Low-res two-view results - ETH3D Benchmark

This table lists the benchmark results for the low-res two-view scenario. This benchmark evaluates the Middlebury stereo metrics (for all metrics, smaller is better):

bad 0.5, 1.0, 2.0, 4.0 [%]: Fraction of pixels with errors larger than the given number of disparities.
Average error [px]: The per-pixel average disparity error.
Root mean square error [px]: The per-pixel root mean square disparity error.
50%, 90%, 95%, 99% error quantile [px]: The highest disparity error within the given percentage of best pixels (for 50%, this is the median error).
Time [s]: The runtime of the method.

The mask determines whether the metric is evaluated for all pixels with ground truth, or only for pixels which are visible in both images (non-occluded).
The coverage selector allows to limit the table to results for all pixels (dense), or a given minimum fraction of pixels.

Methods with suffix _ROB may participate in the Robust Vision Challenge.

Click one or more dataset result cells or column headers to show visualizations. Most visualizations are only available for training datasets. The visualizations may not work with mobile browsers.

Method	all	lakes. 1l	lakes. 1s	sand box 1l	sand box 1s	stora. room 1l	stora. room 1s	stora. room 2l	stora. room 2s	stora. room 2 1l	stora. room 2 1s	stora. room 2 2l	stora. room 2 2s	stora. room 3l	stora. room 3s	tunnel 1l	tunnel 1s	tunnel 2l	tunnel 2s	tunnel 3l	tunnel 3s

DN-CSS_ROB	0.64 1	0.32 2	1.05 12	0.55 5	0.40 12	0.54 1	0.45 3	1.21 1	0.93 1	1.03 2	1.47 16	0.69 1	0.63 1	1.76 12	0.49 1	0.18 1	0.14 2	0.25 9	0.22 9	0.22 5	0.20 4

MLCV	0.72 2	0.34 4	1.10 15	0.51 1	0.23 1	0.91 3	0.39 2	1.33 2	1.23 6	1.46 13	1.55 19	1.01 3	0.64 2	1.95 16	0.80 8	0.18 1	0.12 1	0.18 3	0.14 3	0.19 2	0.17 1

iResNet	0.75 3	0.40 11	1.36 26	0.67 16	0.29 2	0.92 4	0.59 8	1.52 4	1.25 9	1.28 7	1.48 17	0.94 2	0.76 4	1.62 7	0.78 6	0.18 1	0.14 2	0.18 3	0.15 5	0.22 5	0.23 7

iResNet_ROB	0.77 4	0.30 1	0.99 11	0.57 8	0.36 6	0.83 2	0.36 1	2.33 32	1.58 24	1.27 6	1.30 6	1.25 6	0.84 5	1.65 8	0.79 7	0.20 5	0.14 2	0.17 2	0.12 1	0.18 1	0.27 9

iResNetv2_ROB	0.80 5	0.37 8	1.71 34	0.56 6	0.37 7	0.94 5	0.64 10	1.73 11	1.34 12	1.23 5	1.42 12	1.26 7	0.92 9	1.85 15	0.56 2	0.23 9	0.15 5	0.19 5	0.14 3	0.24 7	0.19 3

StereoDRNet-Refined	0.80 5	0.41 13	0.88 8	0.62 12	0.42 14	1.33 15	0.48 4	1.52 4	1.17 3	1.45 12	1.44 14	1.44 12	0.90 7	1.26 1	1.24 14	0.20 5	0.16 7	0.27 11	0.24 10	0.25 8	0.28 12

DLCB_ROB	0.86 7	0.40 11	0.78 7	0.68 19	0.54 22	1.11 7	0.84 14	1.49 3	1.23 6	1.64 17	1.65 21	1.70 21	0.91 8	1.70 10	1.05 11	0.25 11	0.24 14	0.26 10	0.27 14	0.27 10	0.24 8

