This commit is contained in:
Simon Meister 2017-11-18 17:54:44 +01:00
parent 73470ee4a8
commit a9f4ac0d56
6 changed files with 116 additions and 1 deletions

1
.gitignore vendored
View File

@ -15,5 +15,6 @@ thesis.out
thesis.run.xml
thesis.toc
thesis.bcf
pdfa.xmpi
downloads

Binary file not shown.

Binary file not shown.

Binary file not shown.

View File

@ -13,6 +13,8 @@
type=bsc, % für Bachelorarbeit
]{tudthesis}
\usepackage[a-1b]{pdfx}
\usepackage[T1]{fontenc}
\usepackage[utf8]{inputenc} % korrekte Darstellung von Umlauten u. Sonderzeichen
\usepackage[stable]{footmisc} % mehr Optionen für Fußnoten
@ -88,7 +90,7 @@
\author{\myname}
\thesistitle{\mytitleen}{\mytitlede}
\birthplace{Erbach}
\date{23.11.2017}
\date{21.11.2017}
\referee{\myprof}{M.Sc. Junhwa Hur}
\department{\myinstitute}
\group{\myfaculty}

112
thesis.xmpdata Normal file
View File

@ -0,0 +1,112 @@
% Replace the following information with your document's actual
% metadata. If you do not want to set a value for a certain parameter,
% just omit it.
%
% Symbols permitted in metadata
% =============================
%
% Within the metadata, all printable ASCII characters except
% '\', '{', '}', and '%' represent themselves. Also, all printable
% Unicode characters from the basic multilingual plane (i.e., up to
% code point U+FFFF) can be used directly with the UTF-8 encoding.
% Consecutive whitespace characters are combined into a single
% space. Whitespace after a macro such as \copyright, \backslash, or
% \sep is ignored. Blank lines are not permitted. Moreover, the
% following markup can be used:
%
% '\ ' - a literal space (for example after a macro)
% \% - a literal '%'
% \{ - a literal '{'
% \} - a literal '}'
% \backslash - a literal '\'
% \copyright - the (c) copyright symbol
%
% The macro \sep is only permitted within \Author, \Keywords, and
% \Org. It is used to separate multiple authors, keywords, etc.
%
% List of supported metadata fields
% =================================
%
% Here is a complete list of user-definable metadata fields currently
% supported, and their meanings. More may be added in the future.
%
% General information:
%
% \Author - the document's human author. Separate multiple
% authors with \sep.
% \Title - the document's title.
% \Keywords - list of keywords, separated with \sep.
% \Subject - the abstract.
% \Org - publishers.
%
% Copyright information:
%
% \Copyright - a copyright statement.
% \CopyrightURL - location of a web page describing the owner
% and/or rights statement for this document.
% \Copyrighted - 'True' if the document is copyrighted, and
% 'False' if it isn't. This is automatically set
% to 'True' if either \Copyright or \CopyrightURL
% is specified, but can be overridden. For
% example, if the copyright statement is "Public
% Domain", this should be set to 'False'.
%
% Publication information:
%
% \PublicationType - The type of publication. If defined, must be
% one of book, catalog, feed, journal, magazine,
% manual, newsletter, pamphlet. This is
% automatically set to "journal" if \Journaltitle
% is specified, but can be overridden.
% \Journaltitle - The title of the journal in which the document
% was published.
% \Journalnumber - The ISSN for the publication in which the
% document was published.
% \Volume - Journal volume.
% \Issue - Journal issue/number.
% \Firstpage - First page number of the published version of
% the document.
% \Lastpage - Last page number of the published version of
% the document.
% \Doi - Digital Object Identifier (DOI) for the
% document, without the leading "doi:".
% \CoverDisplayDate - Date on the cover of the journal issue, as a
% human-readable text string.
% \CoverDate - Date on the cover of the journal issue, in a
% format suitable for storing in a database field
% with a 'date' data type.
\Title{Motion R-CNN: Instance-level 3D Motion Estimation with Region-based CNNs}
\Author{Simon Meister}
\Copyright{Copyright \copyright\ 2017 "Simon Meister"}
\Keywords{optical flow\sep
instance segmentation\sep
deep learning}
\Subject{With the advent of deep learning, it has become popular to re-purpose generic deep networks for classical
computer vision problems involving pixel-wise estimation.
Following this trend, many recent end-to-end deep learning approaches to optical flow and scene flow
predict complete, high resolution flow fields with a generic network for dense, pixel-wise prediction,
thereby ignoring the inherent structure of the underlying motion estimation problem and any physical
constraints within the scene.
We introduce a scalable end-to-end deep learning approach for dense motion estimation that respects the
structure of the scene as being composed of distinct objects, thus combining the representation learning
benefits and speed of end-to-end deep networks with a physically plausible scene model inspired by
slanted plane energy-minimization approaches to scene flow.
Building on recent advances in region-based convolutional networks (R-CNNs), we integrate motion
estimation with instance segmentation. Given two consecutive frames from a monocular RGB-D camera,
our resulting end-to-end deep network detects objects with precise per-pixel object masks and estimates
the 3D motion of each detected object between the frames. By additionally estimating a global camera
motion in the same network, we compose a dense optical flow field based on instance-level and global
motion predictions. We train our network on the synthetic Virtual KITTI dataset, which provides ground
truth for all components of our system.}
\setRGBcolorprofile{sRGB_IEC61966-2-1_black_scaled.icc}
{sRGB_IEC61966-2-1_black_scaled}
{sRGB IEC61966 v2.1 with black scaling}
{http://www.color.org}