-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpresentation.tex
388 lines (362 loc) · 19.5 KB
/
presentation.tex
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
\documentclass[10pt, export]{beamer}
\usepackage[utf8]{inputenc}
\usepackage[T1]{fontenc}
\usepackage{standalone}
\usepackage[acronym]{glossaries}
\usepackage{enumitem}
\usepackage{multirow}
\usepackage{multicol}
\usepackage{array}
\newcolumntype{x}[1]{>{\centering\let\newline\\\arraybackslash\hspace{0pt}}p{#1}}
\usepackage{booktabs, colortbl}
\usepackage{siunitx}
\usepackage{mathrsfs, amsmath}
\usepackage{graphicx}
\usepackage{subcaption}
\usepackage[font={small, color=IGNDarkGrey}, labelformat=empty]{caption}
\DeclareCaptionFont{footnotesize}{\footnotesize}
\usepackage[font={footnotesize, color=IGNDarkGrey}, labelformat=empty]{subcaption}
\usepackage{adjustbox}
\usepackage{tikz}
\usetikzlibrary{calc, 3d, shapes, arrows, shadows, decorations}
\usepackage{pgfplots, pgfplotstable}
\usepackage{pifont, textcomp}
\newcommand{\cmark}{{\color{green} \ding{51}}}%
\newcommand{\xmark}{{\color{red} \ding{55}}}%
\definecolor{Moins4}{RGB}{183,10,10}
\definecolor{Moins3}{RGB}{255,0,0}
\definecolor{Moins2}{RGB}{255,122,12}
\definecolor{Moins1}{RGB}{255,181,12}
\definecolor{Zero}{RGB}{255,255,255}
\definecolor{Plus1}{RGB}{58,255,12}
\definecolor{Plus2}{RGB}{58,155,12}
\usepackage{hyperref}
\usepackage[useregional]{datetime2}
\usepackage[
backend=biber,
style=alphabetic,
backref=true,
sorting=none,
autocite=footnote,
maxnames=2,
hyperref=true,
natbib=false,
abbreviate=false,
doi=false,
isbn=false,
url=false
]{biblatex}
\addbibresource{references.bib}
\setbeamerfont{footnote}{size=\tiny}
\usetheme{ign}
\newacronym{acr::lod}{LoD}{Level of Detail}
\newacronym{acr::elod}{eLoD}{evaluation Level of Detail}
\newacronym{acr::lidar}{LiDAR}{Light Detection and Ranging}
\newacronym{acr::dsm}{DSM}{Digital Surface Model}
\newacronym{acr::gui}{GUI}{Graphical User Interface}
\title{The necessary yet complex evaluation of 3D city models}
\subtitle{a semantic approach}
\date{\tiny \DTMdisplaydate{2019}{5}{27}{1}}
\author{
\scriptsize \underline{\textbf{Oussama Ennafii}} \inst{1, 2}, Arnaud Le Bris \inst{1}, Cl\'ement Mallet \inst{1}, Florent Lafarge \inst{2}
}
\institute{
\scriptsize \inst{1} Univ. Paris Est, LaSTIG STRUDEL, IGN, ENSG\\
\inst{1} name.surname@ign.fr\\
~\\
\scriptsize \inst{2} Inria, TITANE\\
\inst{2} name.surname@inria.fr
}
\begin{document}
\begin{frame}[plain]
\titlepage{}
\end{frame}
\section{Introduction}
\begin{frame}{Motivation}
\begin{figure}
\begin{center}
\includestandalone[mode=buildnew, width=\textwidth]{context}
\end{center}
\end{figure}
\end{frame}
\begin{frame}{3D representation of buildings}
\begin{figure}
\begin{center}
\includestandalone[mode=buildnew, width=.9\textwidth]{building_models-manhattan}
\end{center}
\end{figure}
\end{frame}
\begin{frame}{3D representation of buildings}
\begin{figure}
\begin{center}
\includestandalone[mode=buildnew, width=.9\textwidth]{building_models-piecewise}
\end{center}
\end{figure}
\end{frame}
\begin{frame}{3D representation of buildings}
\begin{figure}
\begin{center}
\includestandalone[mode=buildnew, width=.9\textwidth]{building_models-grammar}
\end{center}
\end{figure}
\end{frame}
\begin{frame}{3D representation of buildings}
\begin{figure}
\begin{center}
\includestandalone[mode=buildnew, width=.9\textwidth]{building_models-all}
\end{center}
\end{figure}
\end{frame}
\begin{frame}{Problem statement}
\begin{block}{\textcolor{purple}{\textbf{Goal}}}
\underline{Detect} and \underline{describe} semantic errors that affect building 3D models.