PSMNet_ROB	0.90 8	0.52 22	1.17 21	0.72 27	0.61 28	1.27 13	1.12 23	1.76 14	1.12 2	1.09 3	1.28 5	1.45 13	0.89 6	1.65 8	1.42 21	0.33 18	0.25 15	0.37 21	0.39 27	0.35 17	0.29 14

CBMV_ROB	0.90 8	0.44 15	0.63 4	0.51 1	0.37 7	1.25 12	0.49 5	1.77 15	1.27 10	1.41 11	2.00 24	1.33 8	1.08 19	2.17 17	0.76 5	0.43 27	0.42 32	0.47 31	0.44 31	0.41 24	0.36 21

CBMV	0.95 10	0.44 15	0.66 6	0.53 4	0.35 5	1.53 19	1.63 36	1.69 8	1.42 18	2.00 24	1.60 20	1.43 10	1.17 22	1.59 5	0.99 10	0.34 20	0.34 27	0.39 24	0.33 21	0.33 16	0.32 17
Konstantinos Batsos, Changjiang Cai, Philippos Mordohai: CBMV: A Coalesced Bidirectional Matching Volume for Disparity Estimation. Computer Vision and Pattern Recognition (CVPR) 2018
ETE_ROB	0.96 11	0.64 32	1.16 19	0.69 20	0.48 17	1.14 9	1.13 24	1.77 15	1.19 4	1.91 22	1.45 15	1.73 22	1.05 16	1.60 6	1.14 12	0.32 17	0.26 17	0.38 23	0.34 23	0.39 22	0.41 26

XPNet_ROB	0.96 11	0.49 21	1.09 14	0.72 27	0.57 23	1.11 7	1.14 25	1.66 7	1.23 6	1.85 21	1.39 11	1.64 19	1.11 21	1.80 14	1.42 21	0.38 26	0.30 21	0.33 19	0.29 18	0.37 19	0.34 20

StereoDRNet	0.98 13	0.47 19	1.70 33	0.76 32	0.94 36	1.51 18	0.91 16	1.92 24	1.32 11	1.61 16	1.13 1	1.47 15	0.97 11	1.75 11	1.46 24	0.28 13	0.23 13	0.29 13	0.27 14	0.35 17	0.27 9

NCCL2	1.00 14	0.56 24	1.06 13	0.75 31	0.57 23	1.31 14	1.72 37	1.77 15	1.36 13	1.50 15	1.17 2	1.91 23	1.17 22	1.54 3	1.29 16	0.35 23	0.31 23	0.45 29	0.44 31	0.40 23	0.40 25

DRN-Test	1.00 14	0.44 15	1.13 17	0.70 23	0.81 32	1.62 23	0.79 13	2.14 29	1.44 19	1.78 19	1.33 7	1.43 10	1.05 16	2.18 18	1.53 27	0.27 12	0.21 11	0.31 17	0.30 20	0.32 15	0.28 12

LALA_ROB	1.01 16	0.59 27	1.16 19	0.67 16	0.51 21	1.63 24	1.26 29	1.69 8	1.38 15	1.81 20	1.37 9	1.94 25	1.02 14	1.58 4	1.38 20	0.37 25	0.27 19	0.43 28	0.37 25	0.43 25	0.37 22

DISCO	1.03 17	0.34 4	1.11 16	0.69 20	0.45 16	2.07 31	0.84 14	2.13 28	1.50 21	1.15 4	1.33 7	2.40 30	1.07 18	2.60 21	1.51 26	0.23 9	0.18 9	0.22 8	0.20 7	0.31 13	0.27 9

NaN_ROB	1.13 18	0.55 23	1.15 18	0.64 15	0.48 17	1.61 22	2.03 39	1.72 10	1.48 20	2.09 25	1.27 4	1.40 9	1.08 19	3.94 27	1.30 17	0.28 13	0.34 27	0.28 12	0.28 16	0.30 11	0.33 19

