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

3Dnovator	80.37 7	69.73 1	65.51 3	72.55 10	82.32 4	65.67 15	57.53 3	73.48 1	69.67 13

Pnet-new-		62.28 15	64.63 4	60.72 23	68.45 20	45.71 43	56.04 4	73.22 2	67.99 15

3Dnovator+	83.92 2	69.39 3	62.05 5	74.29 7	81.76 5	64.89 17	52.69 7	71.42 3	76.22 1

OpenMVS	76.72 13	64.07 11	59.05 6	67.42 15	75.57 14	62.58 19	47.98 10	70.11 4	64.12 19

AttMVS		64.84 10	57.49 8	69.74 13	88.51 2	50.94 34	45.77 15	69.21 5	69.78 12

DeepC-MVS	82.31 4	65.89 7	57.22 9	71.66 11	74.01 15	70.18 7	47.73 11	66.72 6	70.79 10
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	80.27 8	65.96 6	54.23 14	73.79 8	75.88 12	71.59 6	45.37 17	63.08 7	73.89 5
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
MVSNet		65.82 8	52.08 16	74.99 4	88.07 3	85.21 1	41.08 20	63.08 7	51.67 38

DeepPCF-MVS	81.24 5	66.76 5	54.74 13	74.77 5	76.18 11	73.22 5	46.42 14	63.06 9	74.91 3

tmmvs		66.88 4	56.70 10	73.68 9	81.20 7	66.01 13	50.58 8	62.81 10	73.82 6

TAPA-MVS	77.73 12	58.58 19	53.61 15	61.89 19	64.81 26	60.73 21	45.40 16	61.82 11	60.14 28
Andrea Romanoni, Matteo Matteucci: TAPA-MVS: Textureless-Aware PAtchMatch Multi-View Stereo. ICCV 2019
COLMAP_ROB	83.01 3	61.51 17	50.83 19	68.63 14	65.07 24	66.61 9	40.64 21	61.02 12	74.20 4
Johannes L. SchÃ¶nberger, Enliang Zheng, Marc Pollefeys, Jan-Michael Frahm: Pixelwise View Selection for Unstructured Multi-View Stereo. ECCV 2016
mvs_zhu_1030		62.39 14	50.41 20	70.37 12	75.70 13	66.39 12	40.10 22	60.73 13	69.02 14

LTVRE_ROB	86.10 1	69.45 2	58.19 7	76.95 2	81.50 6	78.07 3	55.83 5	60.55 14	71.29 9
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
test_1124		61.69 16	67.72 2	57.67 34	71.65 17	46.06 42	75.31 2	60.13 15	55.30 33

TAPA-MVS(SR)		57.16 22	54.76 12	58.76 32	67.29 22	48.19 38	50.15 9	59.37 16	60.81 24

test_1120		63.22 13	68.85 1	59.47 31	78.81 10	51.99 33	79.45 1	58.25 17	47.61 46

tm-dncc		63.90 12	48.20 23	74.37 6	81.20 7	66.01 13	38.20 24	58.21 18	75.91 2

P-MVSNet		54.95 27	50.02 22	58.24 33	66.31 23	47.74 40	41.88 19	58.16 19	60.67 25

OpenMVS_ROB	70.19 17	51.98 34	50.29 21	53.10 35	65.07 24	39.38 48	42.82 18	57.77 20	54.85 34

test_1126		58.81 18	54.80 11	61.48 21	70.94 18	56.27 27	53.41 6	56.19 21	57.23 31

Pnet-blend++		56.52 24	51.51 17	59.87 26	62.39 29	66.50 10	47.44 12	55.57 22	50.71 40

Pnet-blend		56.52 24	51.51 17	59.87 26	62.39 29	66.50 10	47.44 12	55.57 22	50.71 40

CasMVSNet(base)		55.44 26	43.59 29	63.33 18	63.97 27	65.66 16	32.47 37	54.72 24	60.37 27

CIDER		46.95 37	44.89 26	48.33 42	50.37 43	46.20 41	35.21 31	54.58 25	48.41 44
Qingshan Xu and Wenbing Tao: Learning Inverse Depth Regression for Multi-View Stereo with Correlation Cost Volume. AAAI 2020
CasMVSNet(SR_A)		53.46 30	44.06 28	59.72 29	57.37 33	62.68 18	34.23 34	53.90 26	59.09 30

CasMVSNet(SR_B)		65.31 9	44.19 27	79.39 1	91.10 1	74.56 4	35.00 33	53.38 27	72.51 7

vp_mvsnet		42.75 43	30.63 44	50.83 38	53.01 41	47.99 39	8.57 73	52.69 28	51.49 39