\end{block}
~\\
\begin{itemize}[label=$\blacktriangleright$, font=\color{IGNGreen}, itemsep=2em]
\item<2-> Semantic errors \textbf{independent} from \underline{reconstruction methods} and \underline{urban scenes}.
\item<3-> \underline{Transferability}, and \underline{scalability}, of the evaluation method.
\end{itemize}
\end{frame}
\section{Methodology}
\begin{frame}{Main ideas behind our approach}
\begin{itemize}[label=$\blacktriangleright$, font=\color{IGNGreen}, itemsep=2em]
\item<1-> \underline{Compile} errors that affect building models in a \underline{taxonomy};
\item<2-> Evaluation at \underline{building level};
\item<3-> \underline{No reference} 3D models $\longrightarrow$ a \underline{supervised} classification problem;
\item<4-> Study in a \underline{2.5D overhead (aerial)} modeling setting.
\end{itemize}
\end{frame}
\begin{frame}{Taxonomy structure}
Two criteria determine the taxonomy structure:
\begin{itemize}[label=$\blacktriangleright$, font=\color{IGNGreen}, itemsep=2em]
\item<2-> the \textbf{\acrfull{acr::lod}};
\begin{figure}
\begin{center}
\includestandalone[mode=buildmissing, width=\textwidth]{lods}
\end{center}
\end{figure}
\item<3-> the \textbf{finesse}: the semantic evaluation level.
\end{itemize}
\end{frame}
\begin{frame}{Error taxonomy (\texttt{finesse} $= 0$)}
\begin{figure}
\includestandalone[mode=buildmissing, width=\textwidth]{finesse_0_taxonomy}
\end{figure}
\end{frame}
\begin{frame}{Error taxonomy (\texttt{finesse} $= 1$)}
\begin{figure}
\includestandalone[mode=buildmissing, width=\textwidth]{finesse_1_taxonomy}
\end{figure}
\end{frame}
\begin{frame}{Error taxonomy (\texttt{finesse} $= 2$)}
\only<1|handout:0>{
\begin{figure}
\includestandalone[mode=buildmissing, width=\textwidth]{finesse_2_taxonomy}
\end{figure}
}
\only<2>{
\begin{figure}
\includestandalone[mode=buildmissing, width=\textwidth]{finesse_2_taxonomy_lods}
\end{figure}
}
\end{frame}
\begin{frame}{Error taxonomy (\texttt{finesse} $= 3$, $\leq$ \acrshort{acr::lod}-1)}
\begin{figure}
\includestandalone[mode=buildmissing, width=\textwidth]{finesse_3_bul_taxonomy}
\end{figure}
\end{frame}
\begin{frame}{Building \texttt{atomic} errors: 2.5D overhead reconstruction case}
\only<1|handout:1>{
\begin{figure}
\begin{center}
\begin{subfigure}{.4\textwidth}
\centering
\fbox{\includegraphics[height=.45\textheight]{images/errors/building/under_segmentation}}
\caption{\label{fig::bul_under} Under segmentation (\texttt{BUS}).}
\end{subfigure}
\begin{subfigure}{.5\textwidth}
\centering
\fbox{\includegraphics[height=.45\textheight]{images/errors/building/over_segmentation}}
\caption{\label{fig::bul_over} Over segmentation (\texttt{BOS}).}
