Method

Deep learning based stereo matching method using sing-view videos [SMV]
[Anonymous Submission]

Submitted on 28 Jan. 2019 11:41 by
[Anonymous Submission]

Running time:1.6 min
Environment:8 cores @ 3.5 Ghz (Python)

Method Description:
This paper proposes an unsupervised approach to construct a deep learning based stereo matching method using sing-view videos (SMV). From videos, a set of corresponding points are computed between images, and image patches that center at the computed points are extracted. Negative and positive samples constitute a dataset to train a similarity network that is then used as a matching cost function.

In addition, we propose a local-global matching cost network that exploits the first feature maps (local features) accompanying with last feature features (global features) as output feature of the proposed network. The concatenated features are connected to full-connected layers and the network outputs a similarity measure of an image patch pair as a matching cost. Computed matching costs are aggregated using semi-global matching and cross-based cost aggregation, followed by sub-pixel interpolation, left-right consistency check, median and bilateral filtering.

We evaluate the proposed stereo matching methods using popular stereo matching datasets, including KITTI 2012 and 2015, and Middlebury. We submit the disparity maps to their benchmark servers to evaluate the performance of SMV. We also compared the generalization of SMV and baseline methods using the training sets of the three datasets.
Parameters:
Network Parameters
input_patch_size=11x11
average_pooling_size=7x7
ckernel_size=3
num_clayers=5
num_fc_layers=3
num_fmaps=112
num_fc_units=384
Latex Bibtex:

Detailed Results

This page provides detailed results for the method(s) selected. For each of the first 20 test images, the number of erroneous pixels at all thresholds is depicted in the table. Underneath, the left input image, the disparity / end-point error map and the estimated (and interpolated) disparity / optical flow map are shown. The error map scales linearly between 0 (black) and >=5 (white) pixels error. Red denotes all occluded pixels, falling outside the image boundaries. The false color map is scaled to the largest ground truth disparity / flow value.

Test Set Average

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 7.92 % 9.75 % 1.1 px 1.3 px
3 pixels 4.23 % 5.77 % 1.1 px 1.3 px
4 pixels 3.06 % 4.26 % 1.1 px 1.3 px
5 pixels 2.54 % 3.48 % 1.1 px 1.3 px
This table as LaTeX

Reflective Regions

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 34.10 % 37.51 % 4.8 px 5.7 px
3 pixels 24.74 % 28.37 % 4.8 px 5.7 px
4 pixels 20.20 % 23.80 % 4.8 px 5.7 px
5 pixels 17.51 % 20.92 % 4.8 px 5.7 px
This table as LaTeX

Test Image 0

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 9.11 % 11.48 % 1.0 px 1.2 px
3 pixels 5.04 % 7.15 % 1.0 px 1.2 px
4 pixels 3.47 % 5.29 % 1.0 px 1.2 px
5 pixels 2.57 % 3.94 % 1.0 px 1.2 px
This table as LaTeX





Test Image 1

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 7.87 % 11.26 % 1.1 px 1.3 px
3 pixels 4.31 % 6.98 % 1.1 px 1.3 px
4 pixels 3.26 % 5.04 % 1.1 px 1.3 px
5 pixels 2.83 % 3.90 % 1.1 px 1.3 px
This table as LaTeX





Test Image 2

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 8.26 % 9.95 % 1.0 px 1.1 px
3 pixels 4.33 % 5.80 % 1.0 px 1.1 px
4 pixels 3.09 % 3.93 % 1.0 px 1.1 px
5 pixels 2.52 % 2.94 % 1.0 px 1.1 px
This table as LaTeX





Test Image 3

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 7.23 % 9.72 % 0.8 px 1.0 px
3 pixels 3.17 % 5.54 % 0.8 px 1.0 px
4 pixels 1.83 % 3.70 % 0.8 px 1.0 px
5 pixels 1.28 % 2.68 % 0.8 px 1.0 px
This table as LaTeX





Test Image 4

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 5.73 % 7.57 % 0.7 px 0.8 px
3 pixels 1.89 % 3.22 % 0.7 px 0.8 px
4 pixels 0.98 % 1.39 % 0.7 px 0.8 px
5 pixels 0.61 % 0.77 % 0.7 px 0.8 px
This table as LaTeX





Test Image 5

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 3.90 % 5.04 % 0.7 px 0.8 px
3 pixels 1.63 % 2.02 % 0.7 px 0.8 px
4 pixels 1.12 % 1.24 % 0.7 px 0.8 px
5 pixels 0.92 % 0.96 % 0.7 px 0.8 px
This table as LaTeX





