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

AttMVS		96.22 5	91.52 16	99.36 1	99.59 1	99.50 3	87.68 15	95.36 22	98.98 2

test_1124		97.51 1	96.38 1	98.27 8	99.58 2	96.12 27	95.79 1	96.96 11	99.10 1

MVSNet		96.11 6	92.12 13	98.77 2	99.42 3	99.78 2	86.41 18	97.84 4	97.10 22

CasMVSNet(SR_B)		93.45 24	85.74 39	98.59 4	99.20 4	99.92 1	79.66 34	91.83 38	96.66 29

Pnet-new-		96.49 4	93.65 6	98.39 7	99.16 5	97.29 21	90.27 7	97.02 10	98.71 5

test_1126		95.78 9	92.58 10	97.92 11	98.89 6	98.35 14	89.40 9	95.76 19	96.51 31

test_1120		97.11 2	96.16 2	97.74 13	98.72 7	96.98 22	94.90 2	97.43 5	97.50 18

Pnet-blend++		94.52 17	94.40 4	94.61 30	98.03 8	98.91 9	91.73 4	97.06 8	86.88 54

Pnet-blend		94.52 17	94.40 4	94.61 30	98.03 8	98.91 9	91.73 4	97.06 8	86.88 54

3Dnovator	98.27 2	96.99 3	94.47 3	98.67 3	97.83 10	99.24 6	92.01 3	96.93 12	98.94 4

vp_mvsnet		86.36 49	70.17 60	97.15 19	97.73 11	96.84 24	42.33 78	98.02 3	96.86 26

mvs_zhu_1030		95.16 15	90.25 19	98.44 6	97.68 12	99.49 4	83.22 25	97.28 6	98.16 10

tm-dncc		93.68 22	87.92 27	97.51 16	97.62 13	96.79 25	83.04 28	92.81 33	98.14 12

tmmvs		95.25 13	91.86 15	97.52 15	97.62 13	96.79 25	88.50 11	95.23 23	98.14 12

OpenMVS_ROB	95.38 14	94.28 21	91.29 18	96.27 23	97.52 15	94.78 36	87.11 16	95.47 20	96.52 30

P-MVSNet		94.83 16	89.24 23	98.55 5	97.36 16	99.33 5	83.98 22	94.50 24	98.97 3

LTVRE_ROB	98.40 1	95.34 11	92.18 12	97.44 17	97.15 17	98.03 17	91.46 6	92.90 32	97.14 21
Andreas Kuhn, Heiko HirschmÃ¼ller, Daniel Scharstein, Helmut Mayer: A TV Prior for High-Quality Scalable Multi-View Stereo Reconstruction. International Journal of Computer Vision 2016
HY-MVS	95.94 13	96.04 7	93.13 7	97.98 10	97.11 18	98.27 15	88.06 13	98.20 2	98.56 7

3Dnovator+	97.89 3	95.95 8	92.88 8	97.99 9	96.94 19	98.52 13	89.88 8	95.89 16	98.50 8

TAPA-MVS	96.21 11	95.20 14	92.05 14	97.30 18	96.93 20	98.17 16	87.93 14	96.17 14	96.81 27
Andrea Romanoni, Matteo Matteucci: TAPA-MVS: Textureless-Aware PAtchMatch Multi-View Stereo. ICCV 2019
TAPA-MVS(SR)		95.29 12	92.68 9	97.03 20	96.70 21	96.89 23	88.16 12	97.21 7	97.50 18

OpenMVS	96.65 7	95.59 10	92.21 11	97.84 12	96.27 22	98.97 8	88.62 10	95.80 18	98.29 9

MVSNet_plusplus		75.31 57	60.60 74	85.11 53	95.60 23	62.72 61	30.90 82	90.30 47	97.02 24

LPCS		94.29 20	89.33 22	97.59 14	95.31 24	98.86 12	86.05 19	92.61 34	98.60 6

DeepPCF-MVS	96.93 5	93.60 23	89.75 21	96.17 26	95.24 25	95.47 32	85.12 20	94.38 26	97.79 15

DeepC-MVS	97.60 4	94.32 19	91.48 17	96.21 24	95.12 26	95.94 30	86.92 17	96.03 15	97.57 17
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
CasMVSNet(base)		91.29 30	83.46 43	96.51 21	94.80 27	99.23 7	76.50 45	90.42 45	95.49 37

DeepC-MVS_fast	96.85 6	93.36 25	89.11 24	96.20 25	94.70 28	95.90 31	84.83 21	93.38 28	98.01 14
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
PVSNet_LR		91.95 27	87.47 29	94.93 29	94.42 29	95.25 33	79.53 35	95.42 21	95.13 38

