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	89.96 1	79.79 1	72.17 1	84.87 1	84.94 1	85.37 1	66.97 1	77.38 1	84.30 1

DeepC-MVS_fast	89.43 2	78.34 2	69.56 4	84.20 2	84.26 2	84.39 2	64.46 3	74.66 4	83.95 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	88.79 3	77.94 3	69.90 3	83.31 3	82.66 7	83.75 3	64.27 4	75.52 3	83.50 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	84.62 6	76.12 4	70.08 2	80.15 7	80.96 8	80.06 9	63.78 5	76.37 2	79.45 14
Andrea Romanoni, Matteo Matteucci: TAPA-MVS: Textureless-Aware PAtchMatch Multi-View Stereo. ICCV 2019
PCF-MVS	84.11 10	75.39 5	66.57 6	81.27 5	83.29 4	80.62 7	62.19 6	70.95 6	79.89 11
Andreas Kuhn, Shan Lin, Oliver Erdler: Plane Completion and Filtering for Multi-View Stereo Reconstruction. GCPR 2019
3Dnovator+	87.14 4	74.49 6	63.75 14	81.65 4	83.81 3	82.92 4	60.57 10	66.93 18	78.22 15

3Dnovator	86.66 5	74.48 7	64.71 9	81.00 6	82.81 5	82.90 5	61.47 8	67.94 14	77.28 16

COLMAP(base)		74.23 8	65.63 8	79.96 9	78.64 11	79.30 11	61.52 7	69.74 10	81.95 4

PLC	84.53 7	73.65 9	66.64 5	78.33 15	75.81 21	78.67 13	61.39 9	71.89 5	80.49 8
Jie Liao, Yanping Fu, Qingan Yan, Chunxia xiao: Pyramid Multi-View Stereo with Local Consistency. Pacific Graphics 2019
COLMAP(SR)		73.53 10	63.77 13	80.04 8	79.79 9	78.44 14	60.49 11	67.05 17	81.87 5

TAPA-MVS(SR)		73.30 11	64.27 10	79.32 11	82.74 6	75.33 19	59.44 12	69.10 11	79.88 12

GSE		73.02 12	64.14 11	78.95 12	77.59 14	78.43 15	59.39 13	68.88 12	80.83 7

ACMP	84.23 8	72.76 13	62.76 16	79.42 10	76.30 19	80.59 8	54.92 19	70.60 7	81.37 6
Qingshan Xu and Wenbing Tao: Planar Prior Assisted PatchMatch Multi-View Stereo. AAAI 2020
LTVRE_ROB	82.13 13	72.22 14	66.24 7	76.21 18	77.23 15	74.60 23	65.16 2	67.32 16	76.79 18
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	84.12 9	72.07 15	62.37 18	78.54 14	75.71 22	79.93 10	54.53 20	70.21 9	79.97 10
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
OpenMVS	83.78 11	71.88 16	61.70 19	78.66 13	78.88 10	81.24 6	56.67 17	66.73 19	75.86 19

COLMAP_ROB	80.39 16	71.80 17	62.64 17	77.90 16	74.64 23	78.97 12	57.39 16	67.89 15	80.10 9
Johannes L. Schönberger, Enliang Zheng, Marc Pollefeys, Jan-Michael Frahm: Pixelwise View Selection for Unstructured Multi-View Stereo. ECCV 2016
ACMH+	81.04 14	71.42 18	63.93 12	76.41 17	78.38 12	71.06 29	58.99 14	68.87 13	79.80 13

IB-MVS	80.51 15	69.16 19	63.18 15	73.15 22	76.86 18	71.66 28	56.10 18	70.26 8	70.91 25
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
LPCS		68.73 20	60.59 20	74.16 21	72.17 25	74.57 24	57.78 15	63.39 23	75.74 20

HY-MVS	83.01 12	68.57 21	58.63 22	75.20 19	77.06 16	73.92 25	53.36 22	63.89 22	74.62 21

ACMH	80.38 17	68.42 22	59.50 21	74.37 20	77.00 17	69.10 34	54.34 21	64.66 20	77.02 17
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
OpenMVS_ROB	74.94 19	63.99 23	53.23 25	71.15 26	75.82 20	68.57 35	51.19 25	55.27 28	69.07 28