Test Image 6

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 10.10 % 12.34 % 1.4 px 1.5 px
3 pixels 5.21 % 7.01 % 1.4 px 1.5 px
4 pixels 3.84 % 4.68 % 1.4 px 1.5 px
5 pixels 3.38 % 3.87 % 1.4 px 1.5 px
This table as LaTeX





Test Image 7

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 9.66 % 11.90 % 1.3 px 1.6 px
3 pixels 5.83 % 8.12 % 1.3 px 1.6 px
4 pixels 4.48 % 6.51 % 1.3 px 1.6 px
5 pixels 3.52 % 5.22 % 1.3 px 1.6 px
This table as LaTeX





Test Image 8

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 4.55 % 6.35 % 0.7 px 0.8 px
3 pixels 1.39 % 2.10 % 0.7 px 0.8 px
4 pixels 0.87 % 1.38 % 0.7 px 0.8 px
5 pixels 0.73 % 1.18 % 0.7 px 0.8 px
This table as LaTeX





Test Image 9

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 15.11 % 17.22 % 1.7 px 2.0 px
3 pixels 9.21 % 11.46 % 1.7 px 2.0 px
4 pixels 6.37 % 8.69 % 1.7 px 2.0 px
5 pixels 4.87 % 7.11 % 1.7 px 2.0 px
This table as LaTeX





Test Image 10

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 3.96 % 5.63 % 0.6 px 0.7 px
3 pixels 1.72 % 3.11 % 0.6 px 0.7 px
4 pixels 1.10 % 2.14 % 0.6 px 0.7 px
5 pixels 0.81 % 1.60 % 0.6 px 0.7 px
This table as LaTeX





Test Image 11

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 12.90 % 16.10 % 1.8 px 1.9 px
3 pixels 6.85 % 10.29 % 1.8 px 1.9 px
4 pixels 5.07 % 8.30 % 1.8 px 1.9 px
5 pixels 4.32 % 7.05 % 1.8 px 1.9 px
This table as LaTeX





Test Image 12

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 7.99 % 10.16 % 1.5 px 1.7 px
3 pixels 5.53 % 7.44 % 1.5 px 1.7 px
4 pixels 4.18 % 5.71 % 1.5 px 1.7 px
5 pixels 3.74 % 5.07 % 1.5 px 1.7 px
This table as LaTeX





Test Image 13

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 9.09 % 10.38 % 1.3 px 1.5 px
3 pixels 6.01 % 7.29 % 1.3 px 1.5 px
4 pixels 4.81 % 6.03 % 1.3 px 1.5 px
5 pixels 3.81 % 4.96 % 1.3 px 1.5 px
This table as LaTeX





Test Image 14

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 6.18 % 7.67 % 0.8 px 0.8 px
3 pixels 2.67 % 3.82 % 0.8 px 0.8 px
4 pixels 1.79 % 2.75 % 0.8 px 0.8 px
5 pixels 1.31 % 2.21 % 0.8 px 0.8 px
This table as LaTeX





Test Image 15

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 8.64 % 10.34 % 1.0 px 1.1 px
3 pixels 4.55 % 5.27 % 1.0 px 1.1 px
4 pixels 2.81 % 3.03 % 1.0 px 1.1 px
5 pixels 2.18 % 2.22 % 1.0 px 1.1 px
This table as LaTeX





Test Image 16

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 7.07 % 7.48 % 0.9 px 0.9 px
3 pixels 1.63 % 1.76 % 0.9 px 0.9 px
4 pixels 0.58 % 0.57 % 0.9 px 0.9 px
5 pixels 0.39 % 0.39 % 0.9 px 0.9 px
This table as LaTeX





Test Image 17

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 10.46 % 11.59 % 1.4 px 1.4 px
3 pixels 6.19 % 6.80 % 1.4 px 1.4 px
4 pixels 4.36 % 4.71 % 1.4 px 1.4 px
5 pixels 3.35 % 3.65 % 1.4 px 1.4 px
This table as LaTeX





Test Image 18

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 4.30 % 4.94 % 0.6 px 0.6 px
3 pixels 1.18 % 1.51 % 0.6 px 0.6 px
4 pixels 0.55 % 0.64 % 0.6 px 0.6 px
5 pixels 0.37 % 0.39 % 0.6 px 0.6 px
This table as LaTeX





Test Image 19

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 4.60 % 7.36 % 0.8 px 0.9 px
3 pixels 2.26 % 3.31 % 0.8 px 0.9 px
4 pixels 1.60 % 1.68 % 0.8 px 0.9 px
5 pixels 1.30 % 1.27 % 0.8 px 0.9 px
This table as LaTeX







eXTReMe Tracker