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	96.53 2	93.94 1	90.03 3	96.55 2	96.08 8	96.95 4	85.43 3	94.62 6	96.62 3

test_1124		93.84 2	93.94 1	93.78 15	98.74 2	85.65 41	93.23 1	94.65 5	96.94 1

test_1120		93.24 3	93.60 2	93.00 18	97.25 6	85.92 40	91.86 2	95.34 2	95.83 7

Pnet-new-		92.40 4	89.38 4	94.42 11	97.52 5	88.91 29	83.32 5	95.44 1	96.83 2

3Dnovator+	96.13 3	92.29 5	87.63 5	95.40 6	94.86 14	95.31 11	82.29 6	92.97 9	96.03 5

AttMVS		91.69 6	84.94 15	96.20 3	99.02 1	93.24 14	77.86 14	92.02 16	96.33 4

LTVRE_ROB	96.88 1	91.35 7	86.15 8	94.81 7	95.87 9	94.76 12	85.06 4	87.25 30	93.80 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
MVSNet		91.32 8	84.12 17	96.13 4	98.67 3	99.19 1	73.29 21	94.95 3	90.53 35

test_1126		91.28 9	86.68 7	94.36 12	96.53 7	93.15 16	81.50 7	91.85 17	93.39 23

OpenMVS	94.22 8	91.21 10	86.10 9	94.61 8	93.08 20	95.94 7	79.58 10	92.61 11	94.82 15

tmmvs		91.01 11	85.87 10	94.43 10	95.72 10	92.39 19	80.42 8	91.32 18	95.17 12

mvs_zhu_1030		90.16 12	81.94 21	95.63 5	94.87 13	97.32 3	69.88 29	94.00 8	94.70 16

TAPA-MVS(SR)		89.94 13	87.22 6	91.74 24	93.91 18	88.71 31	80.31 9	94.13 7	92.61 25

DeepC-MVS	95.41 4	89.62 14	85.22 13	92.56 20	91.17 26	92.38 21	78.02 13	92.43 14	94.13 20
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
CasMVSNet(SR_B)		89.61 15	77.50 31	97.68 1	98.62 4	99.07 2	66.98 34	88.02 27	95.36 10

TAPA-MVS	93.32 12	89.43 16	84.97 14	92.40 21	92.59 21	92.76 17	77.17 16	92.77 10	91.86 28
Andrea Romanoni, Matteo Matteucci: TAPA-MVS: Textureless-Aware PAtchMatch Multi-View Stereo. ICCV 2019
Pnet-blend++		89.08 17	85.48 11	91.49 25	94.54 15	95.61 9	78.52 11	92.44 12	84.31 42

Pnet-blend		89.08 17	85.48 11	91.49 25	94.54 15	95.61 9	78.52 11	92.44 12	84.31 42

DeepPCF-MVS	94.58 5	89.02 19	82.96 19	93.05 17	91.93 23	92.37 22	76.15 18	89.78 22	94.87 14

tm-dncc		88.78 20	80.08 25	94.58 9	95.72 10	92.39 19	72.62 24	87.54 28	95.64 8

P-MVSNet		88.77 21	81.05 22	93.92 14	95.26 12	91.17 25	72.77 23	89.32 23	95.32 11

LPCS		88.71 22	80.50 24	94.18 13	91.38 24	95.63 8	75.15 20	85.84 36	95.53 9

HY-MVS	91.43 15	88.66 23	84.44 16	91.47 27	91.38 24	89.67 28	76.53 17	92.35 15	93.36 24

DeepC-MVS_fast	94.34 7	88.57 24	82.00 20	92.95 19	91.15 27	92.65 18	75.30 19	88.70 24	95.05 13
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
OpenMVS_ROB	91.80 14	87.38 25	83.78 18	89.79 31	94.43 17	83.34 43	77.69 15	89.88 21	91.59 30

GSE		86.24 26	77.00 32	92.39 22	89.33 30	93.63 13	71.06 27	82.94 42	94.22 19

COLMAP_ROB	94.48 6	85.43 27	78.72 27	89.90 30	84.17 39	91.10 26	68.97 32	88.47 26	94.42 18
Johannes L. SchÃ¶nberger, Enliang Zheng, Marc Pollefeys, Jan-Michael Frahm: Pixelwise View Selection for Unstructured Multi-View Stereo. ECCV 2016
CasMVSNet(base)		85.15 28	74.67 37	92.14 23	88.81 31	96.80 5	62.49 43	86.85 31	90.81 34

COLMAP(base)		85.07 29	78.37 28	89.55 32	86.00 33	88.85 30	71.31 26	85.43 38	93.78 22

