Initialize a repository for writeup

templates/baposter-template/examples/code/Makefile

@@ -0,0 +1,7 @@
+ALL: poster_landscape.pdf 
+
+%.pdf: %.tex Makefile
+	pdflatex --enable-write18 $< &&	pdflatex $< && pdflatex $<
+
+clean:
+	  rm -f *.aux	*.bbl	*.blg	*.log	poster_*.pdf 

templates/baposter-template/examples/code/papers.bib

@@ -0,0 +1,305 @@
+@article{matthews:aamr,
+  address = {Hingham, MA, USA},
+  author = {Iain Matthews and Simon Baker},
+  citeulike-article-id = {1848202},
+  doi = {10.1023/B:VISI.0000029666.37597.d3},
+  issn = {0920-5691},
+  journal = {IJCV},
+  keywords = {2d, aam},
+  month = {November},
+  number = {2},
+  pages = {135--164},
+  priority = {2},
+  publisher = {Kluwer Academic Publishers},
+  title = {{A}ctive {A}ppearance {M}odels {R}evisited},
+  url = {http://dx.doi.org/10.1023/B:VISI.0000029666.37597.d3},
+  volume = {60},
+  year = {2004}
+}
+
+
+@article{hager98:project_out,
+  author = {Hager, G. D.  and Belhumeur, P. N. },
+  booktitle = {Pattern Analysis and Machine Intelligence, IEEE Transactions on},
+  citeulike-article-id = {1865605},
+  journal = {Pattern Analysis and Machine Intelligence, IEEE Transactions on},
+  keywords = {aam, project_out_trick, tracking},
+  number = {10},
+  pages = {1025--1039},
+  priority = {5},
+  title = {Efficient region tracking with parametric models of geometry and illumination},
+  url = {http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=722606},
+  volume = {20},
+  year = {1998}
+}
+
+
+
+@inproceedings{sami:icia,
+  address = {Washington, DC, USA},
+  author = {Romdhani, Sami   and Vetter, Thomas  },
+  booktitle = {ICCV '03},
+  citeulike-article-id = {1869162},
+  isbn = {0769519504},
+  keywords = {aam, inverse_compositional, morphable_model},
+  priority = {5},
+  publisher = {IEEE Computer Society},
+  title = {Efficient, {R}obust and {A}ccurate {F}itting of a {3D} {M}orphable {M}odel},
+  url = {http://portal.acm.org/citation.cfm?id=946247.946642},
+  year = {2003}
+}
+
+@techreport{sami:selective_recovery,
+  author = {Romdhani, Sami   and Canterakis, Nikolaos   and Vetter, Thomas  },
+  booktitle = {Technical Report Nr 3},
+  citeulike-article-id = {1896147},
+  institution = {University of Basel},
+  keywords = {morphable_model, weak_perspective},
+  month = {July},
+  organization = {Computer Science Dept},
+  priority = {5},
+  title = {Selective vs. Global Recovery of Rigid and Non-Rigid Motion},
+  year = {2003}
+}
+
+
+@inproceedings{baker01:equivalence,
+  abstract = {There are two major formulations of image alignment using gradient descent. The first estimates an additive increment to the parameters (the additive approach), the second an incremental warp (the compositional approach). We first prove that these two formulations are equivalent. A very efficient algorithm was proposed by Hager and Belhumeur (1998) using the additive approach that unfortunately can only be applied to a very restricted class of warps. We show that using the compositional approach an equally efficient algorithm (the inverse compositional algorithm) can be derived that can be applied to any set of warps which form a group. While most warps used in computer vision form groups, there are a certain warps that do not. Perhaps most notable is the set of piecewise affine warps used in flexible appearance models (FAMs). We end this paper by extending the inverse compositional algorithm to apply to FAMs.},
+  author = {Baker, S.  and Matthews, I. },
+  booktitle = {CVPR '01},
+  citeulike-article-id = {1967072},
+  journal = {CVPR '01},
+  keywords = {aam, icia},
+  pages = {I-1090--I-1097 vol.1},
+  priority = {5},
+  title = {Equivalence and {E}fficiency of {I}mage {A}lignment {A}lgorithms},
+  url = {http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=990652},
+  volume = {1},
+  year = {2001}
+}
+
+@article{cootes01:aam, 
+  author = {Cootes, T. F.  and Edwards, G. J.  and Taylor, C. J. }, 
+  journal = {PAMI},
+  keywords = {aam, model},
+  number = {6},
+  pages = {681--685},
+  title = {{A}ctive {A}ppearance {M}odels}, 
+  volume = {23}, 
+  year = {2001} 
+}
+
+  
+
+@article{cootes:tps,
+  abstract = {Face images are difficult to interpret because they are highly variable. Sources of variability include individual appearance, 3D pose, facial expression, and lighting. We describe a compact parametrized model of facial appearance which takes into account all these sources of variability. The model represents both shape and gray-level appearance, and is created by performing a statistical analysis over a training set of face images. A robust multiresolution search algorithm is used to fit the model to faces in new images. This allows the main facial features to be located, and a set of shape, and gray-level appearance parameters to be recovered. A good approximation to a given face can be reconstructed using less than 100 of these parameters. This representation can be used for tasks such as image coding, person identification, 3D pose recovery, gender recognition, and expression recognition. Experimental results are presented for a database of 690 face images obtained under widely varying conditions of 3D pose, lighting, and facial expression. The system performs well on all the tasks listed above},
+  author = {Lanitis, A.  and Taylor, C. J.  and Cootes, T. F. },
+  citeulike-article-id = {827645},
+  journal = {Pattern Analysis and Machine Intelligence, IEEE Transactions on},
+  keywords = {3d, face},
+  number = {7},
+  pages = {743--756},
+  priority = {5},
+  title = {Automatic interpretation and coding of face images using flexible models},
+  url = {http://ieeexplore.ieee.org/xpls/abs\_all.jsp?arnumber=598231},
+  volume = {19},
+  year = {1997}
+}
+
+
+@techreport{baker01:icia_tr,
+  address = {Pittsburgh, PA},
+  author = {Simon},
+  citeulike-article-id = {2390678},
+  institution = {Robotics Institute, Carnegie Mellon University},
+  keywords = {aam, icia},
+  month = {February},
+  number = {CMU-RI-TR-01-03},
+  priority = {0},
+  title = {Aligning Images Incrementally Backwards},
+  url = {http://www.ri.cmu.edu/pubs/pub\_3491.html},
+  year = {2001}
+}
+
+
+@misc{minka:matrix,
+  abstract = {This paper contains a large number of matrix identities which cannot be absorbed by mere reading. The reader is encouraged to take time and check each equation by hand and work out the examples. This is advanced material; see Searle (1982) for basic results. 1 Derivatives},
+  author = {Minka, T. },
+  citeulike-article-id = {899368},
+  keywords = {linear\_algebra},
+  priority = {0},
+  title = {Old and new matrix algebra useful for statistics},