PLC	73.85 16	54.06 28	45.01 25	60.08 24	55.47 37	60.30 23	37.58 25	52.44 29	64.49 17
Jie Liao, Yanping Fu, Qingan Yan, Chunxia xiao: Pyramid Multi-View Stereo with Local Consistency. Pacific Graphics 2019
COLMAP(base)		57.56 20	45.74 24	65.45 16	63.08 28	61.37 20	39.37 23	52.10 30	71.91 8

ACMM	79.39 9	52.37 33	40.99 35	59.96 25	58.56 32	60.72 22	31.00 38	50.98 31	60.62 26
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
ACMP	79.16 10	51.84 35	40.13 38	59.65 30	57.20 34	58.68 25	30.46 39	49.80 32	63.09 21
Qingshan Xu and Wenbing Tao: Planar Prior Assisted PatchMatch Multi-View Stereo. AAAI 2020
PCF-MVS	74.62 15	56.56 23	43.50 30	65.27 17	73.58 16	59.98 24	37.23 27	49.76 33	62.24 22
Andreas Kuhn, Shan Lin, Oliver Erdler: Plane Completion and Filtering for Multi-View Stereo Reconstruction. GCPR 2019
ACMH+	77.89 11	47.08 36	42.29 32	50.27 40	54.95 38	39.16 49	35.24 30	49.34 34	56.70 32

ACMH	76.49 14	45.64 38	41.63 34	48.31 43	53.97 40	38.96 50	34.00 36	49.25 35	52.00 36
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
LPCS		53.86 29	42.82 31	61.22 22	67.38 21	54.15 28	37.38 26	48.27 36	62.14 23

MVSNet_plusplus		36.25 50	26.59 50	42.70 47	54.00 39	14.25 70	5.76 80	47.42 37	59.84 29

COLMAP(SR)		52.61 32	41.79 33	59.83 28	58.75 31	54.00 29	37.15 28	46.43 38	66.74 16

GSE		53.18 31	40.31 36	61.77 20	69.15 19	52.25 32	35.20 32	45.43 39	63.90 20

IB-MVS	62.13 19	39.31 44	35.63 40	41.76 50	46.97 46	38.95 51	26.86 44	44.41 40	39.36 50
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
HY-MVS	64.64 18	44.49 39	40.14 37	47.40 44	52.03 42	42.24 46	36.35 29	43.93 41	47.92 45

PVSNet_LR		43.87 41	33.54 42	50.75 39	49.15 44	50.81 35	23.32 47	43.77 42	52.30 35

R-MVSNet		37.45 47	34.59 41	39.35 51	41.64 49	42.85 45	25.44 46	43.75 43	33.57 54

Pnet_fast		44.24 40	35.70 39	49.94 41	46.98 45	58.15 26	29.16 41	42.24 44	44.70 48

MVSCRF		34.84 51	29.04 49	38.71 52	41.35 51	44.92 44	18.79 51	39.30 45	29.86 55

PVSNet	58.17 21	38.39 46	31.41 43	43.04 46	41.89 48	37.34 52	29.69 40	33.13 46	49.88 42

PVSNet_0	51.08 22	39.05 45	29.60 45	45.34 45	44.29 47	39.84 47	26.44 45	32.77 47	51.89 37

ANet		43.24 42	29.49 46	52.40 36	55.77 35	52.26 30	27.30 43	31.68 48	49.19 43

MVSNet_++		21.10 57	16.58 61	24.11 58	22.79 58	10.88 76	1.86 84	31.30 49	38.65 52

test_1205		36.84 48	29.16 47	41.97 49	41.52 50	49.04 36	27.34 42	30.98 50	35.34 53

BP-MVSNet		32.65 53	23.64 51	38.65 53	40.81 52	33.99 54	19.28 49	28.00 51	41.15 49
Christian Sormann, Patrick KnÃ¶belreiter, Andreas Kuhn, Mattia Rossi, Thomas Pock, Friedrich Fraundorfer: BP-MVSNet: Belief-Propagation-Layers for Multi-View-Stereo. 3DV 2020
MVS_test_1		21.23 56	16.85 58	24.15 57	31.31 55	18.73 61	6.46 79	27.23 52	22.39 61

Pnet-eth		20.12 59	18.76 57	21.02 60	24.61 57	21.44 57	13.26 56	24.26 53	17.02 67

