Method

Geometry and Context Network [GC-NET]


Submitted on 27 Jan. 2017 17:02 by
Alex Kendall (University of Cambridge)

Running time:0.9 s
Environment:Nvidia GTX Titan X

Method Description:
We propose a novel deep learning architecture for
regressing disparity from a rectified pair of
stereo images. We leverage knowledge of the
problem's geometry to form a cost volume using
deep feature representations. We learn to
incorporate contextual information using 3-D
convolutions over this volume. Disparity values
are regressed from the cost volume using a
proposed differentiable soft argmin operation,
which allows us to train our method end-to-end to
sub-pixel accuracy without any additional post-
processing or regularization. We evaluate our
method on the Scene Flow and KITTI datasets and
on KITTI we set a new state-of-the-art benchmark,
while being significantly faster than competing
approaches.
Parameters:
See paper.
Latex Bibtex:
@inproceedings{kendall2017end,
title={End-to-End Learning of Geometry and Context for
Deep Stereo Regression},
author={Kendall, Alex and Martirosyan, Hayk and
Dasgupta, Saumitro and Henry, Peter and Kennedy, Ryan
and Bachrach, Abraham and Bry, Adam},
booktitle = {Proceedings of the International Conference on
Computer Vision ({ICCV})},
year={2017}
}

Detailed Results

This page provides detailed results for the method(s) selected. For the first 20 test images, the percentage of erroneous pixels is depicted in the table. We use the error metric described in Object Scene Flow for Autonomous Vehicles (CVPR 2015), which considers a pixel to be correctly estimated if the disparity or flow end-point error is <3px or <5% (for scene flow this criterion needs to be fulfilled for both disparity maps and the flow map). Underneath, the left input image, the estimated results and the error maps are shown (for disp_0/disp_1/flow/scene_flow, respectively). The error map uses the log-color scale described in Object Scene Flow for Autonomous Vehicles (CVPR 2015), depicting correct estimates (<3px or <5% error) in blue and wrong estimates in red color tones. Dark regions in the error images denote the occluded pixels which fall outside the image boundaries. The false color maps of the results are scaled to the largest ground truth disparity values / flow magnitudes.

Test Set Average

Error D1-bg D1-fg D1-all
All / All 2.21 6.16 2.87
All / Est 2.21 6.16 2.87
Noc / All 2.02 5.58 2.61
Noc / Est 2.02 5.58 2.61
This table as LaTeX

Test Image 0

Error D1-bg D1-fg D1-all
All / All 1.40 1.24 1.38
All / Est 1.40 1.24 1.38
Noc / All 1.33 1.24 1.31
Noc / Est 1.33 1.24 1.31
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 1

Error D1-bg D1-fg D1-all
All / All 2.20 3.80 2.38
All / Est 2.20 3.80 2.38
Noc / All 2.10 3.80 2.29
Noc / Est 2.10 3.80 2.29
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 2

Error D1-bg D1-fg D1-all
All / All 2.68 3.28 2.71
All / Est 2.68 3.28 2.71
Noc / All 2.45 3.28 2.49
Noc / Est 2.45 3.28 2.49
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 3

Error D1-bg D1-fg D1-all
All / All 2.42 0.61 2.25
All / Est 2.42 0.61 2.25
Noc / All 2.38 0.61 2.21
Noc / Est 2.38 0.61 2.21
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 4

Error D1-bg D1-fg D1-all
All / All 1.96 1.73 1.92
All / Est 1.96 1.73 1.92
Noc / All 1.94 1.73 1.91
Noc / Est 1.94 1.73 1.91
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 5

Error D1-bg D1-fg D1-all
All / All 5.29 3.41 5.12
All / Est 5.29 3.41 5.12
Noc / All 4.38 3.41 4.29
Noc / Est 4.38 3.41 4.29
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 6

Error D1-bg D1-fg D1-all
All / All 5.24 3.93 5.10
All / Est 5.24 3.93 5.10
Noc / All 5.35 3.93 5.20
Noc / Est 5.35 3.93 5.20
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 7

Error D1-bg D1-fg D1-all
All / All 0.83 3.01 1.25
All / Est 0.83 3.01 1.25
Noc / All 0.84 3.01 1.27
Noc / Est 0.84 3.01 1.27
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 8

Error D1-bg D1-fg D1-all
All / All 0.95 1.97 1.14
All / Est 0.95 1.97 1.14
Noc / All 0.95 1.97 1.14
Noc / Est 0.95 1.97 1.14
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 9

Error D1-bg D1-fg D1-all
All / All 1.01 1.49 1.14
All / Est 1.01 1.49 1.14
Noc / All 1.03 1.57 1.16
Noc / Est 1.03 1.57 1.16
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 10

Error D1-bg D1-fg D1-all
All / All 0.88 5.57 1.95
All / Est 0.88 5.57 1.95
Noc / All 0.89 5.57 1.97
Noc / Est 0.89 5.57 1.97
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 11

Error D1-bg D1-fg D1-all
All / All 0.79 1.60 0.94
All / Est 0.79 1.60 0.94
Noc / All 0.79 1.60 0.94
Noc / Est 0.79 1.60 0.94
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 12

Error D1-bg D1-fg D1-all
All / All 0.77 1.59 0.82
All / Est 0.77 1.59 0.82
Noc / All 0.65 1.59 0.71
Noc / Est 0.65 1.59 0.71
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 13

Error D1-bg D1-fg D1-all
All / All 0.79 0.05 0.70
All / Est 0.79 0.05 0.70
Noc / All 0.68 0.05 0.60
Noc / Est 0.68 0.05 0.60
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 14

Error D1-bg D1-fg D1-all
All / All 1.67 1.29 1.66
All / Est 1.67 1.29 1.66
Noc / All 1.52 1.29 1.52
Noc / Est 1.52 1.29 1.52
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 15

Error D1-bg D1-fg D1-all
All / All 3.09 0.91 2.89
All / Est 3.09 0.91 2.89
Noc / All 3.14 0.91 2.94
Noc / Est 3.14 0.91 2.94
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 16

Error D1-bg D1-fg D1-all
All / All 5.05 0.18 4.34
All / Est 5.05 0.18 4.34
Noc / All 4.79 0.18 4.11
Noc / Est 4.79 0.18 4.11
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 17

Error D1-bg D1-fg D1-all
All / All 1.36 0.43 1.26
All / Est 1.36 0.43 1.26
Noc / All 1.36 0.43 1.26
Noc / Est 1.36 0.43 1.26
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 18

Error D1-bg D1-fg D1-all
All / All 5.10 2.57 3.90
All / Est 5.10 2.57 3.90
Noc / All 4.92 2.57 3.80
Noc / Est 4.92 2.57 3.80
This table as LaTeX

Input Image

D1 Result

D1 Error


Test Image 19

Error D1-bg D1-fg D1-all
All / All 0.89 1.59 0.97
All / Est 0.89 1.59 0.97
Noc / All 0.86 1.59 0.95
Noc / Est 0.86 1.59 0.95
This table as LaTeX

Input Image

D1 Result

D1 Error




eXTReMe Tracker