+  url = {http://citeseer.ist.psu.edu/minka97old.html},
+  year = {1997}
+}
+
+
+@article{matthews:kanade20,
+  author = {Simon Baker  and Iain Matthews},
+  citeulike-article-id = {1238610},
+  doi = {10.1023/B:VISI.0000011205.11775.fd},
+  journal = {IJCV},
+  longjournal = {International Journal of Computer Vision},
+  month = {February},
+  number = {3},
+  pages = {221--255},
+  priority = {5},
+  title = {{L}ucas-{K}anade 20 {Y}ears {O}n: {A} {U}nifying {F}ramework},
+  theurl = {http://portal.acm.org/citation.cfm?id=964568.964604},
+  volume = {56},
+  year = {2004}
+}
+
+@inproceedings{wu:boosted_ranking,
+  abstract = {Face alignment seeks to deform a face model to match it with the features of the image of a face by optimizing an appropriate cost function. We propose a new face model that is aligned by maximizing a score function, which we learn from training data, and that we impose to be concave. We show that this problem can be reduced to learning a classifier that is able to say whether or not by switching from one alignment to a new one, the model is approaching the correct fitting. This relates to the ranking problem where a number of instances need to be ordered. For training the model, we propose to extend GentleBoost [23] to rank-learning. Extensive experimentation shows the superiority of this approach to other learning paradigms, and demonstrates that this model exceeds the alignment performance of the state-of-the-art.},
+  author = {Wu, Hao   and Liu, Xiaoming   and Doretto, Gianfranco  },
+  booktitle = {CVPR '08},
+  citeulike-article-id = {3132874},
+  doi = {10.1109/CVPR.2008.4587753},
+  journal = {CVPR '08},
+  keywords = {aam, descriptors},
+  pages = {1--8},
+  posted-at = {2008-08-18 15:05:08},
+  priority = {0},
+  title = {Face {A}lignment via {B}oosted {R}anking {M}odel},
+  url = {http://dx.doi.org/10.1109/CVPR.2008.4587753},
+  year = {2008}
+}
+
+
+@inproceedings{Saragih07:aam,
+  abstract = {The Active Appearance Model (AAM) is a powerful generative method for modeling and registering deformable visual objects. Most methods for AAM fitting utilize a linear parameter update model in an iterative framework. Despite its popularity, the scope of this approach is severely restricted, both in fitting accuracy and capture range, due to the simplicity of the linear update models used. In this paper, we present an new AAM fitting formulation, which utilizes a nonlinear update model. To motivate our approach, we compare its performance against two popular fitting methods on two publicly available face databases, in which this formulation boasts significant performance improvements.},
+  author = {Saragih, J.  and Goecke, R. },
+  booktitle = {ICCV '07},
+  citeulike-article-id = {3132996},
+  doi = {10.1109/ICCV.2007.4409106},
+  journal = {ICCV 2007},
+  keywords = {aam},
+  pages = {1--8},
+  posted-at = {2008-08-18 16:15:13},
+  priority = {0},
+  title = {A {N}onlinear {D}iscriminative {A}pproach to {AAM} {F}itting},
+  url = {http://dx.doi.org/10.1109/ICCV.2007.4409106},
+  year = {2007}
+}
+
+@inproceedings{Liu07:aam,
+  author = {Liu, Xiaoming},
+  booktitle = {CVPR '07},
+  journal = {CVPR '07},
+  pages = {1--8},
+  title = {{G}eneric {F}ace {A}lignment using {B}oosted {A}ppearance {M}odel},
+  url = {http://dx.doi.org/10.1109/CVPR.2007.383265},
+  year = {2007}
+}
+
+
+%  address = {New York, NY, USA},
+  @inproceedings{blanz:model,
+    author = {Blanz, Volker   and Vetter, Thomas  },
+    booktitle = {SIGGRAPH '99},
+    citeulike-article-id = {423122},
+    doi = {10.1145/311535.311556},
+    isbn = {0201485605},
+    keywords = {morphable_model},
+    pages = {187--194},
+    priority = {0},
+    publisher = {ACM Press},
+    title = {A {M}orphable {M}odel for the {S}ynthesis of {3D} {F}aces},
+    url = {http://portal.acm.org/citation.cfm?id=311556},
+    year = {1999}
+  }
+
+@article{vemuri98:motion,
+  author = {Vemuri, B. C.  and Huang, S.  and Sahni, S.  and Leonard, C. M.  and Mohr, C.  and Gilmore, R.  and Fitzsimmons, J. },
+  citeulike-article-id = {3484827},
+  doi = {http://dx.doi.org/10.1016/S1361-8415(98)80004-2},
+  issn = {1361-8415},
+  journal = {Medical Image Analysis},
+  keywords = {icia},
+  month = {March},
+  pages = {79--98},
+  posted-at = {2008-11-06 15:07:17},
+  priority = {0},
+  publisher = {Elsevier},
+  title = {An efficient motion estimator with application to medical image registration},
+  url = {http://dx.doi.org/10.1016/S1361-8415(98)80004-2},
+  year = {1998}
+}
+
+@inproceedings{burkhardt86:motion,
+  author = {H. Burkhardt and N. Diehl},
+  title = {Simultaneous {E}stimation of {R}otation and {T}ranslation in {I}mage {S}equences},
+  booktitle = {Proc. of the European Signal Processing Conference, EUSIPCO-86},
+  address = {Den Haag},
+  year = {1986}
+}
+
+@phdthesis{diehl:diss,
+  author = {N. Diehl},
+  title = {Methoden zur allgemeinen {B}ewegungssch{\"a}tzung in {B}ildfolgen},
+  school = {TU Hamburg-Harburg},
+  year = {1988},
+  note = {Published as Fortschrittsbericht (Reihe 10, Nr. 92) VDI-Zeitschriften, VDI-Verlag}
+}
+
+@inproceedings{Nguyen_2008_6186,
+  author = "Minh Hoai Nguyen and Fernando de la Torre Frade",
+  title = "Learning Image Alignment without Local Minima for Face Detection and Tracking",
+  booktitle = "8th IEEE International Conference on Automatic Face and Gesture Recognition",
+  month = "September",
+  year = "2008"
+}
+
+@inproceedings{xm2vts,
+  MONTH = {March},
+  YEAR = {1999},
+  AUTHOR = {K. Messer and J. Matas and J. Kittler and J. Luettin and G. Maitre},
+  TITLE = {{XM2VTSDB}: {T}he {E}xtended {M2VTS} {D}atabase},
+  Booktitle  = {2nd Int. Conf. on Audio and Video-based Biometric Person Authentication },
+  PAGES = {}
+}
+
+@TECHREPORT{imm,
+  author       = "M. M. Nordstr{\o}m and M. Larsen and J. Sierakowski and M. B. Stegmann",
+  title        = "The {IMM} Face Database - An Annotated Dataset of 240 Face Images",
+  year         = "2004",
+  month        = "may",
+  keywords     = "annotated image dataset, face images, statistical models of shape",
+  number       = "",
+  series       = "",
+  institution  = "{IMM}, {TU} {D}enmark {DTU}",
+  address      = "",
+  type         = "",
+  url          = "http://www2.imm.dtu.dk/pubdb/p.php?3160",
+  abstract     = "This note describes a dataset consisting of 240 annotated monocular images of 40 different human faces. Points of correspondence are placed on each image so the dataset can be readily used for building statistical models of shape. Format specifications and terms of use are also given in this note."