\end{subfigure}
\end{center}
\end{figure}
}
\only<2|handout:2>{
\begin{figure}
\begin{center}
\begin{subfigure}{.45\textwidth}
\centering
\fbox{\includegraphics[height=.5\textheight]{images/errors/building/border}}
\caption{\label{fig::bul_footprint} Imprecise border (\texttt{BIB}).}
\end{subfigure}
\begin{subfigure}{.45\textwidth}
\centering
\fbox{\includegraphics[height=.5\textheight]{images/errors/building/topology}}
\caption{\label{fig::bul_height} Inaccurate topology (\texttt{BIT}).}
\end{subfigure}
\end{center}
\end{figure}
}
\end{frame}
\begin{frame}{Error taxonomy (\texttt{finesse} $= 3$, $\leq$ \acrshort{acr::lod}-2)}
\begin{figure}
\includestandalone[mode=buildmissing, width=\textwidth]{finesse_3_all_taxonomy}
\end{figure}
\end{frame}
\begin{frame}{Facet \texttt{atomic} errors: 2.5D overhead reconstruction case}
\only<1|handout:1>{
\begin{figure}
\begin{center}
\begin{subfigure}{.31\textwidth}
\centering
\fbox{\includegraphics[height=.4\textheight]{images/errors/facet/under_segmentation}}
\caption{\label{fig::fac_under} Under segmentation (\texttt{FUS})}
\end{subfigure}
\begin{subfigure}{.31\textwidth}
\centering
\fbox{\includegraphics[height=.4\textheight]{images/errors/facet/over_segmentation}}
\caption{\label{fig::fac_over} Over segmentation (\texttt{FOS})}
\end{subfigure}
\begin{subfigure}{.31\textwidth}
\centering
\fbox{\includegraphics[height=.4\textheight]{images/errors/facet/geometry}}
\caption{\label{fig::fac_height} Imprecise geometry (\texttt{FIG})}
\end{subfigure}
\end{center}
\end{figure}
}
\only<2|handout:2>{
\begin{figure}
\begin{center}
\begin{subfigure}{.45\textwidth}
\centering
\fbox{\includegraphics[height=.65\textwidth]{images/errors/facet/border}}
\caption{\label{fig::fac_footprint} Imprecise borders (\texttt{FIB})}
\end{subfigure}
\begin{subfigure}{.45\textwidth}
\centering
\fbox{\includegraphics[height=.65\textwidth]{images/errors/facet/topology}}
\caption{\label{fig::fac_height} Inaccurate topology (\texttt{FIT})}
\end{subfigure}
\end{center}
\end{figure}
}
\end{frame}
\begin{frame}{The evaluation pipeline}
\begin{figure}
\includestandalone[mode=buildnew, height=.81\textheight]{graphical_pipeline}
\end{figure}
\end{frame}
\section{Experiments}
\begin{frame}{Datasets: urban scenes}
\begin{figure}
\begin{center}
\includestandalone[mode=buildnew, width=.9\textwidth]{dataset}
\end{center}
\end{figure}
\end{frame}
\begin{frame}{Ablation study: Elancourt example}
\begin{table}
\scriptsize
\begin{center}
\scriptsize
\begin{tabular}{|x{.5cm} | x{.8cm} x{.8cm} | x{.9cm} x{.9cm} | x{.8cm} x{.8cm} | x{.8cm} x{.8cm}|}
\hline
