-
Notifications
You must be signed in to change notification settings - Fork 15
/
Copy pathfact_sheets.tex
195 lines (129 loc) · 6.91 KB
/
fact_sheets.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
\documentclass{article}
\usepackage[utf8]{inputenc}
\usepackage{url}
\usepackage{hyperref}
\usepackage{graphicx}
\title{Emotion challenge fact sheet}
\author{iCV}
\date{February 2017}
\begin{document}
\maketitle
\section{Team details}
\begin{itemize}
\item Team name
CBSR-CASIA
\item Team leader name
Jianzhu, Guo
\item Team leader address, phone number and email
Address: No.95 Zhongguancun East Road, Haidian District, Beijing 100190, China \\
Phone: 86-15600118066 \\
Email: [email protected]
\item Rest of the team members
Shuai, Zhou (equal contribution)\\
Jilin, Wu (equal contribution)\\
Jun, Wan (corresponding author) \\
Xiangyu, Zhu \\
Zhen, Lei
\item Team website URL (if any)
\url{http://www.cbsr.ia.ac.cn/users/jwan/research.html}
\item Affiliation
National Laboratory of Pattern Recognition (NLPR), Institute of Automation, Chinese Academy of Sciences (CASIA).
\end{itemize}
\section{Contribution details}
\begin{itemize}
\item Title of the contribution
\textbf{Multi-modality Network with Visual and Geometry based information for Micro Emotion Recognition}
\item Final score
It is not available now.
\item General method description
We fisrt extract the each ID's lamdmark feature, and subsract it's mean as the geometrical information, then we get each ID's landmark offset of $R^{136\times1}$. And we use an modified AlexNet to extract the $R^{256\times1}$ feature of the $224\times224$ image on the second-last full connected layer. Then we concact the two features, and connect to a full connected layer. We use Hinge Loss to train the net, the ground truth is a single label indicating the complementary and dominant emotion.
\item References
[1] Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton. "ImageNet classification with deep convolutional neural networks." neural information processing systems (2012): 1097-1105.
[2] Kazemi, Vahid, and Josephine Sullivan. "One millisecond face alignment with an ensemble of regression trees." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2014.
\item Representative image / diagram of the method
%\textbf{add flowchart here}
\begin{figure}
\centering
\includegraphics[width=12cm]{figures/net.png}
\caption{Architecture of the network architecture}
\label{fig:architecture}
\end{figure}
\item Describe data preprocessing techniques applied (if any)
%\textbf{add face detection and landmark here}
Face detection and face landmark extraction using dlib and face crop and weakly aligned using the eyes center and the upper flip. We crop the image to $224 \times 224$ size.
\end{itemize}
\section{Face Landmarks Detection}
\subsection{Features / Data representation}
Describe features used or data representation model FOR FACE LANDMARKS DETECTION (if any)
\subsection{Dimensionality reduction}
Dimensionality reduction technique applied FOR FACE LANDMARKS DETECTION (if any)
\subsection{Compositional model}
Compositional model used, i.e. pictorial structure FOR FACE LANDMARKS DETECTION (if any)
\subsection{Learning strategy}
Learning strategy applied FOR FACE LANDMARKS DETECTION (if any)
\subsection{Other techniques}
Other technique/strategy used not included in previous items FOR FACE LANDMARKS DETECTION (if any)
\subsection{Method complexity}
Method complexity FOR FACE LANDMARKS DETECTION
\section{Dominant emotion recognition}
\subsection{Features / Data representation}
Describe features used or data representation model FOR DOMINANT EMOTION RECOGNITION (if any)
\subsection{Dimensionality reduction}
Dimensionality reduction technique applied FOR DOMINANT EMOTION RECOGNITION (if any)
\subsection{Compositional model}
Compositional model used, i.e. pictorial structure FOR DOMINANT EMOTION RECOGNITION (if any)
\subsection{Learning strategy}
Learning strategy applied FOR DOMINANT EMOTION RECOGNITION (if any)
\subsection{Other techniques}
Other technique/strategy used not included in previous items FOR DOMINANT EMOTION RECOGNITION (if any)
\subsection{Method complexity}
Method complexity FOR DOMINANT EMOTION RECOGNITION
\section{Complementary emotion recognition}
\subsection{Features / Data representation}
Describe features used or data representation model FOR COMPLEMENTARY EMOTION RECOGNITION (if any)
\subsection{Dimensionality reduction}
Dimensionality reduction technique applied FOR COMPLEMENTARY EMOTION RECOGNITION (if any)
\subsection{Compositional model}
Compositional model used, i.e. pictorial structure FOR COMPLEMENTARY EMOTION RECOGNITION (if any)
\subsection{Learning strategy}
Learning strategy applied FOR COMPLEMENTARY EMOTION RECOGNITION (if any)
\subsection{Other techniques}
Other technique/strategy used not included in previous items FOR COMPLEMENTARY EMOTION RECOGNITION (if any)
\subsection{Method complexity}
Method complexity FOR COMPLEMENTARY EMOTION RECOGNITION
\section{Joint dominant and complementary emotion recognition}
\subsection{Features / Data representation}
Describe features used or data representation model FOR JOINT DOMINANT AND COMPLEMENTARY EMOTION RECOGNITION (if any)
Landmark offset and AlextNet feature extraction.
\subsection{Dimensionality reduction}
Dimensionality reduction technique applied FOR JOINT DOMINANT AND COMPLEMENTARY EMOTION RECOGNITION (if any)
\subsection{Compositional model}
Compositional model used, i.e. pictorial structure FOR JOINT DOMINANT AND COMPLEMENTARY EMOTION RECOGNITION (if any)
\subsection{Learning strategy}
Learning strategy applied FOR JOINT DOMINANT AND COMPLEMENTARY EMOTION RECOGNITION (if any)
\subsection{Other techniques}
Other technique/strategy used not included in previous items FOR JOINT DOMINANT AND COMPLEMENTARY EMOTION RECOGNITION (if any)
\subsection{Method complexity}
Method complexity FOR JOINT DOMINANT AND COMPLEMENTARY EMOTION RECOGNITION
\section{Global Method Description}
\begin{itemize}
\item Total method complexity: all stages
\item Which pre-trained or external methods have been used (for any stage, if any)
\item Which additional data has been used in addition to the provided training and validation data (at any stage, if any)
\item Qualitative advantages of the proposed solution
If takes advantage of the geometry feature and texture pattern.
\item Results of the comparison to other approaches (if any)
\item Novelty degree of the solution and if is has been previously published
\end{itemize}
\section{Other details}
\begin{itemize}
\item Language and implementation details (including platform, memory, parallelization requirements)
We use Python with OpenCV and Dlib package and Caffe deep learning framework.
\item Detailed list of prerequisites for compilation
\item Human effort required for implementation, training and validation?
\item Training/testing expended time?
Training takes about an hour on Titan-X GPU, validation of 7000 images takes about half minute.
\item General comments and impressions of the challenge?
The task is hard, and it's hard to find effective texture feature to do this challenge well.
\end{itemize}
\end{document}