.85 30	68.11 27	73.01 30

P-MVSNet		70.42 32	71.20 23	69.90 39	66.73 42	72.02 40	64.84 24	77.55 18	70.95 33

ANet-0.75		67.51 33	58.66 34	73.41 36	70.12 38	79.37 32	50.72 35	66.61 30	70.74 34

AttMVS		65.89 34	60.14 32	69.73 40	64.68 45	75.04 39	58.80 31	61.47 37	69.47 36

mvs_zhu_1030		64.43 35	51.10 38	73.32 37	71.88 37	78.97 35	42.87 38	59.32 39	69.10 37

ANet		63.85 36	52.65 37	71.32 38	70.12 38	79.37 32	48.53 37	56.77 42	64.48 43

Pnet-new-		63.42 37	42.13 53	77.61 30	77.71 32	79.30 34	41.06 41	43.20 62	75.82 28

CasMVSNet(SR_A)		63.08 38	45.33 45	74.92 33	74.36 33	83.73 23	37.04 52	53.62 45	66.66 40

CasMVSNet(base)		61.88 39	43.72 48	73.99 35	74.03 35	81.95 29	34.99 57	52.45 47	65.99 41

Pnet_fast		59.94 40	35.26 68	76.39 31	74.15 34	82.99 26	25.02 68	45.51 56	72.03 32

CPR_FA		58.26 41	56.35 36	59.54 49	56.43 53	55.34 52	50.61 36	62.09 36	66.83 39

unsupervisedMVS_cas		57.32 42	44.73 47	65.71 42	63.76 46	71.21 41	36.03 54	53.44 46	62.15 46

MVSNet		56.22 43	46.49 42	62.70 45	57.79 50	67.16 47	38.93 45	54.05 43	63.16 45

A1Net		54.85 44	59.98 33	51.43 52	47.46 62	42.97 67	52.72 34	67.24 29	63.86 44

MVSNet_plusplus		54.10 45	36.12 66	66.09 41	78.36 31	49.55 56	20.40 73	51.84 48	70.36 35

Pnet-blend		53.68 46	37.04 61	64.76 43	68.91 40	70.02 43	24.13 69	49.96 52	55.36 50

Pnet-blend++		53.68 46	37.04 61	64.76 43	68.91 40	70.02 43	24.13 69	49.96 52	55.36 50

MVSCRF		51.99 48	37.64 58	61.55 47	58.88 47	68.35 45	34.99 57	40.28 65	57.43 49

MVSNet_++		50.85 49	37.40 59	59.81 48	65.34 44	46.28 58	9.19 76	65.60 32	67.81 38

test_1124		50.28 50	31.72 70	62.65 46	58.29 49	70.82 42	27.23 66	36.21 70	58.84 47

Snet		49.32 51	36.02 67	58.18 50	66.02 43	50.24 54	27.93 65	44.12 57	58.27 48

CasMVSNet(SR_B)		48.99 52	45.75 43	51.15 53	49.07 60	62.78 48	37.57 46	53.93 44	41.61 61

test3		45.71 53	44.95 46	46.21 60	42.67 66	44.26 61	39.04 44	50.87 49	51.69 52

Pnet-eth		45.62 54	49.15 40	43.27 67	53.31 55	29.96 77	35.03 56	63.26 35	46.53 58

unMVSv1		45.02 55	43.12 50	46.29 59	47.83 61	50.14 55	42.67 39	43.57 58	40.92 62

RMVSNet		44.67 56	50.67 39	40.67 69	50.79 56	41.21 69	42.32 40	59.02 40	30.02 70

MVE	50.73 23	44.67 56	45.57 44	44.07 65	35.90 73	52.30 53	40.89 42	50.26 50	44.00 60
Simon Fuhrmann, Fabian Langguth, Michael Goesele: MVE - A Multi-View Reconstruction Environment. EUROGRAPHICS Workshops on Graphics and Cultural Heritage (2014)
TVSNet		44.65 58	42.86 52	45.84 61	43.41 65	43.54 65	35.52 55	50.20 51	50.56 53

hgnet		44.61 59	38.48 56	48.69 54	57.33 51	60.08 49	37.17 50	39.80 67	28.65 72

DPSNet		44.61 59	38.48 56	48.69 54	57.33 51	60.08 49	37.17 50	39.80 67	28.65 72

SGNet		44.49 61	43.48 49	45.16 63	42.20 69	43.60 64	37.53 47	49.43 54	49.69 54

PSD-MVSNet		44.35 62	43.10 51	45.19 62	42.64 67	43.61 63	37.46 48	48.74 55	49.32 55

MVSNet + Gipuma		44.11 63	38.76 55	47.67 57	50.05 59	45.26 60	34.93 59	42.60 63	47.71 56

F/T MVSNet+Gipuma		43.59 64	37.31 60	47.78 56	50.12 57	45.64 59	34.25 60	40.37 64	47.60 57

PMVS	53.92 22	42.74 65	26.50 72	53.57 51	54.03 54	68.05 46	36.35 53	16.65 76	38.64 63
Y. Furukawa, J. Ponce: Accurate, dense, and robust multiview stereopsis. PAMI (2010)
firsttry		41.11 66	36.23 63	44.36 64	43.82 64	43.77 62	32.28 61	40.18 66	45.51 59

QQQNet		40.79 67	40.37 54	41.07 68	42.37 68	43.46 66	37.29 49	43.44 59	37.38 66

example		40.49 68	30.54 71	47.11 58	58.75 48	58.88 51	29.97 62	31.12 71	23.71 76

metmvs_fine		39.41 69	48.46 41	33.38 73	35.53 74	31.67 76	39.28 43	57.64 41	32.95 69

CCVNet		38.86 70	31.75 69	43.59 66	44.13 63	49.27 57	25.48 67	38.01 69	37.38 66

ternet		37.41 71	36.23 63	38.19 70	37.28 70	38.77 70	29.02 63	43.44 59	38.53 64

SVVNet		37.41 71	36.23 63	38.19 70	37.28 70	38.77 70	29.02 63	43.44 59	38.53 64

unMVSmet		32.09 73	25.89 73	36.22 72	36.96 72	42.02 68	23.79 71	27.99 72	29.68 71

test_1120		27.94 74	22.22 75	31.76 74	20.92 77	37.22 73	18.54 75	25.90 74	37.14 68

confMetMVS		26.62 75	23.72 74	28.55 76	28.79 75	33.00 75	21.33 72	26.10 73	23.86 75

Cas-MVS_preliminary		25.33 76	19.98 76	28.90 75	24.65 76	34.26 74	18.62 74	21.35 75	27.78 74

CMPMVS	62.92 21	8.44 77	0.22 78	13.92 77	4.26 78	37.50 72	0.45 78	0.00 78	0.00 78
M. Jancosek, T. Pajdla: Multi-View Reconstruction Preserving Weakly-Supported Surfaces. CVPR 2011
FADENet		1.48 78	1.79 77	1.27 78	2.11 79	1.06 78	2.23 77	1.34 77	0.65 77

dnet		0.00 79	0.00 79	0.00 79	0.00 80	0.00 79	0.00 79	0.00 78	0.00 78

UnsupFinetunedMVSNet					50.12 57

Low-resolution many-view benchmark