+}
+
+
+@article{wimmer07:learning,
+  author =       {Matthias Wimmer and Freek Stulp and Sylvia Pietzsch and Bernd Radig},
+  title =        {Learning Local Objective Functions for Robust Face Model Fitting},
+  journal =      {PAMI},
+  year =         {2008},
+  volume =       {30},
+  number =       {8},
+  issn =   {0162-8828},
+  pages =    {1357-1370},
+  bib2html_pubtype  =   {Journal},
+  bib2html_rescat   =   {Image Understanding},
+  bib2html_groups   =   {IU},
+  doi =     {http://doi.ieeecomputersociety.org/10.1109/TPAMI.2007.70793},
+  publisher =     {IEEE Computer Society},
+  address =     {Los Alamitos, CA, USA},
+}
+
+@misc{cootes:talkingface,
+  AUTHOR = "T. F. Cootes",
+  TITLE = "Talking Face Video",
+  MONTH = "October",
+  YEAR = {2008},
+  NOTE = {{\smaller[1]{\url{www-prima.inrialpes.fr/FGnet/data/01-TalkingFace/talking\_face.html}}}}
+
+}
+
+

templates/baposter-template/examples/code/poster_landscape.aux

@@ -0,0 +1,5 @@
+\relax 
+\bibstyle{ieee}
+\bibcite{amberg07:nicp}{1}
+\newlabel{eqn:f}{{1}{2}}
+\newlabel{eqn:comp}{{2}{2}}

templates/baposter-template/examples/code/poster_landscape.tex

@@ -0,0 +1,634 @@
+\documentclass[landscape,a0paper,fontscale=0.292]{baposter}
+
+\usepackage[vlined]{algorithm2e}
+\usepackage{times}
+\usepackage{calc}
+\usepackage{url}
+\usepackage{graphicx}
+\usepackage{amsmath}
+\usepackage{amssymb}
+\usepackage{relsize}
+\usepackage{multirow}
+\usepackage{booktabs}
+
+\usepackage{graphicx}
+\usepackage{multicol}
+\usepackage[T1]{fontenc}
+\usepackage{ae}
+
+\graphicspath{{images/}}
+
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+ %%%% Some math symbols used in the text
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+ % Format 
+ \newcommand{\RotUP}[1]{\begin{sideways}#1\end{sideways}}
+
+
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+ % Multicol Settings
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+ \setlength{\columnsep}{0.7em}
+ \setlength{\columnseprule}{0mm}
+
+
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+ % Save space in lists. Use this after the opening of the list
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+ \newcommand{\compresslist}{%
+ \setlength{\itemsep}{1pt}%
+ \setlength{\parskip}{0pt}%
+ \setlength{\parsep}{0pt}%
+ }
+
+
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+ % Formating
+ \newcommand{\Matrix}[1]{\begin{bmatrix} #1 \end{bmatrix}}
+ \newcommand{\Vector}[1]{\begin{pmatrix} #1 \end{pmatrix}}
+
+ \newcommand*{\norm}[1]{\mathopen\| #1 \mathclose\|}% use instead of $\|x\|$
+ \newcommand*{\abs}[1]{\mathopen| #1 \mathclose|}% use instead of $\|x\|$
+ \newcommand*{\normLR}[1]{\left\| #1 \right\|}% use instead of $\|x\|$
+
+ \newcommand*{\SET}[1]  {\ensuremath{\mathcal{#1}}}
+ \newcommand*{\FUN}[1]  {\ensuremath{\mathcal{#1}}}
+ \newcommand*{\MAT}[1]  {\ensuremath{\boldsymbol{#1}}}
+ \newcommand*{\VEC}[1]  {\ensuremath{\boldsymbol{#1}}}
+ \newcommand*{\CONST}[1]{\ensuremath{\mathit{#1}}}
+
+ \DeclareMathOperator*{\argmax}{arg\,max}
+ \DeclareMathOperator*{\diag}{diag}
+ \DeclareMathOperator*{\argmin}{arg\,min}
+ \DeclareMathOperator*{\vectorize}{vec}
+ \DeclareMathOperator*{\reshape}{reshape}
+
+ %-----------------------------------------------------------------------------
+ % Differentiation
+ \newcommand*{\Nabla}[1]{\nabla_{\!#1}}
+
+ \renewcommand*{\d}{\mathrm{d}}
+ \newcommand*{\dd}{\partial}
+
+ \newcommand*{\At}[2]{\ensuremath{\left.#1\right|_{#2}}}
+ \newcommand*{\AtZero}[1]{\At{#1}{\pp=\VEC 0}}
+
+ \newcommand*{\diffp}[2]{\ensuremath{\frac{\dd #1}{\dd #2}}}
+ \newcommand*{\diffpp}[3]{\ensuremath{\frac{\dd^2 #1}{\dd #2 \dd #3}}}
+ \newcommand*{\diffppp}[4]{\ensuremath{\frac{\dd^3 #1}{\dd #2 \dd #3 \dd #4}}}
+ \newcommand*{\difff}[2]{\ensuremath{\frac{\d #1}{\d #2}}}
+ \newcommand*{\diffff}[3]{\ensuremath{\frac{\d^2 #1}{\d #2 \d #3}}}
+ \newcommand*{\difffp}[3]{\ensuremath{\frac{\dd\d #1}{\d #2 \dd #3}}}
+ \newcommand*{\difffpp}[4]{\ensuremath{\frac{\dd^2\d #1}{\d #2 \dd #3 \dd #4}}}
+
+ \newcommand*{\diffpAtZero}[2]{\ensuremath{\AtZero{\diffp{#1}{#2}}}}
+ \newcommand*{\diffppAtZero}[3]{\ensuremath{\AtZero{\diffpp{#1}{#2}{#3}}}}
+ \newcommand*{\difffAt}[3]{\ensuremath{\At{\difff{#1}{#2}}{#3}}}
+ \newcommand*{\difffAtZero}[2]{\ensuremath{\AtZero{\difff{#1}{#2}}}}
+ \newcommand*{\difffpAtZero}[3]{\ensuremath{\AtZero{\difffp{#1}{#2}{#3}}}}
+ \newcommand*{\difffppAtZero}[4]{\ensuremath{\AtZero{\difffpp{#1}{#2}{#3}{#4}}}}
+
+ %-----------------------------------------------------------------------------
+ % Defined
+ % How should the defined operator look like (:= or ^= ==)
+ % (I want back my :=, it is so much better than ^= because (1) it has a
+ % direction and (2) everyone here uses it.)
