Snapshot of 9/14/2015 before updating ELAS results. The old ELAS results were worse because of a bug in the conversion to grayscale images (using ImageMagick's "convert" under linux). The conversion is now done via the code in the SDK 1.6.
test densetest sparsetraining densetraining sparse
Metric:
bad 0.5bad 1.0bad 2.0bad 4.0avgerrrmsA50A90A95A99timetime/MPtime/GD
Mask:
nonoccall
Reset sort
ReferenceDescriptionParametersRunning Environment
Loading...
[stat] error
bad 2.0 (%)
Weight
Date
Name
Res
Avg
Austr
AustrP
Bicyc2
Class
ClassE
Compu
Crusa
CrusaP
Djemb
DjembL
Hoops
Livgrm
Nkuba
Plants
Stairs
MP: 5.6
nd: 290
im0
im1
GT
nonocc
MP: 5.6
nd: 290
im0
im1
GT
nonocc
MP: 5.6
nd: 250
im0
im1
GT
nonocc
MP: 5.7
nd: 610
im0
im1
GT
nonocc
MP: 5.7
nd: 610
im0
im1
GT
nonocc
MP: 1.5
nd: 256
im0
im1
GT
nonocc
MP: 5.5
nd: 800
im0
im1
GT
nonocc
MP: 5.5
nd: 800
im0
im1
GT
nonocc
MP: 5.7
nd: 320
im0
im1
GT
nonocc
MP: 5.7
nd: 320
im0
im1
GT
nonocc
MP: 5.7
nd: 410
im0
im1
GT
nonocc
MP: 5.9
nd: 320
im0
im1
GT
nonocc
MP: 5.5
nd: 570
im0
im1
GT
nonocc
MP: 5.6
nd: 320
im0
im1
GT
nonocc
MP: 5.2
nd: 450
im0
im1
GT
nonocc
Warning: Undefined variable $ids in /home/vision/public_html/stereo/snap3/2015_09_14/eval3/get_table.php on line 399
OpenCV 2.4.8 StereoSGBM method, full variant (2 passes). Reimplementation of H. Hirschmüller's SGM method (CVPR 2006; PAMI 2008).
OpenCV's "semi-global block matching" method; memory-intensive 2-pass version, which can only handle the quarter-size images. The matching cost is the sum of absolute differences over small windows. Aggregation is performed by dynamic programming along paths in 8 directions. Post filter as implemented in OpenCV. Dense results are created by hole-filling along scanlines.
SAD window: 3x3 pixel
Truncation value for pre-filter: 63
P1/P2: 8*3*3*3/32*3*3*3
Uniqueness ratio: 10
Speckle window size: 100
Speckle range: 32
Full DP: true
C/C++; 1 core, i7@3.3 GHz
07/25/14
1
SGBM2
Q
1
26.6
14
27.9
7
12.1
18
17.8
20
13.7
7
74.5
18
16.2
14
30.3
11
26.3
17
11.0
18
64.4
19
37.9
15
25.8
11
25.3
15
29.3
17
43.7
16
OpenCV 2.4.8 StereoSGBM method, single-pass variant. Reimplementation and modification of H. Hirschmüller's SGM method (CVPR 2006; PAMI 2008).
OpenCV's "semi-global block matching" method; memory efficient single-pass version. The matching cost is the sum of absolute differences over small windows. Aggregation is performed by dynamic programming along paths in only 5 of 8 directions. Post filter as implemented in OpenCV. Dense results are created by hole-filling along scanlines.
SAD window: 3x3 pixel
Truncation value for pre-filter: 63
P1/P2: 8*3*3*3/32*3*3*3
Uniqueness ratio: 10
Speckle window size: 100
Speckle range: 32
Full DP: false
The images are Census transformed and the Hamming distance is used as pixelwise matching cost. Aggregation is performed by a kind of dynamic programming along 8 paths that go from all directions through the image. Small disparity patches are invalidated. Interpolation is also performed along 8 paths.
A prior disparity image is calculated by matching a set of reliable support points and triangulating between them. A maximum a-posterior approach refines the disparities. The disparities for the left and right image are checked for consistency and disparity segments below a size of 50 pixels removed.
Standard parameters of Libelas as provided with the MiddEval3-SDK.