PCF-MVS	92.86 18	89.27 41	82.32 44	93.90 33	94.10 30	92.56 44	78.41 39	86.23 54	95.05 39
Andreas Kuhn, Shan Lin, Oliver Erdler: Plane Completion and Filtering for Multi-View Stereo Reconstruction. GCPR 2019
IB-MVS	91.63 19	90.18 37	87.85 28	91.73 46	94.00 31	92.76 43	81.28 31	94.41 25	88.44 51
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
Pnet-eth		79.24 53	78.85 53	79.51 56	93.83 32	64.76 58	68.57 55	89.13 52	79.93 61

GSE		92.65 26	86.87 31	96.50 22	93.80 33	97.95 20	83.16 26	90.59 44	97.74 16

PMVS	91.26 20	75.40 56	44.56 82	95.95 27	93.75 34	95.97 29	55.22 67	33.90 82	98.15 11
Y. Furukawa, J. Ponce: Accurate, dense, and robust multiview stereopsis. PAMI (2010)
test_1205		87.14 48	81.84 46	90.68 47	93.01 35	96.01 28	77.62 44	86.06 56	83.02 59

CasMVSNet(SR_A)		91.32 29	84.85 40	95.63 28	93.01 35	98.91 9	78.76 37	90.94 42	94.97 40

BP-MVSNet		87.27 46	80.28 51	91.92 45	92.62 37	89.82 50	71.21 50	89.35 50	93.32 42
Christian Sormann, Patrick KnÃ¶belreiter, Andreas Kuhn, Mattia Rossi, Thomas Pock, Friedrich Fraundorfer: BP-MVSNet: Belief-Propagation-Layers for Multi-View-Stereo. 3DV 2020
Pnet_fast		89.94 38	85.97 36	92.58 41	92.52 38	94.98 34	78.33 40	93.61 27	90.25 47

COLMAP(SR)		91.87 28	87.99 26	94.46 32	92.04 39	94.57 37	83.60 24	92.38 35	96.76 28

R-MVSNet		90.65 33	88.39 25	92.16 43	92.01 40	91.05 48	80.93 33	95.85 17	93.43 41

PVSNet	93.40 17	87.96 44	87.30 30	88.40 52	91.26 41	84.79 53	83.16 26	91.43 40	89.14 50

ACMH+	96.62 9	90.69 32	86.75 32	93.32 37	91.07 42	92.46 45	81.20 32	92.31 36	96.43 33

F/T MVSNet+Gipuma		79.48 52	82.20 45	77.68 57	90.98 43	55.72 65	71.27 49	93.13 29	86.32 56

UnsupFinetunedMVSNet					90.98 43

MVS_test_1		75.15 58	66.26 63	81.08 55	90.91 45	64.69 59	42.15 79	90.37 46	87.64 53

ACMH	96.65 7	90.42 35	85.83 38	93.48 36	90.64 46	92.90 41	78.75 38	92.91 31	96.89 25
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
COLMAP(base)		90.88 31	86.46 33	93.83 34	90.50 47	93.89 38	81.31 30	91.60 39	97.10 22

MVSNet + Gipuma		78.76 54	81.53 48	76.91 58	90.19 48	55.24 66	71.20 51	91.86 37	85.28 57

PVSNet_0	89.98 21	87.18 47	84.23 41	89.14 51	89.98 49	85.88 52	78.21 43	90.25 48	91.56 45

PLC	94.65 16	90.20 36	86.30 34	92.80 39	89.29 50	92.80 42	81.48 29	91.12 41	96.30 34
Jie Liao, Yanping Fu, Qingan Yan, Chunxia xiao: Pyramid Multi-View Stereo with Local Consistency. Pacific Graphics 2019
CIDER		89.94 38	90.13 20	89.81 48	89.08 51	90.51 49	83.89 23	96.37 13	89.84 49
Qingshan Xu and Wenbing Tao: Learning Inverse Depth Regression for Multi-View Stereo with Correlation Cost Volume. AAAI 2020
ACMM	96.08 12	89.28 40	84.14 42	92.70 40	88.68 52	93.12 40	78.33 40	89.95 49	96.30 34
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
COLMAP_ROB	96.50 10	90.54 34	85.96 37	93.59 35	88.41 53	94.98 34	78.87 36	93.05 30	97.37 20
Johannes L. SchÃ¶nberger, Enliang Zheng, Marc Pollefeys, Jan-Michael Frahm: Pixelwise View Selection for Unstructured Multi-View Stereo. ECCV 2016
unMVSmet		75.98 55	76.68 54	75.52 59	88.21 54	63.35 60	67.51 57	85.84 57	74.99 62

ANet-0.75		82.50 51	67.03 62	92.82 38	87.52 55	97.97 18	56.76 63	77.30 65	92.96 43