+ %
+ % Use :=
+ %\newcommand*{\defined}{\ensuremath{\mathrel{\mathop{:}}=}}
+ %\newcommand*{\definedRight}{\ensuremath{=\mathrel{\mathop{:}}}}
+ % Use ^=
+ \newcommand*{\defined}{\ensuremath{\triangleq}}
+ \newcommand*{\definedRight}{\ensuremath{\triangleq}}
+ % Use = with three bars
+ %\newcommand*{\defined}{\ensuremath{?}}
+ %\newcommand*{\definedRight}{\ensuremath{?}}
+
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+ % Symbols used in the paper
+
+ %-----------------------------------------------------------------------------
+ % The Methods
+ \newcommand*{\ICIA}{\emph{ICIA}}
+ \newcommand*{\CoDe}{\emph{CoDe}}
+ \newcommand*{\LinCoDe}{\emph{LinCoDe}}
+ \newcommand*{\CoNe}{\emph{CoNe}}
+ \newcommand*{\CoLiNe}{\emph{CoLiNe}}
+ \newcommand*{\LinCoLiNe}{\emph{LinCoLiNe}}
+
+ % inter eye distance
+ \newcommand*{\ied}{IED}
+
+ %-----------------------------------------------------------------------------
+ % Koerper
+ %%\newcommand*{\RR}{\mathbb{R}}
+ %\newcommand*{\RR}{{I\hspace{-3.5pt}R}}
+ %\newcommand*{\RR}{{\mathrm{I\hspace{-2.7pt}R}}}
+
+ \font\dsfnt=dsrom12
+
+ \DeclareSymbolFont{nark}{U}{dsrom}{m}{n}
+ \DeclareMathSymbol{\NN}{\dsfnt}{nark}{`N}
+ \DeclareMathSymbol{\RR}{\dsfnt}{nark}{`R}
+ \DeclareMathSymbol{\ZZ}{\dsfnt}{nark}{`Z}
+
+ %-----------------------------------------------------------------------------
+ % Domains
+ \newcommand*{\D}{\mathcal{D}}
+ \newcommand*{\I}{\mathcal{I}}
+
+ %-----------------------------------------------------------------------------
+ % Texture coordinates
+ \newcommand*{\rr}{\VEC{r}}
+
+ %-----------------------------------------------------------------------------
+ % Parameters
+ \newcommand*{\pt}{\VEC{\tau}}
+ \newcommand*{\pr}{\VEC{\rho}}
+ \newcommand*{\pp}{\VEC{p}}
+ \newcommand*{\qq}{\VEC{q}}
+ \newcommand*{\xx}{\VEC{x}}
+ \newcommand*{\deltaq}{\Delta \qq}
+ \newcommand*{\deltap}{\Delta \pp}
+ \newcommand*{\zz}{\VEC{z}}
+ \newcommand*{\pa}{\VEC{\alpha}}
+ \newcommand*{\qa}{\VEC{\alpha}}
+ \newcommand*{\pb}{\VEC{\beta}}
+
+ %-----------------------------------------------------------------------------
+ % Optimal appearance parameters
+ \newcommand*{\pbh}[1]{\ensuremath{\hat{\pb}({#1})}}
+
+ %-----------------------------------------------------------------------------
+ % Warp basis
+ \newcommand*{\M}[1]{\ensuremath{M({#1})}}
+ \newcommand*{\LL}[1]{\ensuremath{L({#1})}}
+
+ %-----------------------------------------------------------------------------
+ % Matrices of the texture model
+ \newcommand*{\AM}[1]{\ensuremath{\Lambda(#1)}}               % Lambda(beta) 
+ \newcommand*{\AMr}[2]{\ensuremath{\Lambda(#1; #2)}}        % Lambda(r, beta)
+
+ \newcommand*{\As}{A}         % Continuous Basis symbol
+ \newcommand*{\afs}{a}        % Continuous mean symbol
+ \newcommand*{\A}[1]{\As(#1)}         % Continuous Basis
+ \newcommand*{\af}[1]{\afs(#1)}        % Continuous mean
+
+
+ %-----------------------------------------------------------------------------
+ % Matrices of the shape model
+ \newcommand*{\MU}{\VEC{\mu}}
+ \newcommand*{\MM}{\MAT{M}}
+
+ %-----------------------------------------------------------------------------
+ %% The project out matrix and operator
+ \newcommand*{\INT}{\MAT{P}}
+ \newcommand*{\INTf}{P}
+
+ %-----------------------------------------------------------------------------
+ % The identity matrix
+ \newcommand*{\EYEtwo}{\Matrix{1 & 0\\0&1}}
+ \newcommand*{\EYE}{\MAT E}
+ \newcommand*{\EYEf}{E}
+
+ % Wether to use subscripts or brackets for some function arguments
+ % can be decided by commenting out the corresponding functions underneath
+ %-----------------------------------------------------------------------------
+ % Mapping
+ \newcommand*{\Cs}[1]{\ensuremath{C^{#1}}} % C symbol
+ \newcommand*{\C}[2]{\ensuremath{C^{#1}(#2)}} % Use C with brackets
+
+ %-----------------------------------------------------------------------------
+ % Objective function
+ \newcommand*{\Fs}{\ensuremath{F}}              % F symbol
+ \newcommand*{\F}[1]{\ensuremath{\Fs(#1)}}       % Use F with brackets    F(q)
+
+ %-----------------------------------------------------------------------------
+ % Approximated objective functions
+ \newcommand*{\FFs}{\tilde{F}}                     % ~F symbol
+ \newcommand*{\FF}[1]{\ensuremath{\FFs(#1)}}       % Use ~F with brackets    F(q)
+
+ %-----------------------------------------------------------------------------
+ % residual function
+ \newcommand*{\es}{\ensuremath{f}}              % R symbol
+
+ \newcommand*{\e}[1]{\ensuremath{\es(#1)}}         % R(q)
+ \newcommand*{\er}[2]{\ensuremath{\es(#1; #2)}}    % R(r; q)
+
+ %-----------------------------------------------------------------------------
+ % Approximated residual functions
