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):

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 Infoalllakes. 1llakes. 1ssand box 1lsand box 1sstora. room 1lstora. room 1sstora. room 2lstora. room 2sstora. room 2 1lstora. room 2 1sstora. room 2 2lstora. room 2 2sstora. room 3lstora. room 3stunnel 1ltunnel 1stunnel 2ltunnel 2stunnel 3ltunnel 3s
sorted bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort by
DN-CSS_ROBtwo views0.43
1
0.26
3
0.62
9
0.40
4
0.29
7
0.44
1
0.27
1
0.77
1
0.75
1
0.72
2
1.09
6
0.48
1
0.43
1
0.69
1
0.37
1
0.15
1
0.12
1
0.21
4
0.19
4
0.18
3
0.16
2
iResNet_ROBtwo views0.52
2
0.21
1
0.53
7
0.37
1
0.21
1
0.61
2
0.29
3
1.51
24
1.32
20
0.90
4
1.02
4
0.80
3
0.62
3
0.76
2
0.43
2
0.16
3
0.12
1
0.13
2
0.09
1
0.15
1
0.18
4
NOSS_ROBtwo views0.57
3
0.39
14
0.39
2
0.44
9
0.30
8
0.73
3
0.54
8
1.03
4
1.08
10
0.67
1
0.87
1
0.76
2
0.57
2
0.77
3
0.47
3
0.38
22
0.38
26
0.44
23
0.42
25
0.37
23
0.34
21
DLCB_ROBtwo views0.58
4
0.28
5
0.47
6
0.49
12
0.32
11
0.77
5
0.50
7
0.99
2
0.92
2
1.04
8
1.14
8
1.21
12
0.69
6
0.88
6
0.69
8
0.20
5
0.20
7
0.22
5
0.22
7
0.21
5
0.18
4
CBMV_ROBtwo views0.59
5
0.37
12
0.38
1
0.39
2
0.28
6
0.74
4
0.28
2
1.00
3
0.98
6
0.94
5
1.25
14
0.81
4
0.78
9
0.87
5
0.50
4
0.36
21
0.37
25
0.43
22
0.40
23
0.33
16
0.30
16
SGM-Foresttwo views0.62
6
0.32
8
0.39
2
0.43
8
0.30
8
1.02
15
0.55
9
1.09
5
1.05
8
1.06
9
1.18
10
1.10
10
0.67
5
0.88
6
0.62
5
0.31
15
0.33
22
0.33
15
0.30
14
0.30
12
0.28
11
Johannes L. Schönberger, Sudipta Sinha, Marc Pollefeys: Learning to Fuse Proposals from Multiple Scanline Optimizations in Semi-Global Matching. ECCV 2018
PSMNet_ROBtwo views0.64
7
0.37
12
0.65
11
0.56
19
0.42
22
0.97
10
0.83
19
1.21
13
0.92
2
0.79
3
1.02
4
1.08
8
0.63
4
0.82
4
0.91
18
0.29
14
0.20
7
0.31
12
0.31
18
0.26
9
0.23
6
CBMVpermissivetwo views0.64
7
0.35
9
0.41
4
0.42
6
0.25
5
1.01
13
0.77
17
1.13
8
1.09
11
1.17
11
1.22
13
1.04
7
0.75
8
0.88
6
0.63
6
0.28
12
0.30
18
0.34
16
0.30
14
0.26
9
0.26
10
Konstantinos Batsos, Changjiang Cai, Philippos Mordohai: CBMV: A Coalesced Bidirectional Matching Volume for Disparity Estimation. Computer Vision and Pattern Recognition (CVPR) 2018
NaN_ROBtwo views0.69
9
0.44
19
0.69
14
0.51
15
0.33
13
0.97
10
1.06
28
1.22
15
1.19
13
1.24
14
0.96
2
1.08
8
0.78
9
1.20
14
0.67
7
0.24
6
0.30
18
0.23
6
0.24
9
0.23
6
0.25
8
ETE_ROBtwo views0.70
10
0.49
21
0.73
16
0.56
19
0.36
15
0.79
6
0.87
21
1.14
9
0.93
4
1.19
12
1.18
10
1.39
16
0.82
13
0.99
11
0.80
11
0.26
10
0.21
9
0.31
12
0.26
12
0.33
16
0.34
21
XPNet_ROBtwo views0.70
10
0.35
9
0.68
13
0.52
16
0.40
16
0.79
6
0.85
20
1.16
10
0.94
5
1.33
16
1.15
9
1.38
15
0.83
14
1.03
12
0.90
16
0.32
17
0.25
15
0.26
8
0.24
9
0.31
13
0.28
11
NCCL2two views0.72
12
0.41
15
0.65
11
0.64
24
0.44
23
0.98
12
1.20
30
1.11
6
1.00
7
0.98
6
0.96
2
1.49
18
0.87
16
0.90
9
0.79
10
0.31
15
0.26
16
0.39
21
0.39
22
0.32
14
0.32
19
LALA_ROBtwo views0.74
13
0.43
17
0.69
14
0.54
17
0.40
16
1.01
13
0.95
25