&\multicolumn{2}{x{1.6cm}|}{\textbf{Geom.}} & \multicolumn{2}{x{1.8cm}|}{\textbf{Geom. $\cup$ Hei.}} & \multicolumn{2}{x{1.6cm}|}{\textbf{Geom. $\cup$ Im.}} & \multicolumn{2}{x{1.6cm}|}{\textbf{All}}\\
\cline{2-9}
&\textbf{Rec.} & \textbf{Prec.} & \textbf{Rec.} & \textbf{Prec.} & \textbf{Rec.} & \textbf{Prec.} & \textbf{Rec.} & \textbf{Prec.}\\
\hline
\texttt{BOS} & \textbf{93.96} & 76.15 & 91.43 & \textbf{77.76} & 91.51 & 76.08 & 90.83 & 76.14 \\
\hline
\strut\\[-\normalbaselineskip]
\only<3|handout:0>{\\[-\normalbaselineskip]\rowcolor{IGNGreen}}\texttt{BUS} & 32.98 & \textbf{76.47} & \textbf{41.86} & 75.57 & 40.38 & 71.00 & 39.32 & 71.81 \\
\hline
\strut\\[-\normalbaselineskip]
\only<5|handout:0>{\\[-\normalbaselineskip]\rowcolor{IGNRed}}\texttt{BIB} & 12.32 & 67.57 & 12.81 & \textbf{68.42} & 16.26 & 67.35 & \textbf{16.75} & 68.0 \\
\hline
\strut\\[-\normalbaselineskip]
\only<4|handout:0>{\\[-\normalbaselineskip]\rowcolor{orange}}\only<5|handout:0>{\\[-\normalbaselineskip]\rowcolor{IGNRed}}\texttt{BIT} & \textbf{25.25} & 92.59 & 20.20 & 90.91 & 20.20 & \textbf{95.24} & 11.11 & 91.67 \\
\hline
\hline
\texttt{FOS} & 98.91 & 99.07 & 98.91 & \textbf{99.30} & \textbf{98.99} & 98.84 & 98.91 & 98.84 \\
\hline
\strut\\[-\normalbaselineskip]
\only<5|handout:0>{\\[-\normalbaselineskip]\rowcolor{IGNRed}}\texttt{FUS} & \textbf{1.90} & 54.55 & 0.63 & \textbf{66.67} & 1.61 & 50 & 1.27 & \textbf{66.67} \\
\hline
\strut\\[-\normalbaselineskip]
\only<5|handout:0>{\\[-\normalbaselineskip]\rowcolor{IGNRed}}\texttt{FIB} & \textbf{9.17} & 87.5 & 0 & --- & 8.30 & 82.61 & 7.42 & \textbf{100} \\
\hline
\strut\\[-\normalbaselineskip]
\only<5|handout:0>{\\[-\normalbaselineskip]\rowcolor{IGNRed}}\texttt{FIT} & \textbf{6.67} & \textbf{100} & 8.73 & 95.24 & 3.33 & \textbf{100} & 3.33 & \textbf{100} \\
\hline
\texttt{FIG} & \textbf{80.54} & 73.14 & 80.45 & \textbf{72.62} & 78.69 & 72.12 & 79.02 & 71.82 \\
\hline
\end{tabular}
\end{center}
\end{table}
\only<1>{
\begin{center}
\footnotesize Results in percentage obtained using \texttt{RF} (\underline{1,000 trees} \& \underline{max depth 4}) and \underline{10-fold cross. val}.
\end{center}
}
\begin{itemize}[label=$\blacktriangleright$, font=\color{IGNGreen}]
\item[\textcolor{IGNDarkGreen}{$\blacktriangleright$}]<2-> \footnotesize In most cases \textbf{Geom.} is sufficient;
\item<3-> \footnotesize Except for \texttt{BUS} $\rightarrow$ other modalities have a big impact ($\approx$ + 10\%).
\item[\textcolor{orange}{$\blacktriangleright$}]<4-> \footnotesize Only \textbf{Geom.} is even \underline{the best} for \texttt{BIT}.
\item[\textcolor{IGNRed}{$\blacktriangleright$}]<5-> \footnotesize Highly unbalanced labels are \underline{hard} to detect.