COLMAP(SR)		84.90 30	79.01 26	88.83 34	86.24 32	87.81 36	72.86 22	85.16 39	92.45 26

CasMVSNet(SR_A)		84.65 31	76.33 36	90.20 29	84.56 37	96.27 6	65.84 37	86.81 32	89.77 38

PLC	91.02 16	83.72 32	77.86 29	87.63 35	83.31 41	87.52 39	70.85 28	84.87 40	92.08 27
Jie Liao, Yanping Fu, Qingan Yan, Chunxia xiao: Pyramid Multi-View Stereo with Local Consistency. Pacific Graphics 2019
PCF-MVS	89.43 18	82.97 33	73.73 39	89.13 33	90.40 28	87.68 37	68.35 33	79.10 45	89.30 39
Andreas Kuhn, Shan Lin, Oliver Erdler: Plane Completion and Filtering for Multi-View Stereo Reconstruction. GCPR 2019
ACMH+	93.58 10	82.41 34	77.68 30	85.56 42	85.31 35	80.89 47	69.59 30	85.78 37	90.47 36

ACMH	93.61 9	82.01 35	76.51 35	85.67 41	85.78 34	80.27 49	66.29 36	86.73 34	90.97 33
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
ACMM	93.33 11	81.98 36	74.23 38	87.15 36	82.31 45	88.08 35	65.33 38	83.12 41	91.07 31
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
CIDER		81.10 37	81.01 23	81.16 47	78.95 51	81.68 46	71.35 25	90.68 19	82.84 48
Qingshan Xu and Wenbing Tao: Learning Inverse Depth Regression for Multi-View Stereo with Correlation Cost Volume. AAAI 2020
PVSNet_LR		81.05 38	72.92 41	86.47 38	84.32 38	87.64 38	58.56 46	87.27 29	87.46 40

ACMP	92.54 13	80.70 39	71.88 43	86.59 37	80.49 49	88.25 34	63.03 41	80.72 44	91.01 32
Qingshan Xu and Wenbing Tao: Planar Prior Assisted PatchMatch Multi-View Stereo. AAAI 2020
R-MVSNet		80.37 40	76.91 33	82.68 44	83.10 42	81.84 45	65.20 39	88.62 25	83.11 47

Pnet_fast		79.81 41	72.73 42	84.52 43	82.40 44	88.35 33	58.67 45	86.79 33	82.82 49

IB-MVS	85.98 20	79.62 42	76.62 34	81.62 46	85.10 36	80.82 48	66.55 35	86.69 35	78.96 52
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
vp_mvsnet		79.40 43	61.80 49	91.14 28	93.13 19	89.83 27	28.94 76	94.66 4	90.45 37

ANet		78.35 44	67.13 47	85.83 40	81.24 48	92.34 23	57.90 47	76.37 48	83.92 45

test_1205		76.84 45	69.16 46	81.96 45	82.96 43	88.41 32	62.91 42	75.41 50	74.53 55

PVSNet	86.72 19	76.20 46	73.67 40	77.88 52	79.20 50	72.57 54	69.58 31	77.77 46	81.87 50

PVSNet_0	81.89 21	75.64 47	69.86 44	79.48 50	78.69 52	74.99 51	63.13 40	76.60 47	84.77 41

MVSCRF		75.50 48	69.78 45	79.31 51	77.67 54	82.99 44	49.66 51	89.90 20	77.28 53

BP-MVSNet		73.12 49	63.32 48	79.66 49	81.44 46	74.39 52	52.82 48	73.82 51	83.14 46
Christian Sormann, Patrick KnÃ¶belreiter, Andreas Kuhn, Mattia Rossi, Thomas Pock, Friedrich Fraundorfer: BP-MVSNet: Belief-Propagation-Layers for Multi-View-Stereo. 3DV 2020
A-TVSNet + Gipuma		72.66 50	60.91 51	80.50 48	75.15 55	84.76 42	59.05 44	62.76 59	81.58 51

PMVS	89.60 17	72.53 51	40.67 72	93.77 16	92.28 22	93.18 15	48.78 53	32.55 80	95.84 6
Y. Furukawa, J. Ponce: Accurate, dense, and robust multiview stereopsis. PAMI (2010)
ANet-0.75		70.67 52	47.80 65	85.91 39	81.25 47	92.34 23	38.12 63	57.47 63	84.15 44

MVSNet_plusplus		65.61 53	51.21 58	75.21 53	89.61 29	44.24 62	19.91 80	82.51 43	91.77 29

unsupervisedMVS_cas		61.20 54	49.84 63	68.78 54	63.77 59	73.15 53	38.43 62	61.26 60	69.42 56