1.21
13
1.07
9
1.22
13
1.12
7
1.56
20
0.79
11
1.04
13
0.86
13
0.33
18
0.22
10
0.35
18
0.30
14
0.35
21
0.31
18
CSANtwo views0.83
14
0.49
21
0.77
18
0.60
22
0.41
20
1.14
18
1.06
28
1.19
12
1.46
24
1.32
15
1.43
16
1.37
14
1.19
20
1.27
17
0.86
13
0.39
23
0.31
20
0.36
19
0.37
20
0.35
21
0.34
21
PWC_ROBbinarytwo views0.87
15
0.52
26
1.28
30
0.50
14
0.32
11
0.91
8
0.37
4
1.36
19
1.46
24
1.61
20
2.59
25
1.16
11
1.01
19
1.34
19
1.40
29
0.25
7
0.18
5
0.28
11
0.20
5
0.29
11
0.28
11
FBW_ROBtwo views0.88
16
0.50
25
0.85
20
0.55
18
0.40
16
1.17
19
0.78
18
1.50
23
1.37
21
1.33
16
1.27
15
1.48
17
1.00
18
2.57
32
0.98
20
0.25
7
0.32
21
0.45
24
0.30
14
0.34
18
0.28
11
PWCDC_ROBbinarytwo views0.89
17
0.49
21
0.80
19
0.74
27
0.40
16
1.03
16
0.39
5
1.33
17
2.35
29
1.02
7
3.77
29
0.81
4
0.92
17
1.20
14
0.90
16
0.34
19
0.22
10
0.24
7
0.20
5
0.34
18
0.29
15
MDST_ROBtwo views0.91
18
0.27
4
0.85
20
0.63
23
0.41
20
2.53
31
0.71
15
1.60
27
1.13
12
2.83
28
1.73
18
0.97
6
0.69
6
1.61
22
0.73
9
0.25
7
0.22
10
0.32
14
0.28
13
0.25
7
0.24
7
PDISCO_ROBtwo views0.91
18
0.48
20
1.06
25
0.99
30
0.97
30
1.67
29
0.63
11
1.64
28
1.59
27
1.06
9
1.18
10
1.31
13
0.79
11
1.55
21
0.97
19
0.49
26
0.22
10
0.45
24
0.40
23
0.45
25
0.37
24
SGM_ROBbinarytwo views0.97
20
0.28
5
0.41
4
0.39
2
0.22
2
1.55
25
0.73
16
1.51
24
1.25
18
2.25
24
1.92
20
2.58
26
1.56
23
2.02
28
1.19
25
0.27
11
0.27
17
0.27
10
0.25
11
0.25
7
0.25
8
Heiko Hirschmueller: Stereo processing by semiglobal matching and mutual information. TPAMI 2008, Volume 30(2), pp. 328-341
SANettwo views1.07
21
0.41
15
1.06
25
0.48
11
0.30
8
1.41
22
0.94
24
1.42
21
3.36
33
1.58
18
2.43
23
2.33
24
1.70
25
1.34
19
1.07
23
0.28
12
0.24
14
0.26
8
0.23
8
0.34
18
0.30
16
WCMA_ROBtwo views1.14
22
0.36
11
0.76
17
0.49
12
0.44
23
1.23
20
0.67
12
1.11
6
1.23
16
2.84
29
3.95
33
3.28
30
1.83
27
1.25
16
0.99
21
0.44
24
0.34
23
0.34
16
0.37
20
0.41
24
0.40
25
MSMD_ROBtwo views1.21
23
0.49
21
0.62
9
0.59
21
0.50
25
1.62
27
0.67
12
1.23
16
1.24
17
2.45
26
3.77
29
3.09
29
2.98
31
1.32
18
0.81
12
0.45
25
0.41
27
0.49
26
0.51
28
0.47
26
0.44
26
pmcnntwo views1.23
24
0.29
7
0.96
23
0.40
4
0.23
3
0.95
9
0.69
14
1.16
10
1.28
19
1.67
21
2.33
22
11.15
36
0.86
15
0.93
10
0.88
15
0.15
1
0.12
1
0.12
1
0.10
2
0.15
1
0.15
1
NVStereoNet_ROBtwo views1.24
25
0.90
29
1.13
28
0.86
28
0.88
28
1.11
17
0.99
26
1.42
21
1.53
26
1.59
19
2.48
24
2.24
23
1.53
22
1.96
27
1.30
27
0.76
30
0.70
30
0.78
29
1.00
33
0.71
29
0.88
29
Nikolai Smolyanskiy, Alexey Kamenev, Stan Birchfield: On the Importance of Stereo for Accurate Depth Estimation: An Efficient Semi-Supervised Deep Neural Network Approach. Arxiv
SPS-STEREOcopylefttwo views1.25
26
0.93
30
1.06
25
1.04
31
1.03
32
1.41
22
0.91
23
1.51
24
1.19
13
2.09
23
1.83
19
1.94
21
1.39
21
1.62
23
1.39
28
0.97
31
0.94
32
0.90
30
0.89
31
0.96
31
0.96
31
K. Yamaguchi, D. McAllester, R. Urtasun: Efficient Joint Segmentation, Occlusion Labeling, Stereo and Flow Estimation. ECCV 2014
