Method

End-to-end Learning of Cost-Volume Aggregation for Real-time Dense Stereo [DeepCostAggr]
[Anonymous Submission]

Submitted on 30 Jan. 2017 20:25 by
[Anonymous Submission]

Running time:0.03 s
Environment:GPU @ 2.5 Ghz (C/C++)

Method Description:
We present a new deep learning-based approach for
dense stereo matching. Compared to previous works, our
approach does not use deep learning of pixel appearance
descriptors, employing very fast classical matching
scores instead. At the same time, our approach uses a
deep convolutional network to predict the local
parameters of cost volume aggregation process, which in
this paper we implement using differentiable domain
transform. By treating such transform as a recurrent
neural network, we are able to train our whole system that
includes cost volume computation, cost-volume
aggregation (smoothing), and winner-takes-all disparity
selection end-to-end. The resulting method is highly
efficient at test time while achieving good matching
accuracy. On the KITTI 2015 benchmark, it achieves a
result of 6.34\% error rate while running at 29 frames per
second rate on a modern GPU.
Parameters:
SAD 1x1, CENSUS 7x7
\alpha = 0.43
\sigma = 4
Latex Bibtex:
@inproceedings{kuzmin2017end,
title={End-to-end Learning of Cost-Volume Aggregation
for
Real-time Dense Stereo},
author={Kuzmin, Andrey and Mikushin, Dmitry and
Lempitsky,
Victor},
booktitle = {2017 IEEE 27th International Workshop on
Machine Learning for Signal Processing (MLSP)},
year={2017}
}

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 7.43 % 8.97 % 1.1 px 1.4 px
3 pixels 5.08 % 6.37 % 1.1 px 1.4 px
4 pixels 3.98 % 5.05 % 1.1 px 1.4 px
5 pixels 3.29 % 4.18 % 1.1 px 1.4 px
This table as LaTeX

Reflective Regions

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 29.59 % 32.87 % 4.2 px 5.2 px
3 pixels 22.19 % 25.58 % 4.2 px 5.2 px
4 pixels 18.14 % 21.47 % 4.2 px 5.2 px
5 pixels 15.47 % 18.70 % 4.2 px 5.2 px
This table as LaTeX

Test Image 0

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 4.59 % 6.79 % 0.8 px 0.9 px
3 pixels 3.42 % 5.34 % 0.8 px 0.9 px
4 pixels 2.98 % 4.35 % 0.8 px 0.9 px
5 pixels 2.36 % 3.18 % 0.8 px 0.9 px
This table as LaTeX





Test Image 1

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 5.20 % 8.21 % 0.9 px 1.0 px
3 pixels 3.08 % 5.22 % 0.9 px 1.0 px
4 pixels 2.38 % 3.81 % 0.9 px 1.0 px
5 pixels 2.06 % 2.87 % 0.9 px 1.0 px
This table as LaTeX





Test Image 2

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 13.15 % 14.31 % 1.8 px 1.9 px
3 pixels 11.37 % 12.19 % 1.8 px 1.9 px
4 pixels 10.00 % 10.54 % 1.8 px 1.9 px
5 pixels 9.05 % 9.52 % 1.8 px 1.9 px
This table as LaTeX





Test Image 3

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 6.63 % 9.16 % 0.8 px 1.0 px
3 pixels 3.99 % 6.35 % 0.8 px 1.0 px
4 pixels 2.81 % 4.76 % 0.8 px 1.0 px
5 pixels 1.72 % 3.28 % 0.8 px 1.0 px
This table as LaTeX





Test Image 4

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 2.26 % 3.59 % 0.6 px 0.7 px
3 pixels 1.45 % 2.26 % 0.6 px 0.7 px
4 pixels 1.24 % 1.43 % 0.6 px 0.7 px
5 pixels 1.01 % 1.20 % 0.6 px 0.7 px
This table as LaTeX





Test Image 5

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 2.63 % 3.22 % 0.7 px 0.7 px
3 pixels 1.41 % 1.81 % 0.7 px 0.7 px
4 pixels 1.17 % 1.34 % 0.7 px 0.7 px
5 pixels 1.00 % 1.16 % 0.7 px 0.7 px
This table as LaTeX





Test Image 6

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 13.04 % 14.41 % 1.9 px 2.1 px
3 pixels 10.05 % 11.13 % 1.9 px 2.1 px
4 pixels 8.54 % 9.43 % 1.9 px 2.1 px
5 pixels 7.52 % 7.98 % 1.9 px 2.1 px
This table as LaTeX