Pnet-eth		61.19 55	57.15 55	63.88 56	83.71 40	47.10 60	41.45 59	72.85 53	60.82 60

MVS_test_1		60.08 56	50.72 59	66.32 55	78.10 53	52.08 56	25.70 79	75.74 49	68.79 57

F/T MVSNet+Gipuma		59.72 57	61.47 50	58.55 57	70.70 56	40.01 65	52.07 49	70.87 54	64.95 58

MVSNet + Gipuma		58.38 58	60.00 52	57.30 58	68.90 58	39.34 67	51.70 50	68.30 56	63.65 59

Snet		53.20 59	57.22 54	50.52 63	62.60 60	36.23 69	49.14 52	65.30 58	52.73 66

Cas-MVS_preliminary		50.29 60	41.63 70	56.06 60	43.18 71	30.41 80	35.30 69	47.96 71	94.60 17

CPR_FA		50.17 61	52.38 56	48.69 64	44.83 70	42.31 63	43.51 58	61.26 60	58.94 61

MVSNet_++		50.06 62	40.45 73	56.46 59	57.82 61	34.95 72	7.47 84	73.43 52	76.61 54

unMVSmet		49.47 63	58.19 53	43.66 67	50.58 66	39.49 66	47.78 54	68.60 55	40.90 74

CCVNet		46.86 64	39.85 76	51.54 62	50.35 67	46.44 61	36.28 68	43.42 77	57.82 62

test_mvsss		46.43 65	32.75 80	55.54 61	53.54 65	61.33 55	19.75 81	45.75 75	51.76 68

hgnet		44.78 66	46.10 66	43.90 65	55.34 63	51.87 57	37.81 64	54.38 65	24.49 80

DPSNet		44.78 66	46.10 66	43.90 65	55.34 63	51.87 57	37.81 64	54.38 65	24.49 80

MVE	73.61 22	42.73 68	50.51 60	37.54 77	28.32 83	41.81 64	47.01 55	54.00 67	42.50 73
Simon Fuhrmann, Fabian Langguth, Michael Goesele: MVE - A Multi-View Reconstruction Environment. EUROGRAPHICS Workshops on Graphics and Cultural Heritage (2014)
A1Net		42.71 69	50.00 61	37.85 76	32.83 77	28.57 82	41.36 60	58.63 62	52.14 67

RMVSNet		42.28 70	49.56 64	37.43 78	46.49 69	36.28 68	46.14 56	52.98 68	29.51 77

TVSNet		41.69 71	43.53 68	40.47 73	34.72 74	35.30 70	36.98 66	50.08 69	51.38 69

QQQNet		41.37 72	40.86 71	41.70 69	32.74 78	34.55 76	34.93 70	46.80 72	57.82 62

confMetMVS		40.92 73	49.99 62	34.88 81	42.05 72	28.82 81	45.31 57	54.67 64	33.76 76

SVVNet		40.48 74	40.38 74	40.54 71	30.67 81	34.64 74	33.96 71	46.80 72	56.31 64

ternet		40.48 74	40.38 74	40.54 71	30.67 81	34.64 74	33.96 71	46.80 72	56.31 64

test3		40.01 76	42.34 69	38.45 75	33.05 76	34.68 73	36.66 67	48.03 70	47.61 70

SGNet		36.94 77	37.96 77	36.27 79	31.95 79	33.84 77	32.37 73	43.54 76	43.01 72

metmvs_fine		36.70 78	52.31 57	26.30 83	26.70 84	24.23 83	38.64 61	65.98 57	27.95 78

firsttry		36.63 79	33.24 79	38.88 74	37.30 73	33.37 78	30.72 74	35.75 79	45.97 71

example		36.07 80	28.13 82	41.37 70	57.61 62	48.26 59	28.65 78	27.61 82	18.25 82

PSD-MVSNet		35.22 81	35.50 78	35.04 80	31.65 80	32.64 79	29.99 75	41.01 78	40.82 75

unMVSv1		30.79 82	29.07 81	31.95 82	33.31 75	35.18 71	28.91 77	29.23 81	27.34 79

CMPMVS	73.10 23	25.96 83	1.21 84	42.46 68	48.37 68	79.01 50	2.43 85	0.00 84	0.00 85
M. Jancosek, T. Pajdla: Multi-View Reconstruction Preserving Weakly-Supported Surfaces. CVPR 2011
FADENet		1.00 84	1.26 83	0.82 85	1.49 86	0.55 85	1.56 86	0.96 83	0.41 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							18.95 82

test_robustmvs				7.77 84	12.93 85	6.88 84	12.31 83		3.50 83

UnsupFinetunedMVSNet					70.70 56

Low-resolution many-view benchmark