+ \newcommand*{\ees}{\tilde{f}}                       % ~R symbol
+ \newcommand*{\ee}[1]{\ensuremath{\ees(#1)}}       % ~R(q)
+ \newcommand*{\eer}[2]{\ensuremath{\ees(#2; #1)}}  % ~R(r; q)
+
+ %-----------------------------------------------------------------------------
+ % Warps
+ \newcommand*{\Vs}{\ensuremath{V}}
+ \newcommand*{\VLins}{\ensuremath{\Vs^{\text{Ortho}}}}
+ \newcommand{\VModels}{\ensuremath{\Vs^{\text{Model}}}}
+ \newcommand*{\Ws}{\ensuremath{W}}
+
+ \newcommand{\V}[1]{\ensuremath{\Vs(#1)}}
+ \newcommand{\VModel}[1]{\ensuremath{\VModels(#1)}}
+ \newcommand{\Vr}[2]{\ensuremath{\Vs(#1; #2)}}
+ \newcommand{\VInvr}[2]{\ensuremath{\Vs^{-1}(#1; #2)}}
+ \newcommand{\VrLin}[2]{\ensuremath{\VLins(#1; #2)}}
+ \newcommand{\W}[1]{\ensuremath{\Ws(#1)}}
+ \newcommand{\Winv}[1]{\ensuremath{\Ws^{-1}(#1)}}
+ \newcommand{\Wr}[2]{\ensuremath{\Ws(#1; #2)}}
+
+
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+%% Begin of Document
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+\begin{document}
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+%% Here starts the poster
+%%---------------------------------------------------------------------------
+%% Format it to your taste with the options
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+\begin{poster}{
+ % Show grid to help with alignment
+ grid=false,
+ % Column spacing
+ colspacing=0.7em,
+ % Color style
+ headerColorOne=cyan!20!white!90!black,
+ borderColor=cyan!30!white!90!black,
+ % Format of textbox
+ textborder=faded,
+ % Format of text header
+ headerborder=open,
+ headershape=roundedright,
+ headershade=plain,
+ background=none,
+ bgColorOne=cyan!10!white,
+ headerheight=0.12\textheight}
+ % Eye Catcher
+ {
+      \includegraphics[width=0.08\linewidth]{track_frame_00010_06}
+      \includegraphics[width=0.08\linewidth]{track_frame_00450_06}
+      \includegraphics[width=0.08\linewidth]{track_frame_04999_06}
+ }
+ % Title
+ {\sc\Huge On Compositional Image Alignment}
+ % Authors
+ {Brian Amberg, Andrew Blake, and Thomas Vetter\\[1em]
+ {\texttt{Brian.Amberg@unibas.ch, ab@microsoft.com, Thomas.Vetter@unibas.ch}}}
+ % University logo
+ {
+  \begin{tabular}{r}
+    \includegraphics[height=0.12\textheight]{logo}\\
+    \raisebox{0em}[0em][0em]{\includegraphics[height=0.03\textheight]{msrlogo}}
+  \end{tabular}
+ }
+
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+%%% Now define the boxes that make up the poster
+%%%---------------------------------------------------------------------------
+%%% Each box has a name and can be placed absolutely or relatively.
+%%% The only inconvenience is that you can only specify a relative position 
+%%% towards an already declared box. So if you have a box attached to the 
+%%% bottom, one to the top and a third one which should be inbetween, you 
+%%% have to specify the top and bottom boxes before you specify the middle 
+%%% box.
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+  \headerbox{Contribution: Fast \emph{and} reliable Face Alignment}{name=contribution,column=0,row=0,span=2}{
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+  Inverse compositional image alignment (ICIA) is fast, but not reliable. We
+  explain ICIA from a different perspective which leads naturally to two new
+  algorithms with a better capture range and comparable speed.
+  }
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+  \headerbox{There is no \emph{inverse} in \emph{ICIA}}{name=abstract,column=0,below=contribution}{
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+    Image aligment minimizes
+    \begin{align}
+      \F{\qq} &\defined \normLR{ \e{\qq,\pb} }^2_{\D},\label{eqn:f}\\
+    \text{with }  \e{\qq} &\defined a - I\circ \W{\qq}\nonumber
+    \end{align}
+    composition with an incremental warp $\Vs$ approximates $\Fs$ around $\qq_0$ as
+    \begin{align}
+      \F{\C{\circ}{\qq_0, \pp}} \approx \FF{\qq_0, \pp} & \defined \normLR{\ee{\qq_0,\pp}}^2_{\D}\label{eqn:comp}\\
+      \text{with } \ee{\qq_0, \pp}&\defined  \INTf(a -  I \circ \W{\qq_0} \circ \V{\pp}) \quad.\nonumber
+    \end{align}
+    The gradient descent or Gauss-Newton update rule then gives an estimate of
+    the incremental warp, which drives the model warp.
+
+    ICIA can be derived by substituting the current backwarped image with the
+    model appearance after taking the derivative. The substitution can be used
+    to get an approximate gradient and/or Hessian, leading to a family of
+    algorithms.
+
+    Additionally we replace the incremental warp $V$ with an orthonormalized
+    warp and regularize in the composition step. The result is a vast
+    improvement in robustness without sacrificing speed.