Test Image 7

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 13.23 % 15.39 % 1.7 px 2.3 px
3 pixels 9.38 % 11.58 % 1.7 px 2.3 px
4 pixels 7.48 % 9.54 % 1.7 px 2.3 px
5 pixels 6.14 % 8.00 % 1.7 px 2.3 px
This table as LaTeX





Test Image 8

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 5.59 % 6.45 % 1.0 px 1.0 px
3 pixels 4.23 % 4.64 % 1.0 px 1.0 px
4 pixels 3.38 % 3.77 % 1.0 px 1.0 px
5 pixels 2.95 % 3.33 % 1.0 px 1.0 px
This table as LaTeX





Test Image 9

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 11.01 % 13.21 % 1.2 px 1.7 px
3 pixels 6.82 % 9.12 % 1.2 px 1.7 px
4 pixels 4.23 % 6.47 % 1.2 px 1.7 px
5 pixels 3.19 % 5.09 % 1.2 px 1.7 px
This table as LaTeX





Test Image 10

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 3.61 % 4.99 % 0.7 px 0.8 px
3 pixels 2.21 % 3.28 % 0.7 px 0.8 px
4 pixels 1.62 % 2.46 % 0.7 px 0.8 px
5 pixels 1.33 % 1.93 % 0.7 px 0.8 px
This table as LaTeX





Test Image 11

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 13.04 % 16.14 % 1.6 px 1.8 px
3 pixels 8.08 % 11.22 % 1.6 px 1.8 px
4 pixels 5.98 % 8.87 % 1.6 px 1.8 px
5 pixels 4.92 % 7.47 % 1.6 px 1.8 px
This table as LaTeX





Test Image 12

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 7.36 % 9.55 % 1.5 px 1.7 px
3 pixels 5.91 % 7.81 % 1.5 px 1.7 px
4 pixels 5.13 % 6.68 % 1.5 px 1.7 px
5 pixels 4.67 % 5.99 % 1.5 px 1.7 px
This table as LaTeX





Test Image 13

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 7.18 % 8.49 % 1.1 px 1.4 px
3 pixels 4.71 % 5.97 % 1.1 px 1.4 px
4 pixels 3.75 % 4.98 % 1.1 px 1.4 px
5 pixels 3.19 % 4.33 % 1.1 px 1.4 px
This table as LaTeX





Test Image 14

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 7.59 % 8.33 % 0.9 px 0.9 px
3 pixels 4.74 % 5.31 % 0.9 px 0.9 px
4 pixels 3.05 % 3.39 % 0.9 px 0.9 px
5 pixels 2.06 % 2.34 % 0.9 px 0.9 px
This table as LaTeX





Test Image 15

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 8.78 % 9.61 % 1.0 px 1.1 px
3 pixels 5.73 % 5.99 % 1.0 px 1.1 px
4 pixels 3.94 % 3.90 % 1.0 px 1.1 px
5 pixels 3.02 % 2.94 % 1.0 px 1.1 px
This table as LaTeX





Test Image 16

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 5.50 % 5.72 % 0.8 px 0.8 px
3 pixels 2.73 % 2.98 % 0.8 px 0.8 px
4 pixels 1.77 % 2.03 % 0.8 px 0.8 px
5 pixels 1.25 % 1.52 % 0.8 px 0.8 px
This table as LaTeX





Test Image 17

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 7.54 % 8.45 % 1.1 px 1.2 px
3 pixels 6.29 % 6.74 % 1.1 px 1.2 px
4 pixels 5.40 % 5.71 % 1.1 px 1.2 px
5 pixels 4.24 % 4.50 % 1.1 px 1.2 px
This table as LaTeX





Test Image 18

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 8.80 % 9.67 % 1.1 px 1.1 px
3 pixels 6.70 % 7.43 % 1.1 px 1.1 px
4 pixels 5.14 % 5.76 % 1.1 px 1.1 px
5 pixels 4.27 % 4.78 % 1.1 px 1.1 px
This table as LaTeX





Test Image 19

Error Out-Noc Out-All Avg-Noc Avg-All
2 pixels 4.08 % 5.76 % 0.7 px 0.7 px
3 pixels 2.17 % 2.91 % 0.7 px 0.7 px
4 pixels 1.25 % 1.38 % 0.7 px 0.7 px
5 pixels 0.78 % 0.77 % 0.7 px 0.7 px
This table as LaTeX







eXTReMe Tracker