C++/SSE; 1 core, i7@3.3 GHz
07/25/14
4
ELAS
F
3
33.1
21
49.3
19
8.96
12
14.5
18
35.1
19
98.0
23
16.4
15
47.4
21
24.2
15
14.9
21
49.7
14
39.7
19
26.2
12
30.1
19
60.8
23
34.6
11
OpenCV 2.4.8 StereoSGBM method, single-pass variant. Reimplementation and modification of H. Hirschmüller's SGM method (CVPR 2006; PAMI 2008).
OpenCV's "semi-global block matching" method; memory efficient single-pass version. The matching cost is the sum of absolute differences over small windows. Aggregation is performed by dynamic programming along paths in only 5 of 8 directions. Post filter as implemented in OpenCV. Dense results are created by hole-filling along scanlines.
SAD window: 3x3 pixel
Truncation value for pre-filter: 63
P1/P2: 8*3*3*3/32*3*3*3
Uniqueness ratio: 10
Speckle window size: 100
Speckle range: 32
Full DP: false
A prior disparity image is calculated by matching a set of reliable support points and triangulating between them. A maximum a-posterior approach refines the disparities. The disparities for the left and right image are checked for consistency and disparity segments below a size of 50 pixels removed.
Standard parameters of Libelas as provided with the MiddEval3-SDK.
C++/SSE; 1 core, i7@3.3 GHz
07/28/14
7
ELAS
H
2
29.3
18
41.8
14
11.1
16
16.6
19
27.9
17
96.0
22
19.8
17
31.6
12
21.4
12
14.1
20
50.8
15
40.2
20
29.2
19
29.7
18
32.3
20
37.3
14
OpenCV 2.4.8 StereoSGBM method, single-pass variant. Reimplementation and modification of H. Hirschmüller's SGM method (CVPR 2006; PAMI 2008).
OpenCV's "semi-global block matching" method; memory efficient single-pass version. The matching cost is the sum of absolute differences over small windows. Aggregation is performed by dynamic programming along paths in only 5 of 8 directions. Post filter as implemented in OpenCV. Dense results are created by hole-filling along scanlines.
SAD window: 3x3 pixel
Truncation value for pre-filter: 63
P1/P2: 8*3*3*3/32*3*3*3
Uniqueness ratio: 10
Speckle window size: 100
Speckle range: 32
Full DP: false
The images are Census transformed and the Hamming distance is used as pixelwise matching cost. Aggregation is performed by a kind of dynamic programming along 8 paths that go from all directions through the image. Small disparity patches are invalidated. Interpolation is also performed along 8 paths.
The images are Census transformed and the Hamming distance is used as pixelwise matching cost. Aggregation is performed by a kind of dynamic programming along 8 paths that go from all directions through the image. Small disparity patches are invalidated. Interpolation is also performed along 8 paths.
Correlation with five, partly overlapping windows on Census transformed images using Hamming distance as matching cost. A left-right consistency check ensures unique matches and filtering small disparity segments removes outliers. Interpolation is done within image rows with the lowest, valid neighboring disparity.
Census window: 7x7 pixel
Correlation window: 9x9 pixel
LR-check: on
Min. segments: 200 pixel
Interpolation: horizontal, lowest neighbor
A fast method for high-resolution stereo matching without exploring the full search space. Plane hypotheses are generated from sparse feature matches. Around each plane, a local plane sweep with +/- 3 disparities levels is performed to establish local disparity hypotheses via SGM using NCC matching costs. Finally, each pixel is assigned to one hypothesis using global optimization, again using SGM.
nRounds=3
The full set of parameters is listed in the paper and the supplemental materials on the project webpage.
A fast method for high-resolution stereo matching without exploring the full search space. Plane hypotheses are generated from sparse feature matches. Around each plane, a local plane sweep with +/- 3 disparities levels is performed to establish local disparity hypotheses via SGM using NCC matching costs. Finally, each pixel is assigned to one hypothesis using global optimization, again using SGM.
nRounds=3
The full set of parameters is listed in the paper and the supplemental materials on the project webpage.
This approach is an adaptive local stereo-method. It is integrated into a hierarchical scheme, which exploits adaptive windows. Sub-pix disparities are estimated,but not refined.
L = 10
t = 35
medianK = [3 3]
censusK = [9 7]
lambda = 45;
Block-matching stereo with Summed Normalized Cross-Correlation (SNCC) measure. Standard post-processed is applied, including a left-right check, error island removal (region growing), hole-filling and median filtering.