+}
+
+
+
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+   \headerbox{Our methods are at the performance/speed sweet point}{name=speed,column=2,row=0,span=2}{
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+   \newlength{\MSZ}
+   \setlength{\MSZ}{0.01\textwidth}
+   \newcommand{\MarkerCircle}[1]{%
+     \tikz{\draw[use as bounding box] (0,0); \draw[fill=#1]             circle(\MSZ);}}
+   \newcommand{\MarkerRectangle}[1]{%
+     \tikz{\draw[use as bounding box] (0,0); \draw[fill=#1]             (-\MSZ,-\MSZ) rectangle +(2\MSZ,2\MSZ);}}
+   \newcommand{\MarkerDiamond}[1]{%
+     \tikz{\draw[use as bounding box] (0,0); \draw[fill=#1,rotate=45]   rectangle +(2\MSZ,2\MSZ);}}
+   \newcommand{\MarkerTriangle}[1]{%
+     \tikz{\draw[use as bounding box] (0,0); \draw[fill=#1]             (-0.866\MSZ,-0.5\MSZ) -- (0\MSZ,1\MSZ) -- (0.866\MSZ,-0.5\MSZ) -- cycle ;}}
+   \newcommand{\MarkerUDTriangle}[1]{%
+     \tikz{\draw[use as bounding box] (0,0); \draw[fill=#1,rotate=180]  (-0.866\MSZ,-0.5\MSZ) -- (0\MSZ,1\MSZ) -- (0.866\MSZ,-0.5\MSZ) -- cycle ;}}
+   \newcommand{\MarkerPlus}[1]{%
+     \tikz{\draw[use as bounding box] (0,0); \draw[fill=#1]         (-\MSZ,-0.25\MSZ) rectangle +(2\MSZ,0.5\MSZ) (-0.25\MSZ,-\MSZ) rectangle +(0.5\MSZ,2\MSZ);}}
+   \newcommand{\MarkerX}[1]{%
+     \tikz{\draw[use as bounding box] (0,0); \draw[fill=#1,rotate=45]         (-\MSZ,-0.25\MSZ) rectangle +(2\MSZ,0.5\MSZ) (-0.25\MSZ,-\MSZ) rectangle +(0.5\MSZ,2\MSZ);}}
+   \begin{tikzpicture}[x=0.00425\linewidth,y=13mm,font=\smaller]
+     % Ticks
+     \begin{scope}[color=black]
+       \foreach \y in {-0.2218487496,-0.1549019600,-0.0969100130,-0.0457574906,0.0000000000,
+                       0.0000000000,0.3010299957,0.4771212547,0.6020599913,0.6989700043,0.7781512504,0.8450980400,0.9030899870,0.9542425094,1.0000000000,
+                       1.0000000000,1.3010299957,1.4771212547,1.6020599913,1.6989700043,1.7781512504,1.8450980400,1.9030899870,1.9542425094,2.0000000000,
+                       2.0000000000,2.3010299957,2.4771212547,2.6020599913,2.6989700043,2.7781512504,2.8450980400,2.9030899870,2.9542425094,3.0000000000} {
+         \draw[color=lightgray!20!white] (0,\y) -- +(100,0); 
+       }
+       \foreach \y/\lbl in {0/$10^0$,1/$10^1$,2/$10^2$,3/$10^3$} {
+         \draw[color=black]     (0,\y) node[anchor=east,color=black] {\lbl} -- +(100,0); 
+         \draw[color=lightgray] (0,\y) -- +(100,0); 
+         \draw[color=black]     (0,\y) -- +(0.5\MSZ,0mm); 
+       }
+       \foreach \x in {0,20,40,60,80,100} {
+         \draw (\x,-0.23) node[anchor=north,color=black] {\x};
+         \draw[color=lightgray] (\x,-0.23) -- +(0,3.23);
+         \draw[color=black] (\x,-0.23) -- +(0mm,0.5\MSZ);
+       }
+     \end{scope}
+     % Border
+     \draw[color=black] (0,3) -- (0,-0.23)  (100,-0.23) -- (100,3);
+     % Axis labels
+     \draw (50,-0.23)  node[below,yshift=-1em]{Success Rate (\%, Larger is better)};
+     \draw (0,1.5) node[above,rotate=90,yshift=1.6em]{Runtime (smaller is better)};
+     \draw (50,3)  node[above]{The main algorithms starting within $20\%$ \ied{}};
+     % Data
+     \foreach \anch/\rot/\xs/\ys/\x/\y/\ttl/\stl in {
+     west/  0  / 1   / 0   /  5.2631578947   /     0.0000000000 /  Original \ICIA{}        / {\MarkerCircle{blue}},
+   %  west/  0  / 1   / 0   / 14.2369727047   /     0.1128364125 /  \ICIA{} + V^{\text{norm}}/ {\MarkerCircle{blue!70!white}},
+     east/  0  /-1   / 0   / 38.9423076923   /     0.8814699511 /  \CoLiNe{}                / {\MarkerRectangle{green}},
+     west/  0  / 1   / 0   / 39.4696029777   /     0.2610806897 /  \LinCoDe{}               / {\MarkerTriangle{red}},
+     west/  0  / 1   / 0   / 56.8548387097   /     0.9091243545 /  \CoDe{}                  / {\MarkerUDTriangle{yellow}},
+     west/  0  / 1   / 0   / 36.4299007444   /     1.3563671940 /  \CoNe{}                  / {\MarkerX{black!50!white}},
+     west/  0  / 1   / 0   / 41.6918429003   /     2.6132022326 /  L-BFGS (with reg)     / {\MarkerPlus{red!50!blue!50!black}}
+     }{
+       \draw (\x,\y) 
+         node[fill=none,anchor=\anch,xshift=\xs\MSZ,yshift=\ys\MSZ,rotate=\rot] {\ttl} 
+         node{\stl};
+       \draw[fill=black] (\x,\y) circle(0.3\MSZ);
+     }
+   \end{tikzpicture}
+   \begin{tikzpicture}[x=0.00425\linewidth,y=13mm,font=\smaller]
+     % Ticks
+     \begin{scope}[color=black]
+       \foreach \y in {-0.2218487496,-0.1549019600,-0.0969100130,-0.0457574906,0.0000000000,
+                       0.0000000000,0.3010299957,0.4771212547,0.6020599913,0.6989700043,0.7781512504,0.8450980400,0.9030899870,0.9542425094,1.0000000000,
+                       1.0000000000,1.3010299957,1.4771212547,1.6020599913,1.6989700043,1.7781512504,1.8450980400,1.9030899870,1.9542425094,2.0000000000,
+                       2.0000000000,2.3010299957,2.4771212547,2.6020599913,2.6989700043,2.7781512504,2.8450980400,2.9030899870,2.9542425094,3.0000000000} {
+         \draw[color=lightgray!20!white] (0,\y) -- +(100,0); 
+       }
+       \foreach \y/\lbl in {0/$10^0$,1/$10^1$,2/$10^2$,3/$10^3$} {
+         \draw[color=black]     (0,\y) node[anchor=east,color=black] {\lbl} -- +(100,0); 
+         \draw[color=lightgray] (0,\y) -- +(100,0); 
+         \draw[color=black]     (0,\y) -- +(0.5\MSZ,0mm); 
+       }
+       \foreach \x in {0,20,40,60,80,100} {
+         \draw (\x,-0.23) node[anchor=north,color=black] {\x};
+         \draw[color=lightgray] (\x,-0.23) -- +(0,3.23);
+         \draw[color=black] (\x,-0.23) -- +(0mm,0.5\MSZ);
+       }
+     \end{scope}
+     % Border
+     \draw[color=black] (0,3) -- (0,-0.23)  (100,-0.23) -- (100,3);
+     % Axis labels
+     \draw (50,-0.23)  node[below,yshift=-1em]{Success Rate (\%, Larger is better)};
+     \draw (0,1.5) node[above,rotate=90,yshift=1.6em]{Runtime (smaller is better)};
+     \draw (50,3)  node[above,text width=0.5\linewidth,text centered]{All algorithms with $V^{\text{norm}}$ and regularisation};
+
+     \foreach \anch/\rot/\xs/\ys/\x/\y/\ttl/\stl in {
+       south/0 / 0   /  1.0/       26.4701318852 /         0.6174735535 / \ICIA{} + $V^{\text{norm}}$ / {\MarkerCircle{    blue!50!black}},
+       south/0 / 0   /  1.0/       52.1334367727 /         1.0781366795 / \CoLiNe{}   / {\MarkerRectangle{ green!50!black}},
+       west/ 0 / 1   /  0  /       55.9348332040 /         0.5514820720 / \LinCoDe{}  / {\MarkerTriangle{  red!50!black}},
+       west/ 0 / 1   /  0  /       66.0356865787 /         1.1158139989 / \CoDe{}     / {\MarkerUDTriangle{yellow!50!black}},
+       west/ 0 / 1   /  0  /       66.9511249030 /         1.8260835334 / \CoNe{}     / {\MarkerX{         black!50!white!50!black}},
+       west/ 0 / 1   /  0  /       41.6918429003 /         2.6132022326 / L-BFGS   / {\MarkerPlus{      red!50!blue!50!black}}
+     }{
+       \draw (\x,\y) 
+         node[fill=none,anchor=\anch,xshift=\xs\MSZ,yshift=\ys\MSZ,rotate=\rot] {\ttl} 
+         node{\stl};
+       \draw[fill=black] (\x,\y) circle(0.3\MSZ);
+     }
+   \end{tikzpicture}
+   \begin{multicols}{2}
+   \textbf{Fitting a multiperson AAM. }
+   The best speed--performance tradeoffs come from the two new algorithms
+   \CoDe{} and \LinCoDe{}. Note that \ICIA{} is practically useless on
+   this difficult multi-person dataset with a success rate near zero (left). It
+   can be improved (right) by using the orthonormal incremental warp and
+   regularisation. The \CoDe{} algorithm with regularisation (right) is as
+   accurate as the slow, approximation-free, compositional Gauss-Newton \CoNe{}
+   method but is seven times more efficient.