HSM	1.19 19	0.32 2	0.65 5	0.57 8	0.38 10	1.23 11	0.59 8	2.25 31	1.19 4	1.30 9	1.25 3	1.94 25	1.03 15	9.32 41	0.71 4	0.22 8	0.16 7	0.20 6	0.16 6	0.21 4	0.21 5

PWC_ROB	1.24 20	0.76 37	2.46 38	0.72 27	0.78 31	1.15 10	0.49 5	1.75 13	1.92 34	2.31 28	3.07 31	1.59 17	1.40 25	2.46 19	2.10 36	0.33 18	0.22 12	0.35 20	0.24 10	0.38 21	0.41 26

SGM-Forest	1.24 20	0.38 10	0.60 1	0.52 3	0.37 7	1.56 21	1.30 30	1.65 6	1.36 13	2.11 26	1.79 23	1.63 18	0.93 10	7.48 37	0.95 9	0.36 24	0.37 30	0.39 24	0.34 23	0.37 19	0.32 17
Johannes L. Schönberger, Sudipta Sinha, Marc Pollefeys: Learning to Fuse Proposals from Multiple Scanline Optimizations in Semi-Global Matching. ECCV 2018
NOSS_ROB	1.26 22	0.46 18	0.62 3	0.56 6	0.38 10	1.06 6	0.78 12	1.73 11	1.40 16	1.01 1	1.43 13	1.10 4	0.75 3	10.62 44	0.65 3	0.45 30	0.43 33	0.48 32	0.45 33	0.44 28	0.41 26

PWCDC_ROB	1.29 23	0.75 36	1.33 25	1.05 35	0.88 33	1.73 25	0.52 7	1.89 22	3.28 38	1.29 8	4.88 38	1.22 5	1.46 27	1.77 13	1.45 23	0.60 34	0.26 17	0.29 13	0.25 12	0.52 32	0.37 22

PDISCO_ROB	1.32 24	0.65 33	1.44 27	1.36 40	1.46 42	2.55 35	0.92 17	2.45 35	2.07 37	1.37 10	1.49 18	1.65 20	1.00 13	2.82 22	1.35 19	1.01 38	0.27 19	0.54 33	0.47 34	0.71 37	0.73 37

MFMNet_re	1.40 25	1.14 41	1.68 32	1.39 41	1.07 37	1.48 17	1.43 32	1.87 20	1.82 32	2.26 27	2.19 26	1.51 16	1.86 30	1.52 2	1.25 15	0.93 37	0.83 39	0.84 38	0.80 38	0.99 39	1.05 40

CSAN	1.41 26	0.59 27	1.26 23	0.70 23	0.50 20	1.76 26	1.78 38	1.77 15	1.90 33	2.35 29	2.32 27	1.92 24	1.94 31	5.56 33	1.49 25	0.43 27	0.35 29	0.40 27	0.41 29	0.43 25	0.41 26

SPS-STEREO	1.67 27	1.04 38	1.23 22	1.21 39	1.13 40	1.82 28	1.11 21	2.16 30	1.52 22	3.07 34	2.32 27	2.82 31	1.80 28	4.11 29	2.04 35	1.03 40	0.99 42	0.97 39	0.94 39	1.03 41	1.02 38
K. Yamaguchi, D. McAllester, R. Urtasun: Efficient Joint Segmentation, Occlusion Labeling, Stereo and Flow Estimation. ECCV 2014
MDST_ROB	1.67 27	0.34 4	1.58 31	0.92 33	0.58 27	4.26 42	1.11 21	2.83 38	1.41 17	4.49 38	2.55 29	1.45 13	0.97 11	7.79 38	1.17 13	0.31 15	0.25 15	0.37 21	0.33 21	0.31 13	0.30 16