MeshStereopermissivetwo views1.34
27
0.43
17
0.54
8
0.44
9
0.34
14
1.66
28
0.60
10
1.94
30
1.37
21
4.47
32
3.25
27
4.71
32
1.94
28
1.92
26
1.17
24
0.35
20
0.35
24
0.37
20
0.31
18
0.32
14
0.32
19
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
SGM+DAISYtwo views1.35
28
0.94
31
1.25
29
0.96
29
1.00
31
1.52
24
1.02
27
1.34
18
1.21
15
2.64
27
2.65
26
2.44
25
1.56
23
1.64
25
1.26
26
0.97
31
0.95
33
0.90
30
0.90
32
0.94
30
0.96
31
ELAS_ROBcopylefttwo views1.60
29
0.61
27
1.01
24
0.70
26
0.56
27
2.00
30
1.89
32
1.96
31
2.65
31
3.33
30
3.26
28
3.03
28
3.41
32
2.49
31
2.09
31
0.52
27
0.45
28
0.50
27
0.48
26
0.54
27
0.52
27
A. Geiger, M. Roser, R. Urtasun: Efficient large-scale stereo matching. ACCV 2010
DispFullNettwo views1.60
29
2.63
34
2.75
31
2.56
34
1.79
33
1.24
21
0.47
6
1.37
20
1.45
23
1.87
22
1.50
17
1.55
19
2.46
29
2.04
29
1.04
22
0.59
29
0.18
5
2.32
33
0.68
29
2.29
33
1.14
33
ELAScopylefttwo views1.63
31
0.61
27
0.95
22
0.68
25
0.55
26
3.01
34
1.55
31
2.39
33
2.33
28
3.64
31
3.89
32
2.75
27
2.90
30
2.15
30
2.27
32
0.52
27
0.45
28
0.50
27
0.48
26
0.54
27
0.52
27
A. Geiger, M. Roser, R. Urtasun: Efficient large-scale stereo matching. ACCV 2010
PWCKtwo views1.92
32
1.66
32
3.12
33
1.65
32
0.91
29
2.84
32
2.26
33
2.10
32
2.41
30
2.36
25
2.31
21
2.22
22
1.82
26
3.43
33
2.03
30
1.61
33
0.71
31
1.51
32
0.82
30
1.69
32
0.91
30
DPSimNet_ROBtwo views3.58
33
2.45
33
8.57
34
2.53
33
2.37
34
2.88
33
3.77
34
3.14
34
3.23
32
4.71
33
3.86
31
4.68
31
4.87
33
4.07
34
5.62
33
2.33
34
2.33
34
2.47
34
2.52
34
2.51
34
2.64
34
LE_ROBtwo views4.42
34
0.24
2
2.75
31
0.42
6
0.23
3
1.57
26
0.90
22
1.71
29
17.88
38
16.56
38
4.88
34
5.17
33
18.85
38
1.63
24
14.62
34
0.17
4
0.14
4
0.16
3
0.16
3
0.19
4
0.17
3
DGTPSM_ROBtwo views12.07
35
8.00
35
20.99
35
8.93
35
16.78
35
10.67
35
30.87
37
9.01
35
15.16
34
6.50
34
15.41
37
9.04
34
14.85
34
9.25
35
21.67
35
4.57
35
8.57
35
5.08
35
9.28
35
5.90
35
10.95
35
DPSMNet_ROBtwo views12.08
36
8.00
35
21.00
36
8.97
36
16.79
36
10.67
35
30.89
38
9.02
36
15.17
35
6.51
35
15.41
37
9.04
34
14.85
34
9.27
36
21.67
35
4.58
36
8.57
35
5.09
36
9.29
36
5.90
35
10.95
35
DPSMtwo views17.78
37
18.63
37
24.31
37
20.90
37
19.47
37
25.97
37
36.21
39
21.23
37
16.24
36
13.38
36
14.06
35
19.18
37
16.30
36
20.78
37
25.67
37
9.38
37
9.23
37
9.49
37
9.82
37
13.51
37
11.91
37
DPSM_ROBtwo views17.78
37
18.63
37
24.31
37
20.90
37
19.47
37
25.97
37
36.21
39
21.23
37
16.24
36
13.38
36
14.06
35
19.18
37
16.30
36
20.78
37
25.67
37
9.38
37
9.23
37
9.49
37
9.82
37
13.51
37
11.91
37
MEDIAN_ROBtwo views37.38
39
40.85
40
40.60
40
32.31
39
31.85
39
27.20
39
24.55
35
29.10
39
33.19
39
42.09
40
41.76
40
33.35
39
34.39
39
39.78
40
37.07
39
43.64
40
44.10
40
44.12
40
43.40
40
41.51
40
42.76
40
AVERAGE_ROBtwo views37.58
40
40.13
39
40.15
39
34.81
40
33.82
40
28.55
40
25.10
36
32.02
40
34.06
40
41.28
39
40.66
39
34.46
40
34.81
40
39.18
39
37.57
40
42.78
39
43.44
39
43.82
39
43.03
39
40.47
39
41.51
39