+
+   The experiments were performed with leave one identity out on a mixture of two databases (XM2VTS and IMM).
+   \end{multicols}
+   }
+%
+% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+%   \headerbox{Methods Compared}{name=methods,column=0,below=algorithm}{
+% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+%   \begin{tabular}{rllllll}
+%     Method                              & Hessian               &                                        & Gradient        &                                  & Speed      & Capture Range\\
+%     \midrule
+% \CoDe{} (this paper)                & Not used              &                                        & True:           & $\tilde{J}_{\qq_0}^T\e{\qq_0}$   & Fast       & Large  \\[0.1em]
+% \LinCoDe{} (this paper)             & Not used              &                                        & Linear Approx:  & $\bar{J}^T\e{\qq_0}$             & Very Fast  & Medium \\[0.1em]
+% \CoLiNe{}~\cite{burkhardt86:motion} & Constant Approx.:     & $\bar{J}^T\bar{J}$                     & True:           & $\tilde{J}_{\qq_0}^T\e{\qq_0}$   & Fast       & Medium \\[0.1em]
+% \ICIA{}~\cite{matthews:aamr}        & Constant Approx.:     & $\bar{J}^T\bar{J}$                     & Linear Approx:  & $\bar{J}^T\e{\qq_0}$             & Very Fast  & Small  \\[0.1em]
+% \CoNe{}~\cite{matthews:kanade20}    & Gauss-Newton Approx.: & $\tilde{J}_{\qq_0}^T\tilde{J}_{\qq_0}$ & True:           & $\tilde{J}_{\qq_0}^T\e{\qq_0}$   & Slow       & Large  
+%   \end{tabular}
+%   The methods introduced in this paper are Hessian-free gradient descent methods.
+%  }
+%
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+   \headerbox{References}{name=references,column=0,above=bottom}{
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+     \smaller
+     
+     \bibliographystyle{ieee}
+     \renewcommand{\section}[2]{\vskip 0.05em}
+       \begin{thebibliography}{1}\itemsep=-0.01em
+       \setlength{\baselineskip}{0.4em}
+
+       \bibitem{amberg07:nicp}
+       B.~Amberg, A.~Blake, T.~Vetter
+       \newblock On Compositional Image Alignment with an Application to Activce Appearance Models
+       \newblock In {\em CVPR'09}, 2009.
+
+       \end{thebibliography}
+   }
+
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+   \headerbox{Training + Testing Data}{name=data,column=0,above=references,below=abstract}{
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+   \includegraphics[width=0.2\linewidth]{018_4_2_masked}%
+   \includegraphics[width=0.2\linewidth]{328_2_1_masked}%
+   \includegraphics[width=0.2\linewidth]{319_2_1_masked}%
+   \includegraphics[width=0.2\linewidth]{027_4_2_masked}%
+   \includegraphics[width=0.2\linewidth]{020_1_1_masked}
+   \includegraphics[width=0.2\linewidth]{12_2f_masked}%
+   \includegraphics[width=0.2\linewidth]{21_3m_masked}%
+   \includegraphics[width=0.2\linewidth]{09_6m_masked}%
+   \includegraphics[width=0.2\linewidth]{33_4m_masked}%
+   %\includegraphics[width=0.2\linewidth]{22_3f_masked}
+   $\dots$\\
+   The model was trained from 456 images from the IMM and XM2VTS datasets using
+   120 landmarks. Get the landmarks, model, and source code at:\\
+   \mbox{\url{www.cs.unibas.ch/personen/amberg_brian/aam/}}
+   }
+
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+
+   \headerbox{Tracking 5000 frames with a general model}{name=tracking,column=2,span=2,below=speed,above=bottom}{
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+ {
+ \begin{tabular}{c@{\hspace{0.05em}}c@{\hspace{0.1em}}c@{\hspace{0.1em}}c@{\hspace{0.1em}}c@{\hspace{1em}}c@{\hspace{0.1em}}c@{\hspace{0.1em}}c@{\hspace{0.1em}}c@{\hspace{0.1em}}c}
+   \multicolumn{5}{c}{\smaller \ICIA{} with $\VLins$} &
+   \multicolumn{5}{c}{\smaller \ICIA{} with $\VLins$ + Regularisation}\\[-0.2em]
+   \includegraphics[width=0.095\linewidth]{track_frame_00010_01}&
+   \includegraphics[width=0.095\linewidth]{track_frame_00050_01}&
+   \includegraphics[width=0.095\linewidth]{track_frame_00450_01}&
+   \includegraphics[width=0.095\linewidth]{track_frame_02000_01}&
+   \includegraphics[width=0.095\linewidth]{track_frame_04999_01}&
+   %
+   \includegraphics[width=0.095\linewidth]{track_frame_00010_02}&
+   \includegraphics[width=0.095\linewidth]{track_frame_00050_02}&
+   \includegraphics[width=0.095\linewidth]{track_frame_00450_02}&
+   \includegraphics[width=0.095\linewidth]{track_frame_02000_02}&
+   \includegraphics[width=0.095\linewidth]{track_frame_04999_02}\\[-0.1em]
+   %
+   \multicolumn{5}{c}{\smaller \LinCoDe{}} &
+   \multicolumn{5}{c}{\smaller \LinCoDe{} + Regularisation}\\[-0.2em]
+   \includegraphics[width=0.095\linewidth]{track_frame_00010_03}&
+   \includegraphics[width=0.095\linewidth]{track_frame_00050_03}&
+   \includegraphics[width=0.095\linewidth]{track_frame_00450_03}&
+   \includegraphics[width=0.095\linewidth]{track_frame_02000_03}&
+   \includegraphics[width=0.095\linewidth]{track_frame_04999_03}&
+   %
+   \includegraphics[width=0.095\linewidth]{track_frame_00010_04}&
+   \includegraphics[width=0.095\linewidth]{track_frame_00050_04}&
+   \includegraphics[width=0.095\linewidth]{track_frame_00450_04}&
+   \includegraphics[width=0.095\linewidth]{track_frame_02000_04}&