\end{itemize}
\end{frame}
\begin{frame}{Transferability}
\begin{table}
\scriptsize
\begin{center}
\begin{tabular}{| x{1.5cm}| x{.4cm} | x{.4cm} | x{.4cm} | x{.4cm} | x{.4cm} | x{.4cm} | x{.4cm} | x{.4cm} | x{.4cm} |}
\hline
&\texttt{BOS} & \texttt{BUS} &\texttt{BIB} &\texttt{BIT} &\texttt{FOS} & \texttt{FUS} &\texttt{FIB} &\texttt{FIT} &\texttt{FIG} \\
\hline
Elancourt $\rightarrow$ Nantes &\cellcolor{IGNRed} & \cellcolor{IGNRed}&\cellcolor{IGNRed}& \cellcolor{IGNRed}& \cellcolor{IGNGreen}& \cellcolor{IGNGreen}&\cellcolor{IGNGreen} Im. & \cellcolor{IGNRed}&\cellcolor{IGNGreen}\\
\hline
Elancourt $\rightarrow$ Paris-13 & \cellcolor{IGNRed}& \cellcolor{IGNRed}& \cellcolor{IGNGreen}& \cellcolor{IGNRed}& \cellcolor{IGNGreen}& \cellcolor{IGNGreen}Im.& \cellcolor{IGNGreen}Im.& \cellcolor{IGNRed}&\cellcolor{IGNGreen}\\
\hline
Nantes $\rightarrow$ Paris-13 & \cellcolor{IGNRed}& \cellcolor{IGNRed}& \cellcolor{IGNDarkGrey} & \cellcolor{IGNGreen}& \cellcolor{IGNGreen}& \cellcolor{IGNGreen}& \cellcolor{IGNGreen}& \cellcolor{IGNDarkGrey} &\cellcolor{IGNGreen}\\
\hline
Nantes $\rightarrow$ Elancourt &\cellcolor{IGNGreen} & \cellcolor{IGNRed}& \cellcolor{IGNRed}&\cellcolor{IGNGreen}All & \cellcolor{IGNGreen}& \cellcolor{IGNRed}&\cellcolor{IGNRed}Im. &\cellcolor{IGNGreen}Im. &\cellcolor{IGNGreen}\\
\hline
Paris-13 $\rightarrow$ Nantes &\cellcolor{IGNRed} & \cellcolor{IGNRed}& \cellcolor{IGNDarkGrey} & \cellcolor{IGNGreen}&\cellcolor{IGNGreen} & \cellcolor{IGNRed}& \cellcolor{IGNGreen}& \cellcolor{IGNDarkGrey} & \cellcolor{IGNRed} Hei.\\
\hline
Paris-13 $\rightarrow$ Elancourt &\cellcolor{IGNGreen} &\cellcolor{IGNRed} & \cellcolor{IGNGreen}Im. & \cellcolor{IGNGreen}& \cellcolor{IGNGreen}& \cellcolor{IGNRed}& \cellcolor{IGNRed}Im. & \cellcolor{IGNDarkGrey} & \cellcolor{IGNGreen}\\
\hline
\end{tabular}
\caption{\footnotesize \textcolor{IGNRed}{$\blacksquare$}: Loss in F-score, \textcolor{IGNGreen}{$\blacksquare$}: Stability or gain in F-score.}
\end{center}
\end{table}
\uncover<2->{
\begin{itemize}[label=$\blacktriangleright$, font=\color{IGNGreen}]
\item<2-> \footnotesize Only 8/24 of \texttt{Building errors} transferability cases are \textcolor{IGNGreen}{positive}.
\item<3-> \footnotesize \texttt{Facets errors} are more transferable: 19/30 are \textcolor{IGNGreen}{positive}.
\item<4-> \footnotesize Additional modalities play a \underline{critical role} in error prediction.
\end{itemize}
}
\end{frame}
\section{Conclusion}
\begin{frame}{\textcolor{yellow}{Conclusion} \& \textcolor{purple}{Perspectives}}
\begin{itemize}[label=$\blacktriangleright$, font=\color{yellow}, itemsep=2em]
\item<1-> \textbf{Flexible, robust and hierarchical} taxonomy;
\item<2-> \textbf{Fast, lightweight and modular} pipeline for model evaluation;
\item<3-> Pipeline tested with \textbf{baseline} handcrafted features:
\begin{itemize}[label=$\blacktriangleright$, font=\color{IGNGreen}]
\item<4-> Geometric features are often sufficient for error prediction;
\item<5-> Adding modalities is most helpfull in transferability.
\end{itemize}
\end{itemize}
\uncover<6->{
~\\
Future work:
\begin{itemize}[label=$\blacktriangleright$, font=\color{purple}, itemsep=2em]
\item<6-> Extend to richer features $\longrightarrow$ Graph kernels, Scattering.
\item<7-> Dataset augmentation $\longrightarrow$ \textbf{Simulate errors}.
\end{itemize}
}
\end{frame}
\end{document}