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

test_1124		98.99 1	97.89 4	99.72 3	99.70 5	99.84 5	98.10 1	97.68 24	99.60 8

test_1120		98.89 2	98.31 1	99.28 10	99.17 12	99.79 7	97.81 2	98.81 9	98.87 32

3Dnovator	99.15 2	98.46 3	96.92 5	99.48 7	99.05 14	99.63 14	95.44 6	98.41 16	99.78 3

HY-MVS	98.23 9	98.26 4	96.37 6	99.52 6	99.21 11	99.65 11	93.56 13	99.18 2	99.69 4

AttMVS		98.10 5	95.45 16	99.86 1	99.84 1	99.87 3	93.29 15	97.60 27	99.87 2

Pnet-new-		98.02 6	95.95 10	99.40 8	99.49 8	99.61 15	94.31 7	97.59 28	99.11 25

3Dnovator+	98.92 3	97.89 7	95.99 9	99.16 12	98.56 27	99.31 21	94.11 9	97.86 22	99.63 5

OpenMVS	98.12 10	97.86 8	95.83 11	99.21 11	98.55 28	99.47 18	93.67 11	97.98 20	99.61 6

TAPA-MVS(SR)		97.85 9	96.03 8	99.07 15	98.53 29	99.13 23	93.55 14	98.51 14	99.55 12

test_1126		97.78 10	96.13 7	98.89 22	99.62 6	99.52 16	94.30 8	97.95 21	97.53 45

MVSNet		97.77 11	94.82 23	99.73 2	99.71 4	99.87 3	90.82 28	98.81 9	99.61 6

TAPA-MVS	97.92 13	97.67 12	95.54 13	99.09 14	98.77 20	99.34 20	93.04 16	98.04 19	99.15 22
Andrea Romanoni, Matteo Matteucci: TAPA-MVS: Textureless-Aware PAtchMatch Multi-View Stereo. ICCV 2019
tmmvs		97.58 13	95.49 14	98.98 18	98.85 18	98.54 28	93.76 10	97.22 32	99.53 13

LPCS		97.58 13	95.27 19	99.12 13	98.44 30	99.37 19	93.66 12	96.88 36	99.56 11

PVSNet_LR		97.57 15	95.45 16	98.98 18	99.00 15	98.67 26	92.23 20	98.67 11	99.27 20

LTVRE_ROB	99.19 1	97.54 16	95.77 12	98.71 23	98.01 35	99.31 21	96.08 5	95.46 53	98.82 33
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
P-MVSNet		97.42 17	94.13 29	99.61 4	98.93 16	99.94 1	90.70 29	97.55 29	99.94 1

mvs_zhu_1030		97.36 18	94.00 30	99.61 4	99.39 9	99.82 6	88.90 37	99.10 3	99.60 8

OpenMVS_ROB	97.31 17	97.22 19	95.18 20	98.58 25	98.69 24	97.95 33	92.61 18	97.75 23	99.10 26

DeepC-MVS	98.90 4	97.16 20	95.29 18	98.41 26	98.16 34	97.76 34	92.20 21	98.38 17	99.32 19
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
GSE		96.89 21	94.34 27	98.60 24	97.60 40	98.97 25	92.57 19	96.11 47	99.23 21

tm-dncc		96.71 22	93.39 36	98.92 21	98.85 18	98.54 28	89.98 33	96.81 37	99.36 17

Pnet-blend++		96.65 23	98.04 2	95.73 50	99.77 2	99.65 11	97.16 3	98.93 5	87.77 67

Pnet-blend		96.65 23	98.04 2	95.73 50	99.77 2	99.65 11	97.16 3	98.93 5	87.77 67

DeepC-MVS_fast	98.47 5	96.63 25	94.29 28	98.19 28	97.66 38	97.50 38	91.63 23	96.95 34	99.41 15
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
DeepPCF-MVS	98.42 6	96.59 26	94.42 25	98.03 29	97.86 36	97.10 41	91.48 24	97.36 30	99.14 24

R-MVSNet		96.56 27	94.43 24	97.98 30	97.61 39	97.54 37	90.39 31	98.47 15	98.78 34

BP-MVSNet		96.44 28	93.68 33	98.28 27	98.32 31	98.10 32	89.68 34	97.67 25	98.43 38
Christian Sormann, Patrick KnÃ¶belreiter, Andreas Kuhn, Mattia Rossi, Thomas Pock, Friedrich Fraundorfer: BP-MVSNet: Belief-Propagation-Layers for Multi-View-Stereo. 3DV 2020
COLMAP(SR)		96.17 29	93.90 32	97.69 32	96.39 48	97.64 36	91.16 26	96.64 39	99.04 27