+   \includegraphics[width=0.095\linewidth]{track_frame_04999_04}\\[-0.1em]
+   %
+   \multicolumn{5}{c}{\smaller \CoDe{}} &
+   \multicolumn{5}{c}{\smaller \CoDe{} + Regularisation}\\[-0.2em]
+   \includegraphics[width=0.095\linewidth]{track_frame_00010_05}&
+   \includegraphics[width=0.095\linewidth]{track_frame_00050_05}&
+   \includegraphics[width=0.095\linewidth]{track_frame_00450_05}&
+   \includegraphics[width=0.095\linewidth]{track_frame_02000_05}&
+   \includegraphics[width=0.095\linewidth]{track_frame_04999_05}&
+   %
+   \includegraphics[width=0.095\linewidth]{track_frame_00010_06}&
+   \includegraphics[width=0.095\linewidth]{track_frame_00050_06}&
+   \includegraphics[width=0.095\linewidth]{track_frame_00450_06}&
+   \includegraphics[width=0.095\linewidth]{track_frame_02000_06}&
+   \includegraphics[width=0.095\linewidth]{track_frame_04999_06}\\[-0.5em]
+   \smaller Frame 10 & \smaller Frame 50 & \smaller Frame 450 & \smaller Frame 2000 & \smaller Frame 5000 &
+   \smaller Frame 10 & \smaller Frame 50 & \smaller Frame 450 & \smaller Frame 2000 & \smaller Frame 5000
+   \end{tabular}
+ }
+   \vspace{-1.2em}
+   \begin{multicols}{2}
+   {\textbf{Our algorithm makes fast and robust tracking possible.}
+     We compare face tracking under natural motion, using \ICIA{},
+     \LinCoDe{} and \CoDe{}. The original \ICIA{} fails
+     immediately with this large model and new face data. Substituting the orthonormal
+     incremental warp for the original \ICIA{} warp, the algorithm still loses track
+     very early, whereas \LinCoDe{} and \CoDe{} can track much
+     further. Finally, adding regularisation to all algorithms, \ICIA{} still
+     loses track completely after approximately 500 frames and does not recover
+     the local deformations accurately. In contrast \CoDe{} now tracks the full
+     5000 frame sequence without reinitialization, and \LinCoDe{} tracks for 2500 frames.}
+   
+   The same training dataset was used for both tracking experiments. The
+   training data was aquired with different camera and light settings from
+   different subjects.
+   \end{multicols}
+   }
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+   \headerbox{Low Res Tracking}{name=lowrestracking,column=1,span=1,below=speed,above=bottom}{
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+\begin{tabular}{@{}c@{}c@{}c@{}c@{}c@{}}
+  \multicolumn{5}{c}{\smaller \ICIA{} with $\VLins$}\\[-0.2em]
+  \includegraphics[width=0.2\linewidth]{bush_00010_02}&
+  \includegraphics[width=0.2\linewidth]{bush_00100_02}&
+  \includegraphics[width=0.2\linewidth]{bush_00200_02}&
+  \includegraphics[width=0.2\linewidth]{bush_00300_02}&
+  \includegraphics[width=0.2\linewidth]{bush_00400_02}\\[-0.1em]
+  \multicolumn{5}{c}{\smaller \LinCoDe{}}\\[-0.2em]
+  \includegraphics[width=0.2\linewidth]{bush_00010_05}&
+  \includegraphics[width=0.2\linewidth]{bush_00100_05}&
+  \includegraphics[width=0.2\linewidth]{bush_00200_05}&
+  \includegraphics[width=0.2\linewidth]{bush_00300_05}&
+  \includegraphics[width=0.2\linewidth]{bush_00400_05}\\[-0.1em]
+  \multicolumn{5}{c}{\smaller \CoDe{}}\\[-0.2em]
+  \includegraphics[width=0.2\linewidth]{bush_00010_08}&
+  \includegraphics[width=0.2\linewidth]{bush_00100_08}&
+  \includegraphics[width=0.2\linewidth]{bush_00200_08}&
+  \includegraphics[width=0.2\linewidth]{bush_00300_08}&
+  \includegraphics[width=0.2\linewidth]{bush_00400_08}\\[-0.5em]
+  \smaller Frame 10 & \smaller Frame 100 & \smaller Frame 200 & \smaller Frame 300 & \smaller Frame 400 
+  \end{tabular}
+
+  \vspace{1.25em}
+  \textbf{Tracking a low resolution video with large head motions
+  succeeds with \CoDe{}, where \ICIA{} fails.}\\ All methods used the orthonormal
+  incremental warp, and relatively strong regularisation.  \ICIA{} starts to
+  drift in the early frames, while~\CoDe{} tracks the full sequence. The
+  approximate gradient method \LinCoDe{} also suceeds, but looses
+  track of the details for about 100 frames.
+   }
+
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+  \headerbox{Compositional Alignment}{name=algorithm,column=1,above=lowrestracking,below=contribution}{
+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+  \begin{algorithm}[H]
+    \dontprintsemicolon
+    \linesnumbered
+    \For{Blur and regularisation values}{
+      \nl Initialize $\qq, \qq_{\text{best}}$ and $\kappa$\;
+      \Repeat{converged}{
+        \nl Calculate $\Nabla{\pp}{\tilde{F}(\qq,\VEC 0)}$, $F(\qq)$\;
+        \eIf{$F(\qq) < F(\qq_{\text{best}})$}{
+          %\nl Calculate distance between best warp estimate and current warp estimate to test for convergence\;
+          \nl $\qq_{\text{best}} \gets \qq$\;
+          %\If{More than three successiv updates}{ (Too much detail)
+            \nl Increase $\kappa$\;
+          %}
+        }{
+          \If{$\kappa$ smaller than threshold}{
+            \nl return\;
+          }%{
+             decrease $\kappa$\;
+          %}
+        }
+        \nl Calculate $\pp$ from $\Nabla{\pp}{\tilde{F}(\qq_{best},\pp)}$ and $\kappa$\;
+        \nl $\qq \gets \C{\circ}{\qq, \pp}$
+      }
+    }
+  \end{algorithm}
+   }
+\end{poster}%
+%
+\end{document}