SNCC (first stage 3x3, second stage 11x11)
min correlation threshold = 0.3
region growing threshold = 2.5 disparity
min region size = 200 pixel
median filter = 1x5 and 5x1
Efficient two-pass aggregation with census/gradient cost metric, followed by iterative cost penalization and disparity re-selection to encourage local smoothness of disparities.
census window size = 9 x 7
max census distance = 38.03
max gradient difference = 2.51
census/gradient balance = 0.09
aggregation window size = 33 x 33
aggregation range parameter = 23.39
aggregation spatial parameter = 7.69
refinement window size = 65 x 65
refinement range parameter = 11.30
refinement spatial parameter = 17.20
cost penalty coefficient = 0.0023
median filter window size = 3 x 3
3 iterations of refinement
confidence threshold of 0.1 for sparse maps
CUDA C++, NVIDIA GeForce TITAN Black
10/07/14
16
IDR
H
2
18.4
5
37.5
12
4.08
1
7.49
3
23.3
13
40.6
7
15.7
13
24.5
6
11.3
6
5.46
6
33.1
8
26.0
4
21.5
6
21.7
7
15.3
5
21.2
4
Anonymous. Using local cues to improve dense stereo matching. CVPR 2015 submission 973.
We propose to perform stereo matching as a two-step energy minimization algorithm. We consider two MRF models: a fully connected model defined on the complete set of pixels in an image and a conventional locally connected model. We solve the energy minimization problem for the fully connected model, after which the marginal function of the solution is used as the unary potential in the locally connected MRF model.
Only gradient componenet (6D vector) of color images is used
C++, Intel® Core™ i5-4300U 1.9-GHz CPU
01/21/15
18
TSGO
F
3
39.1
22
34.1
10
16.9
19
20.0
22
43.3
21
55.4
12
14.3
7
54.1
23
49.2
23
33.9
23
66.2
21
45.9
22
39.8
22
42.6
22
47.2
22
52.6
21
Anonymous. Recursive edge-aware filters for stereo matching. CVPR Embedded Vision Workshop 2015, submission ID 9.
A local matching technique utilizing SAD+Census cost measure and a recursive edge-aware aggregation through Successive Weighted Summation. Occlusion handling is provided via left-right cross check and a background favored filling.
smoothness parameter sigma = 24
5x5 Census window, Census weight=0.7, SAD weight=0.3, occlusion threshold=2
Efficient message passing on minimum spanning trees to acquire the maximum a posteriori disparity estimates.
Gradient Coefficient 0.289
Gradient Truncation 1.0
Census Coefficient 0.014
Census Truncation 20
Census Window 7x9
Matlab; i7 core @2.7GHz
04/17/15
21
TMAP
H
2
17.1
4
20.2
5
4.94
6
8.13
5
12.8
5
30.0
4
14.1
6
27.9
10
20.4
11
5.09
2
31.5
7
23.1
2
20.9
4
19.0
3
18.8
10
18.0
2
Anonymous. MeshStereo: A global stereo model with mesh alignment regularization for view interpolation. ICCV 2015 submission 2636.
i7-2600 3.40GHz 8 cores, C++
04/20/15
22
MeshStereo
H
2
13.4
2
5.90
2
4.88
5
10.8
9
12.9
6
10.6
1
13.6
3
12.2
4
9.01
2
5.39
4
27.4
3
23.5
3
17.7
2
21.0
6
15.4
6
20.9
3
J. Žbontar and Y. LeCun. Computing the stereo matching cost with a convolutional neural network. A previous version appeared in CVPR 2015. An extended version will be submitted to JMLR.
Compute the matching cost with a convolutional neural network. Then apply cross-based cost aggregation, semiglobal matching, left-right consistency check, median filter, and a bilateral filter.
DETAILS:
The network is similar to the one described in our CVPR paper differing only in the values of some hyperparameters. The input to the network are two 11 x 11 image patches. Five convolutional layers with 3 x 3 kernels and 112 feature maps extract feature vectors from the input image patches. The two 112-length feature vectors are concatenated into a 224-length vector which is passed through three fully-connected layers with 384 units each. The final (fourth) fully-connected layer projects the output to a single number---the matching cost. One important addition was the use of data augmentation techniques to increase the size of the training set. We tried to use as much training data as possible. Therefore we combined all of the 2001, 2003, 2005, 2006, and 2014 Middlebury datasets obtaining 60 image pairs. For the newer datasets (2005, 2006, and 2014) we also used several illumination and exposure settings.