CIDER		95.78 30	95.47 15	95.98 48	97.01 46	96.43 45	91.90 22	99.04 4	94.51 56
Qingshan Xu and Wenbing Tao: Learning Inverse Depth Regression for Multi-View Stereo with Correlation Cost Volume. AAAI 2020
ACMH+	98.40 8	95.77 31	92.95 37	97.64 33	96.31 49	97.47 40	89.65 35	96.26 44	99.15 22

IB-MVS	95.41 20	95.66 32	94.83 22	96.21 45	98.30 32	97.74 35	91.10 27	98.57 12	92.58 59
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	98.42 6	95.63 33	92.35 43	97.81 31	95.84 51	98.24 31	88.19 42	96.51 41	99.36 17
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
CasMVSNet(SR_B)		95.57 34	90.42 52	99.00 17	99.52 7	99.94 1	84.68 58	96.16 45	97.54 44

Pnet_fast		95.36 35	92.80 42	97.07 36	98.75 21	98.31 30	88.27 41	97.33 31	94.14 57

CasMVSNet(SR_A)		95.29 36	89.69 55	99.02 16	98.59 25	99.48 17	83.78 59	95.60 50	98.99 29

COLMAP(base)		95.23 37	92.81 40	96.85 39	95.06 55	96.52 43	89.63 36	95.99 48	98.96 31

PVSNet	97.47 15	95.18 38	94.37 26	95.72 52	97.83 37	94.18 52	90.68 30	98.06 18	95.13 55

A-TVSNet + Gipuma		95.08 39	93.98 31	95.81 49	95.21 53	96.00 48	91.24 25	96.73 38	96.22 51

PLC	97.35 16	94.93 40	92.81 40	96.35 44	94.63 56	95.74 49	90.06 32	95.56 52	98.69 35
Jie Liao, Yanping Fu, Qingan Yan, Chunxia xiao: Pyramid Multi-View Stereo with Local Consistency. Pacific Graphics 2019
CasMVSNet(base)		94.85 41	88.68 57	98.96 20	98.70 23	99.67 8	82.37 61	94.99 55	98.50 37

MVSCRF		94.80 42	92.89 38	96.08 46	93.77 60	96.33 46	86.18 50	99.60 1	98.14 39

ACMM	98.09 11	94.80 42	91.85 45	96.76 40	95.11 54	96.48 44	87.94 44	95.76 49	98.69 35
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
PVSNet_0	95.53 19	94.76 44	92.88 39	96.02 47	97.06 45	94.33 51	88.13 43	97.63 26	96.66 49

COLMAP_ROB	98.06 12	94.72 45	92.05 44	96.50 42	93.64 61	96.86 42	87.59 45	96.51 41	99.02 28
Johannes L. SchÃ¶nberger, Enliang Zheng, Marc Pollefeys, Jan-Michael Frahm: Pixelwise View Selection for Unstructured Multi-View Stereo. ECCV 2016
ANet		94.58 46	91.17 47	96.86 38	94.49 57	99.66 9	85.16 55	97.18 33	96.43 50

PCF-MVS	96.03 18	94.34 47	90.51 51	96.89 37	96.93 47	95.66 50	87.59 45	93.42 61	98.08 40
Andreas Kuhn, Shan Lin, Oliver Erdler: Plane Completion and Filtering for Multi-View Stereo Reconstruction. GCPR 2019
unMVSmet		94.27 48	90.80 49	96.58 41	98.30 32	93.88 54	86.93 49	94.67 56	97.57 43

ACMP	97.51 14	93.81 49	89.89 53	96.42 43	93.98 59	96.31 47	85.39 54	94.39 58	98.97 30
Qingshan Xu and Wenbing Tao: Planar Prior Assisted PatchMatch Multi-View Stereo. AAAI 2020
Pnet-eth		93.70 50	94.85 21	92.94 55	98.90 17	88.78 56	92.79 17	96.91 35	91.13 60

test_1205		93.55 51	91.37 46	95.00 53	98.73 22	98.60 27	88.30 40	94.43 57	87.67 69

confMetMVS		93.00 52	90.74 50	94.52 54	98.59 25	89.04 55	85.90 52	95.57 51	95.92 52

vp_mvsnet		92.70 53	82.77 71	99.32 9	99.36 10	99.00 24	66.62 78	98.92 7	99.58 10

ANet-0.75		92.64 54	85.79 63	97.20 35	94.49 57	99.66 9	79.52 63	92.06 64	97.46 47

F/T MVSNet+Gipuma		91.90 55	93.60 34	90.76 58	97.37 43	77.09 67	88.37 38	98.83 8	97.83 41