PMVS	80.48 6	57.42 21	29.11 48	76.30 3	79.61 9	79.41 2	34.22 35	24.00 54	69.88 11
Y. Furukawa, J. Ponce: Accurate, dense, and robust multiview stereopsis. PAMI (2010)
Snet		19.16 60	19.74 52	18.78 61	22.45 59	15.11 67	16.74 53	22.73 55	18.76 65

unsupervisedMVS_cas		26.15 54	18.92 55	30.97 55	30.39 56	34.23 53	15.24 54	22.60 56	28.30 56

unMVSmet		11.69 77	16.62 60	8.41 79	8.35 80	8.22 79	12.26 61	20.99 57	8.66 76

F/T MVSNet+Gipuma		18.31 61	19.68 53	17.39 63	17.36 63	13.22 72	18.92 50	20.43 58	21.60 62

CPR_FA		15.63 64	16.76 59	14.87 71	12.15 71	14.63 69	13.13 57	20.39 59	17.84 66

metmvs_fine		9.11 81	14.53 64	5.49 83	5.49 84	4.77 82	8.81 72	20.25 60	6.22 78

A-TVSNet + Gipuma		33.17 52	19.66 54	42.17 48	38.64 54	48.93 37	20.23 48	19.09 61	38.94 51

MVSNet + Gipuma		17.51 62	18.78 56	16.66 65	16.50 66	12.75 73	18.63 52	18.92 62	20.74 63

A1Net		14.88 65	15.39 62	14.54 73	12.69 69	11.35 75	12.32 60	18.45 63	19.59 64

ANet-0.75		36.48 49	14.29 65	51.28 37	55.77 35	52.26 30	10.99 63	17.58 64	45.82 47

hgnet		14.22 70	13.23 66	14.89 69	18.80 61	20.68 59	10.37 68	16.09 65	5.18 80

DPSNet		14.22 70	13.23 66	14.89 69	18.80 61	20.68 59	10.37 68	16.09 65	5.18 80

MVE	40.22 23	12.38 74	14.61 63	10.89 78	7.98 81	12.62 74	13.54 55	15.69 67	12.06 74
Simon Fuhrmann, Fabian Langguth, Michael Goesele: MVE - A Multi-View Reconstruction Environment. EUROGRAPHICS Workshops on Graphics and Cultural Heritage (2014)
TVSNet		13.85 72	12.75 69	14.58 72	11.20 72	15.62 65	10.72 65	14.79 68	16.93 68

test3		13.52 73	12.47 70	14.23 74	10.80 73	15.75 62	10.84 64	14.10 69	16.14 69

confMetMVS		8.11 82	9.85 77	6.95 81	7.36 82	4.72 83	6.48 78	13.22 70	8.77 75

test_mvsss		14.34 69	8.77 79	18.06 62	16.23 67	23.38 55	4.71 81	12.84 71	14.55 70

RMVSNet		9.14 80	12.82 68	6.69 82	7.06 83	7.73 80	12.89 58	12.74 72	5.29 79

QQQNet		14.60 66	11.15 75	16.90 64	10.35 74	15.13 66	9.88 70	12.42 73	25.21 57

SVVNet		14.42 67	11.43 72	16.41 66	10.16 75	15.66 63	10.44 66	12.42 73	23.41 59

ternet		14.42 67	11.43 72	16.41 66	10.16 75	15.66 63	10.44 66	12.42 73	23.41 59

SGNet		11.80 76	10.60 76	12.60 75	9.98 77	14.69 68	8.98 71	12.23 76	13.13 72

PSD-MVSNet		11.08 78	9.72 78	12.00 77	9.76 78	13.96 71	8.14 74	11.29 77	12.27 73

Cas-MVS_preliminary		20.88 58	11.16 74	27.36 56	12.72 68	5.17 81	11.58 62	10.75 78	64.20 18

CCVNet		17.35 63	11.44 71	21.29 59	16.96 65	21.70 56	12.75 59	10.13 79	25.21 57

firsttry		10.67 79	8.36 80	12.20 76	12.25 70	10.50 77	7.87 76	8.86 80	13.86 71

unMVSv1		7.70 83	6.95 82	8.20 80	8.59 79	9.59 78	6.96 77	6.95 81	6.43 77

example		11.91 75	7.31 81	14.98 68	20.88 60	20.72 58	8.09 75	6.53 82	3.34 82

FADENet		0.15 84	0.18 84	0.13 85	0.22 86	0.09 85	0.23 86	0.12 83	0.07 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

CMPMVS	59.41 20	21.70 55	0.34 83	35.93 54	38.78 53	69.01 8	0.69 85	0.00 84	0.00 85
M. Jancosek, T. Pajdla: Multi-View Reconstruction Preserving Weakly-Supported Surfaces. CVPR 2011
test_MVS							4.45 82

test_robustmvs				1.67 84	2.86 85	1.42 84	2.72 83		0.73 83

UnsupFinetunedMVSNet					17.36 63

Low-resolution many-view benchmark