SGM_ROB	1.68 29	0.36 7	0.61 2	0.59 10	0.29 2	2.25 34	1.18 27	2.38 33	1.68 26	3.60 35	2.67 30	3.69 35	2.42 33	8.13 39	1.90 32	0.31 15	0.31 23	0.30 16	0.28 16	0.30 11	0.29 14
Heiko Hirschmueller: Stereo processing by semiglobal matching and mutual information. TPAMI 2008, Volume 30(2), pp. 328-341
DPSNet	1.69 30	0.59 27	4.33 41	0.70 23	0.57 23	2.02 30	1.62 35	3.35 42	1.78 30	1.48 14	1.38 10	2.17 28	2.84 34	4.77 31	3.66 40	0.48 31	0.31 23	0.29 13	0.25 12	0.64 35	0.48 33

FBW_ROB	1.71 31	0.62 31	1.45 28	0.67 16	0.61 28	1.81 27	1.08 20	2.52 36	1.70 27	1.77 18	1.73 22	1.98 27	1.25 24	12.57 45	1.84 31	0.34 20	0.37 30	0.56 36	0.38 26	0.45 29	0.43 31

WCMA_ROB	1.84 32	0.47 19	1.31 24	0.69 20	0.57 23	2.11 32	1.14 25	1.87 20	1.73 29	4.04 36	6.43 40	5.62 39	3.61 37	2.59 20	1.94 33	0.52 32	0.44 35	0.39 24	0.42 30	0.48 30	0.47 32

SANet	1.88 33	0.57 26	2.18 37	0.63 13	0.40 12	2.80 39	1.57 34	2.39 34	5.88 42	2.36 30	3.57 32	3.72 36	2.97 35	4.74 30	1.76 30	0.34 20	0.30 21	0.32 18	0.29 18	0.48 30	0.37 22

NVStereoNet_ROB	1.88 33	1.07 39	1.45 28	1.07 37	1.17 41	1.47 16	1.48 33	1.89 22	1.93 35	1.94 23	4.67 36	3.08 32	1.80 28	6.51 35	1.65 29	1.06 42	0.86 40	1.08 41	1.32 42	0.93 38	1.26 42
Nikolai Smolyanskiy, Alexey Kamenev, Stan Birchfield: On the Importance of Stereo for Accurate Depth Estimation: An Efficient Semi-Supervised Deep Neural Network Approach. Arxiv
MSMD_ROB	1.99 35	0.59 27	0.89 9	0.70 23	0.61 28	3.95 41	0.94 18	1.80 19	1.79 31	4.41 37	4.43 34	8.17 41	3.66 38	3.40 23	1.33 18	0.52 32	0.48 36	0.54 33	0.58 37	0.53 33	0.50 34

MeshStereo	2.26 36	0.56 24	0.92 10	0.61 11	0.49 19	2.60 36	1.02 19	3.18 40	2.05 36	7.40 44	4.98 39	7.58 40	2.98 36	6.37 34	1.96 34	0.43 27	0.43 33	0.45 29	0.39 27	0.43 25	0.42 30
C. Zhang, Z. Li, Y. Cheng, R. Cai, H. Chao, Y. Rui: MeshStereo: A Global Stereo Model with Mesh Alignment Regularization for View Interpolation. ICCV 2015
DispFullNet	2.34 37	2.85 43	3.66 40	3.08 44	2.27 43	1.97 29	0.67 11	2.00 26	1.70 27	2.52 31	2.00 24	2.23 29	5.86 41	3.44 25	1.58 28	1.01 38	0.32 26	2.64 43	1.94 43	2.85 43	2.24 43

pmcnn	2.37 38	0.42 14	2.71 39	0.63 13	0.43 15	1.54 20	1.19 28	2.01 27	1.53 23	2.64 32	7.60 42	18.07 46	1.45 26	3.44 25	2.78 38	0.18 1	0.15 5	0.15 1	0.12 1	0.20 3	0.18 2