MVSNet + Gipuma		91.67 56	93.45 35	90.48 60	97.40 41	76.54 68	88.37 38	98.52 13	97.50 46

CPR_FA		88.38 57	88.27 58	88.46 62	88.24 67	81.63 59	87.03 48	89.51 73	95.52 53

unsupervisedMVS_cas		87.73 58	80.56 75	92.51 56	90.02 66	94.04 53	72.99 72	88.12 75	93.47 58

metmvs_fine		87.49 59	89.83 54	85.92 64	91.28 64	78.30 64	83.54 60	96.12 46	88.18 66

MVS_test_1		87.21 60	82.26 72	90.51 59	97.39 42	76.51 69	67.94 77	96.58 40	97.62 42

Cas-MVS_preliminary		86.14 61	81.69 73	89.11 61	92.91 62	77.14 66	71.77 74	91.61 65	97.27 48

RMVSNet		85.99 62	91.12 48	82.56 69	92.56 63	71.89 71	87.23 47	95.02 54	83.24 75

MVSNet_plusplus		85.44 63	75.33 77	92.19 57	99.09 13	77.97 65	54.19 82	96.48 43	99.49 14

hgnet		84.50 64	86.06 61	83.46 67	85.83 70	78.98 62	84.69 56	87.44 78	85.56 70

DPSNet		84.50 64	86.06 61	83.46 67	85.83 70	78.98 62	84.69 56	87.44 78	85.56 70

Snet		83.91 66	89.25 56	80.34 70	95.66 52	59.91 75	85.81 53	92.69 62	85.46 73

MVE	92.54 22	82.65 67	87.52 59	79.40 71	75.10 73	82.70 58	85.94 51	89.11 74	80.40 79
Simon Fuhrmann, Fabian Langguth, Michael Goesele: MVE - A Multi-View Reconstruction Environment. EUROGRAPHICS Workshops on Graphics and Cultural Heritage (2014)
test_mvsss		82.49 68	73.62 79	88.40 63	90.79 65	84.94 57	59.37 80	87.86 76	89.47 65

PMVS	92.94 21	78.84 69	50.69 82	97.61 34	95.96 50	97.48 39	65.74 79	35.64 82	99.38 16
Y. Furukawa, J. Ponce: Accurate, dense, and robust multiview stereopsis. PAMI (2010)
CCVNet		78.62 70	84.56 65	74.66 72	70.01 80	63.28 73	78.75 64	90.36 71	90.68 61

example		78.46 71	69.47 80	84.45 66	87.90 69	79.95 61	69.21 76	69.72 81	85.51 72

TVSNet		76.50 72	85.63 64	70.41 75	72.28 75	53.63 78	77.80 65	93.46 60	85.31 74

MVSNet_++		76.28 73	63.91 81	84.54 65	87.95 68	70.45 72	33.49 84	94.32 59	95.21 54

QQQNet		76.16 74	83.14 69	71.50 74	70.80 76	53.03 79	75.46 69	90.83 67	90.68 61

A1Net		75.91 75	86.79 60	68.65 79	76.15 72	46.80 83	81.23 62	92.34 63	83.01 76

SVVNet		75.58 76	83.42 67	70.36 76	66.75 82	54.11 76	76.01 67	90.83 67	90.21 63

ternet		75.58 76	83.42 67	70.36 76	66.75 82	54.11 76	76.01 67	90.83 67	90.21 63

firsttry		75.37 78	82.99 70	70.28 78	67.26 81	62.42 74	75.38 70	90.60 70	81.16 77

test3		74.14 79	83.69 66	67.77 80	70.24 79	52.13 81	76.30 66	91.09 66	80.94 78

unMVSv1		73.78 80	74.27 78	73.45 73	74.10 74	75.59 70	73.49 71	75.05 80	70.66 82

SGNet		72.64 81	81.38 74	66.81 81	70.44 78	52.48 80	72.96 73	89.80 72	77.50 80

PSD-MVSNet		71.12 82	79.18 76	65.75 82	70.74 77	51.62 82	70.68 75	87.67 77	74.90 81

CMPMVS	77.52 23	27.18 83	1.74 84	44.14 83	51.76 84	80.67 60	3.48 86	0.00 84	0.00 85
M. Jancosek, T. Pajdla: Multi-View Reconstruction Preserving Weakly-Supported Surfaces. CVPR 2011
FADENet		15.30 84	10.26 83	18.67 85	28.34 86	22.58 85	10.01 85	10.51 83	5.07 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

test_MVS							58.19 81

test_robustmvs				32.21 84	49.22 85	30.09 84	47.07 83		17.32 83

UnsupFinetunedMVSNet					97.37 43

Low-resolution many-view benchmark