ANet		88.09 42	81.47 49	92.50 42	87.52 55	97.97 18	72.29 48	90.64 43	92.00 44

ACMP	95.32 15	87.91 45	81.54 47	92.16 43	86.95 57	93.34 39	75.21 46	87.87 53	96.18 36
Qingshan Xu and Wenbing Tao: Planar Prior Assisted PatchMatch Multi-View Stereo. AAAI 2020
MVSCRF		87.97 43	86.10 35	89.22 49	85.61 58	92.04 46	73.48 47	98.71 1	89.99 48

A-TVSNet + Gipuma		85.81 50	80.78 50	89.16 50	85.45 59	91.39 47	78.22 42	83.34 59	90.64 46

Snet		70.24 62	76.61 55	66.00 64	81.52 60	47.72 73	70.02 52	83.20 60	68.76 69

confMetMVS		70.86 61	79.56 52	65.06 65	79.46 61	52.64 71	69.86 53	89.25 51	63.08 73

unsupervisedMVS_cas		74.92 59	64.56 66	81.83 54	75.67 62	86.23 51	53.74 71	75.37 66	83.59 58

Cas-MVS_preliminary		70.90 60	65.38 64	74.58 60	74.23 63	53.08 70	52.36 72	78.40 63	96.45 32

test_mvsss		64.12 65	49.98 78	73.54 61	73.37 64	73.88 55	34.02 80	65.95 77	73.36 63

example		53.61 78	46.43 81	58.40 70	72.19 65	61.90 63	45.83 77	47.03 81	41.10 82

RMVSNet		65.81 64	74.09 56	60.29 69	71.48 66	53.56 67	68.62 54	79.56 61	55.82 77

MVSNet_++		61.90 69	49.37 79	70.26 62	71.25 67	51.13 72	13.23 84	85.51 58	88.40 52

hgnet		62.55 67	64.18 68	61.47 67	70.70 68	65.58 56	56.76 63	71.59 72	48.14 79

DPSNet		62.55 67	64.18 68	61.47 67	70.70 68	65.58 56	56.76 63	71.59 72	48.14 79

CPR_FA		69.65 63	71.00 58	68.75 63	66.27 70	59.56 64	64.16 58	77.85 64	80.42 60

CCVNet		63.49 66	64.42 67	62.86 66	63.31 71	53.21 68	57.30 61	71.54 74	72.07 64

metmvs_fine		60.40 71	73.46 57	51.70 77	59.47 72	43.63 75	60.85 59	86.07 55	52.00 78

unMVSv1		49.24 82	47.80 80	50.21 80	52.05 73	53.15 69	47.48 76	48.13 80	45.41 81

CMPMVS	75.91 23	27.01 83	1.74 84	43.86 83	51.27 74	80.30 54	3.48 85	0.00 84	0.00 85
M. Jancosek, T. Pajdla: Multi-View Reconstruction Preserving Weakly-Supported Surfaces. CVPR 2011
TVSNet		58.57 72	65.04 65	54.25 73	50.35 75	43.60 76	56.78 62	73.30 67	68.81 68

firsttry		53.54 79	54.29 77	53.04 74	49.77 76	45.66 74	49.70 74	58.88 79	63.70 72

A1Net		58.27 73	69.56 61	50.74 79	48.60 77	35.72 83	60.70 60	78.42 62	67.89 70

QQQNet		58.00 74	63.43 70	54.38 72	48.14 78	42.94 79	53.98 68	72.88 69	72.07 64

test3		56.14 77	63.17 73	51.45 78	47.91 79	42.37 80	55.62 66	70.72 75	64.07 71

SGNet		53.33 80	58.49 75	49.90 81	47.35 80	42.30 81	50.65 73	66.33 76	60.04 74

PSD-MVSNet		51.39 81	55.47 76	48.66 82	47.23 81	41.20 82	47.60 75	63.35 78	57.55 76

MVE	83.40 22	61.54 70	70.38 59	55.65 71	44.73 82	62.23 62	67.58 56	73.18 68	60.00 75
Simon Fuhrmann, Fabian Langguth, Michael Goesele: MVE - A Multi-View Reconstruction Environment. EUROGRAPHICS Workshops on Graphics and Cultural Heritage (2014)
SVVNet		57.08 75	63.37 71	52.89 75	44.50 83	43.08 77	53.86 69	72.88 69	71.08 66

ternet		57.08 75	63.37 71	52.89 75	44.50 83	43.08 77	53.86 69	72.88 69	71.08 66

test_robustmvs				14.79 84	24.07 85	13.33 84	22.91 83		6.99 83

FADENet		2.66 84	2.75 83	2.60 85	4.76 86	1.96 85	3.23 86	2.27 83	1.09 84

dnet		0.00 85	0.00 85	0.00 86	0.00 87	0.00 86	0.00 87	0.00 84	0.00 85

test_MVS							32.87 81

Low-resolution many-view benchmark