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	delivery area	electro	forest	playground	terrains

LTVRE_ROB	86.82 1	74.53 1	78.23 1	72.06 1	62.97 1	76.95 2	80.00 2	59.24 4	93.50 2
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
3Dnovator+	83.71 3	72.24 2	75.03 2	70.38 3	57.83 3	78.07 1	71.19 6	61.87 3	92.23 4

3Dnovator	79.41 10	72.11 3	74.21 3	70.71 2	54.58 5	76.83 3	71.81 4	63.51 2	93.85 1

COLMAP_ROB	85.66 2	69.58 4	73.71 4	66.82 4	58.83 2	75.96 4	60.77 13	63.74 1	88.59 6
Johannes L. SchÃ¶nberger, Enliang Zheng, Marc Pollefeys, Jan-Michael Frahm: Pixelwise View Selection for Unstructured Multi-View Stereo. ECCV 2016
OpenMVS	75.38 16	67.44 5	71.75 5	64.57 7	51.05 8	71.84 7	63.63 11	58.26 6	92.45 3

DeepC-MVS_fast	81.78 5	67.22 6	71.51 6	64.36 8	55.68 4	72.92 5	66.58 8	53.58 8	87.35 7
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
DeepC-MVS	83.59 4	65.98 7	69.37 8	63.72 9	52.36 7	70.11 9	63.91 10	57.15 7	86.38 9
Andreas Kuhn, Christian Sormann, Mattia Rossi, Oliver Erdler, Friedrich Fraundorfer: DeepC-MVS: Deep Confidence Prediction for Multi-View Stereo Reconstruction. 3DV 2020
DeepPCF-MVS	81.61 6	65.12 8	70.29 7	61.68 11	53.86 6	70.31 8	64.65 9	50.07 11	86.72 8

CasMVSNet(SR_A)		62.79 9	58.12 20	65.91 5	34.02 22	59.92 14	87.07 1	50.74 9	82.22 15

CasMVSNet(base)		62.06 10	57.36 24	65.19 6	37.99 21	63.36 12	73.58 3	58.62 5	76.74 25

IB-MVS	71.28 17	61.06 11	65.02 10	58.42 12	39.73 17	60.44 13	67.25 7	47.55 12	90.30 5
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
CasMVSNet(SR_B)		60.35 12	58.12 20	61.84 10	34.02 22	63.53 11	71.26 5	50.74 9	82.22 15

TAPA-MVS	78.00 13	57.53 13	59.15 17	56.45 13	33.25 25	72.19 6	55.89 18	41.27 20	85.06 11
Andrea Romanoni, Matteo Matteucci: TAPA-MVS: Textureless-Aware PAtchMatch Multi-View Stereo. ICCV 2019
TAPA-MVS(SR)		57.42 14	61.79 13	54.51 14	38.52 19	57.98 16	58.15 16	47.40 13	85.06 11

PCF-MVS	76.59 14	57.03 15	62.55 12	53.34 16	46.44 11	56.38 19	62.25 12	41.39 18	78.67 17
Andreas Kuhn, Shan Lin, Oliver Erdler: Plane Completion and Filtering for Multi-View Stereo Reconstruction. GCPR 2019
COLMAP(base)		54.85 16	60.50 15	51.08 17	45.13 12	57.58 18	53.68 19	41.99 15	75.86 26

ACMM	80.67 7	54.69 17	55.75 25	53.98 15	33.39 24	64.20 10	51.32 21	46.41 14	78.11 21
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
LPCS		54.43 18	63.80 11	48.19 19	49.50 9	57.64 17	46.89 24	40.04 22	78.09 22

ACMP	80.00 8	53.70 19	59.82 16	49.62 18	41.23 14	58.13 15	48.91 23	41.82 16	78.40 19
Qingshan Xu and Wenbing Tao: Planar Prior Assisted PatchMatch Multi-View Stereo. AAAI 2020
GSE		52.96 20	65.52 9	44.58 27	47.61 10	51.37 23	45.62 27	36.75 26	83.44 14

PLC	76.06 15	51.83 21	58.21 19	47.57 20	41.31 13	53.45 21	48.97 22	40.30 21	75.12 28
Jie Liao, Yanping Fu, Qingan Yan, Chunxia xiao: Pyramid Multi-View Stereo with Local Consistency. Pacific Graphics 2019
COLMAP(SR)		51.67 22	59.12 18	46.71 22	40.55 15	51.72 22	52.60 20	35.80 27	77.70 23

PMVS	79.51 9	51.62 23	61.72 14	44.90 26	39.36 18	36.39 28	59.11 14	39.19 23	84.08 13
Y. Furukawa, J. Ponce: Accurate, dense, and robust multiview stereopsis. PAMI (2010)
ACMH+	79.05 11	51.57 24	58.12 20	47.21 21	40.39 16	54.48 20	45.83 26	41.33 19	75.84 27

ACMH	78.40 12	50.98 25	58.12 20	46.22 24	38.05 20	50.36 24	46.62 25	41.67 17	78.18 20
Qingshan Xu and Wenbing Tao: Multi-Scale Geometric Consistency Guided Multi-View Stereo. CVPR 2019
BP-MVSNet		49.12 26	54.86 26	45.30 25	32.80 26	40.86 26	57.05 17	37.99 24	76.92 24
Christian Sormann, Patrick KnÃ¶belreiter, Andreas Kuhn, Mattia Rossi, Thomas Pock, Friedrich Fraundorfer: BP-MVSNet: Belief-Propagation-Layers for Multi-View-Stereo. 3DV 2020
CIDER		48.39 27	51.15 28	46.56 23	23.77 27	44.18 25	58.74 15	36.76 25	78.52 18
Qingshan Xu and Wenbing Tao: Learning Inverse Depth Regression for Multi-View Stereo with Correlation Cost Volume. AAAI 2020
A-TVSNet + Gipuma		45.32 28	54.45 27	39.23 28	23.11 28	38.96 27	45.44 28	33.28 28	85.79 10

unsupervisedMVS_cas		27.71 29	27.25 33	28.02 29	17.77 29	22.21 29	32.37 29	29.49 29	36.73 33

hgnet		20.23 30	34.94 30	10.42 31	4.69 31	8.06 31	20.25 31	2.95 32	65.18 30

DPSNet		20.23 30	34.94 30	10.42 31	4.69 31	8.06 31	20.25 31	2.95 32	65.18 30

example		20.20 32	33.62 32	11.26 30	5.50 30	5.26 33	22.01 30	6.49 31	61.74 32

CMPMVS	55.74 18	14.63 33	35.74 29	0.55 34	0.00 34	0.00 34	1.66 34	0.00 34	71.48 29
M. Jancosek, T. Pajdla: Multi-View Reconstruction Preserving Weakly-Supported Surfaces. CVPR 2011
MVE	41.12 19	13.33 34	18.95 34	9.58 33	3.37 33	11.70 30	6.32 33	10.73 30	34.53 34
Simon Fuhrmann, Fabian Langguth, Michael Goesele: MVE - A Multi-View Reconstruction Environment. EUROGRAPHICS Workshops on Graphics and Cultural Heritage (2014)
dnet		0.00 35	0.00 35	0.00 35	0.00 34	0.00 34	0.00 35	0.00 34	0.00 35

Low-resolution many-view benchmark