A-TVSNet + Gipuma		63.67 24	50.37 30	72.54 24	69.76 29	76.21 18	48.72 28	52.03 33	71.64 24

PVSNet_0	73.20 20	63.55 25	51.85 28	71.35 25	72.10 26	69.61 33	47.57 29	56.13 26	72.33 23

CIDER		63.41 26	52.95 26	70.39 29	71.10 28	72.79 26	49.19 26	56.71 25	67.28 29
Qingshan Xu and Wenbing Tao: Learning Inverse Depth Regression for Multi-View Stereo with Correlation Cost Volume. AAAI 2020
BP-MVSNet		63.26 27	52.48 27	70.45 28	73.14 24	64.36 39	45.07 30	59.89 24	73.84 22
Christian Sormann, Patrick Knöbelreiter, Andreas Kuhn, Mattia Rossi, Thomas Pock, Friedrich Fraundorfer: BP-MVSNet: Belief-Propagation-Layers for Multi-View-Stereo. 3DV 2020
PVSNet	78.82 18	62.67 28	53.41 24	68.85 31	69.68 30	66.37 37	51.28 24	55.54 27	70.50 27

test_1126		61.29 29	44.16 33	72.71 23	77.87 13	75.17 20	40.34 32	47.98 37	65.09 30

P-MVSNet		61.04 30	58.45 23	62.77 37	60.78 42	63.39 41	52.67 23	64.23 21	64.13 31

test_1205		58.34 31	47.13 32	65.81 35	67.31 33	74.82 21	40.34 32	53.91 30	55.31 43

AttMVS		58.26 32	51.80 29	62.56 39	58.33 43	65.96 38	49.04 27	54.55 29	63.38 32

R-MVSNet		56.72 33	48.74 31	62.04 40	61.37 39	66.93 36	43.74 31	53.73 31	57.82 40

CasMVSNet(SR_A)		56.09 34	36.56 41	69.11 30	67.36 31	77.93 16	28.99 42	44.12 42	62.05 34

mvs_zhu_1030		55.64 35	40.20 37	65.93 34	63.79 36	72.44 27	31.72 39	48.68 35	61.57 36

Pnet-new-		55.36 36	32.32 46	70.71 27	71.82 27	69.62 32	32.54 38	32.10 58	70.69 26

ANet-0.75		55.32 37	41.08 35	64.81 36	61.58 37	70.76 30	34.14 37	48.01 36	62.08 33

CasMVSNet(base)		55.02 38	35.18 42	68.25 32	67.01 34	76.54 17	27.28 44	43.09 43	61.20 37

ANet		53.16 39	38.86 38	62.70 38	61.58 37	70.76 30	34.74 35	42.99 44	55.74 42

Pnet_fast		50.82 40	26.87 57	66.78 33	64.56 35	74.79 22	17.08 68	36.66 51	61.01 38

MVSNet		49.13 41	37.77 39	56.70 41	50.43 48	63.92 40	30.66 40	44.88 40	55.77 41

unsupervisedMVS_cas		45.48 42	32.83 44	53.91 46	51.79 46	59.72 45	24.95 49	40.71 45	50.24 45

Pnet-blend++		44.30 43	26.07 59	56.45 42	60.87 40	61.19 43	14.59 69	37.56 48	47.29 48

Pnet-blend		44.30 43	26.07 59	56.45 42	60.87 40	61.19 43	14.59 69	37.56 48	47.29 48

MVSNet_plusplus		43.14 45	25.17 63	55.12 45	67.36 31	36.20 57	12.78 72	37.55 50	61.80 35

test_1124		42.16 46	21.84 69	55.71 44	53.19 44	61.81 42	20.50 62	23.19 70	52.14 44

CasMVSNet(SR_B)		42.10 47	37.13 40	45.41 49	43.33 51	57.07 48	29.42 41	44.85 41	35.82 57

MVSCRF		41.80 48	27.23 55	51.52 47	48.62 49	58.89 46	23.51 58	30.95 59	47.03 50

CPR_FA		41.50 49	40.70 36	42.03 52	37.78 56	40.09 53	34.69 36	46.72 39	48.23 47

A1Net		40.64 50	42.66 34	39.29 53	35.99 57	32.67 65	36.23 34	49.09 34	49.23 46

MVSNet_++		40.21 51	29.12 52	47.60 48	52.26 45	32.34 67	5.74 76	52.50 32	58.20 39