ELAS	2.37 38	0.74 34	1.54 30	0.99 34	0.92 34	4.78 43	2.33 40	3.33 41	3.37 39	5.05 40	4.77 37	3.67 34	4.53 40	3.42 24	4.50 42	0.62 35	0.53 37	0.55 35	0.53 36	0.63 34	0.64 36
A. Geiger, M. Roser, R. Urtasun: Efficient large-scale stereo matching. ACCV 2010
SGM+DAISY	2.41 40	1.07 39	2.14 36	1.18 38	1.10 38	2.14 33	1.36 31	1.96 25	1.66 25	5.46 41	7.07 41	4.95 38	4.32 39	5.50 32	2.26 37	1.04 41	1.02 43	0.99 40	0.97 40	1.02 40	1.04 39

ELAS_ROB	2.48 41	0.74 34	1.90 35	1.05 35	0.92 34	2.76 37	4.01 43	2.76 37	4.11 41	4.69 39	4.63 35	4.62 37	5.96 42	3.97 28	3.96 41	0.63 36	0.53 37	0.56 36	0.52 35	0.64 35	0.63 35
A. Geiger, M. Roser, R. Urtasun: Efficient large-scale stereo matching. ACCV 2010
PWCK	2.88 42	2.05 42	5.72 42	2.08 42	1.11 39	3.55 40	3.90 42	2.86 39	3.54 40	3.03 33	3.63 33	3.10 33	2.31 32	8.15 40	3.09 39	2.12 43	0.91 41	1.75 42	0.98 41	2.57 42	1.18 41

LE_ROB	8.05 43	0.37 8	23.53 45	0.74 30	0.33 4	2.77 38	2.87 41	3.71 43	27.24 48	25.39 48	8.87 43	8.46 42	26.71 47	6.65 36	21.99 44	0.21 7	0.19 10	0.20 6	0.21 8	0.25 8	0.22 6

DPSimNet_ROB	11.20 44	4.98 44	35.43 48	2.90 43	4.52 44	7.89 44	17.09 44	4.66 44	6.06 43	14.89 47	10.93 44	17.29 45	36.81 49	13.18 46	15.68 43	2.86 44	6.29 44	5.75 46	5.09 44	3.91 44	7.70 44

DGTPSM_ROB	13.01 45	8.23 45	21.87 43	9.51 45	18.10 45	11.56 45	32.01 47	10.63 45	16.20 44	7.03 42	19.48 47	10.15 43	16.66 43	9.83 42	22.42 45	4.75 45	8.68 45	5.37 44	10.25 45	6.16 45	11.36 45

DPSMNet_ROB	13.02 46	8.23 45	21.91 44	9.53 46	18.11 46	11.56 45	32.02 48	10.64 46	16.23 45	7.05 43	19.49 48	10.16 44	16.66 43	9.85 43	22.43 46	4.77 46	8.68 45	5.38 45	10.26 46	6.17 46	11.36 45

DPSM	19.08 47	19.14 47	25.17 46	22.15 47	20.85 47	28.17 47	37.82 49	26.56 47	16.85 46	14.14 45	14.84 45	21.70 47	18.75 45	21.87 47	26.55 47	10.16 47	9.30 47	10.47 47	10.42 47	14.03 47	12.61 47

DPSM_ROB	19.08 47	19.14 47	25.17 46	22.15 47	20.85 47	28.17 47	37.82 49	26.56 47	16.85 46	14.14 45	14.84 45	21.70 47	18.75 45	21.87 47	26.55 47	10.16 47	9.30 47	10.47 47	10.42 47	14.03 47	12.61 47

AVERAGE_ROB	40.43 49	42.37 49	42.73 49	36.96 50	35.58 50	30.50 49	27.61 45	36.71 50	38.39 49	44.91 49	44.36 49	37.45 50	36.70 48	41.88 49	40.35 49	45.43 49	46.81 49	47.11 49	46.35 49	42.51 49	43.87 49

MEDIAN_ROB	42.00 50	44.40 50	44.29 50	35.15 49	34.95 49	31.28 50	29.60 46	35.88 49	40.02 50	47.20 50	46.64 50	37.37 49	38.48 50	44.35 50	42.45 50	48.03 50	49.39 50	49.71 50	49.03 50	45.14 50	46.65 50

Low-res two-view benchmark