Snet		37.53 52	26.97 56	44.56 51	51.19 47	38.17 54	19.63 66	34.31 55	44.33 51

PMVS	47.18 22	35.10 53	20.52 71	44.82 50	43.36 50	58.67 47	28.88 43	12.16 74	32.42 60
Y. Furukawa, J. Ponce: Accurate, dense, and robust multiview stereopsis. PAMI (2010)
test3		33.53 54	31.08 48	35.17 55	30.55 64	35.73 58	26.13 48	36.03 52	39.22 52

TVSNet		32.96 55	29.87 50	35.02 56	31.41 63	35.01 59	23.96 55	35.79 53	38.64 53

SGNet		31.82 56	29.50 51	33.36 62	29.84 67	34.36 61	24.75 51	34.26 56	35.90 56

Pnet-eth		31.53 57	34.06 43	29.85 69	41.28 53	17.75 75	20.71 61	47.41 38	30.53 66

PSD-MVSNet		31.43 58	29.01 53	33.04 63	29.84 67	34.06 62	24.49 54	33.52 57	35.22 58

F/T MVSNet+Gipuma		31.25 59	25.73 62	34.93 57	34.16 58	33.91 63	23.61 57	27.84 64	36.73 54

MVSNet + Gipuma		31.15 60	26.01 61	34.58 58	33.87 60	33.39 64	23.75 56	28.26 63	36.47 55

hgnet		30.74 61	25.12 64	34.49 59	40.98 54	47.29 49	24.56 52	25.68 67	15.18 73

DPSNet		30.74 61	25.12 64	34.49 59	40.98 54	47.29 49	24.56 52	25.68 67	15.18 73

MVE	39.65 23	30.69 63	31.26 47	30.31 66	23.68 72	36.67 55	26.92 45	35.60 54	30.60 65
Simon Fuhrmann, Fabian Langguth, Michael Goesele: MVE - A Multi-View Reconstruction Environment. EUROGRAPHICS Workshops on Graphics and Cultural Heritage (2014)
QQQNet		30.24 64	27.34 54	32.18 64	30.23 65	34.61 60	24.76 50	29.92 60	31.70 63

CCVNet		29.61 65	21.20 70	35.21 54	32.95 61	40.98 52	17.15 67	25.24 69	31.70 63

firsttry		28.73 66	23.72 68	32.06 65	31.74 62	30.80 70	21.15 60	26.30 66	33.65 59

SVVNet		28.03 67	24.84 66	30.16 67	26.30 70	31.88 68	19.75 64	29.92 60	32.29 61

ternet		28.03 67	24.84 66	30.16 67	26.30 70	31.88 68	19.75 64	29.92 60	32.29 61

unMVSv1		28.02 69	26.35 58	29.13 70	30.17 66	32.63 66	26.14 47	26.55 65	24.58 68

example		27.92 70	19.45 72	33.56 61	42.92 52	46.29 51	19.84 63	19.07 71	11.46 76

RMVSNet		26.78 71	32.62 45	22.89 71	27.05 69	26.01 71	26.90 46	38.35 47	15.62 72

metmvs_fine		22.70 72	30.59 49	17.43 75	17.52 74	16.63 77	22.55 59	38.64 46	18.14 70

unMVSmet		17.94 73	14.89 73	19.98 73	19.45 73	24.72 72	13.90 71	15.87 72	15.75 71

test_1120		17.74 74	12.97 74	20.92 72	13.88 76	22.05 73	12.33 73	13.61 73	26.82 67

Cas-MVS_preliminary		14.58 75	9.47 76	17.99 74	13.62 77	18.53 74	10.67 74	8.27 76	21.82 69

confMetMVS		13.05 76	10.92 75	14.47 76	14.73 75	17.12 76	10.20 75	11.64 75	11.56 75

CMPMVS	59.16 21	7.91 77	0.09 78	13.12 77	3.14 78	36.21 56	0.19 78	0.00 78	0.00 78
M. Jancosek, T. Pajdla: Multi-View Reconstruction Preserving Weakly-Supported Surfaces. CVPR 2011
FADENet		0.56 78	0.62 77	0.51 78	0.86 79	0.42 78	0.78 77	0.45 77	0.27 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					34.16 58

Low-resolution many-view benchmark