-
Notifications
You must be signed in to change notification settings - Fork 2
/
SKDAVS.html
158 lines (156 loc) · 6.67 KB
/
SKDAVS.html
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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<title>SKDAVS</title>
<link rel="stylesheet" type="text/css" href="assets/scripts/bulma.min.css">
<link rel="stylesheet" type="text/css" href="assets/scripts/theme.css">
<link rel="stylesheet" type="text/css" href="https://cdn.bootcdn.net/ajax/libs/font-awesome/4.7.0/css/font-awesome.min.css">
</head>
<body>
<section class="hero is-light" style="">
<div class="hero-body" style="padding-top: 50px;">
<div class="container" style="text-align: center;margin-bottom:5px;">
<h1 class="title">
Spatiotemporal Knowledge Distillation for Efficient
</h1>
<h1 class="title">
Estimation of Aerial Video Saliency
</h1>
<div class="author">Jia Li<sup>1</sup></div>
<div class="author">Kui Fu<sup>1,3</sup></div>
<div class="author">Shengwei Zhao<sup>2,4</sup></div>
<div class="author">Shiming Ge<sup>2</sup></div>
<div class="group">
<a href="http://cvteam.net/">CVTEAM</a>
</div>
<div class="aff">
<p><sup>1</sup>State Key Laboratory of Virtual Reality Technology and Systems, SCSE, Beihang University, Beijing, China</p>
<p><sup>2</sup>Institute of Information Engineering, Chinese Academy of Sciences</p>
<p><sup>3</sup>Peng Cheng Laboratory</p>
<p><sup>4</sup> School of Cyber Security, University of Chinese Academy of Sciences</p>
</div>
<div class="con">
<p style="font-size: 24px; margin-top:5px; margin-bottom: 15px;">
TIP 2020
</p>
</div>
<div class="columns">
<div class="column"></div>
<div class="column"></div>
<div class="column">
<a href="http://cvteam.net/papers/2019-TIP-Spatiotemporal%20Knowledge%20Distillation%20for%20Efficient%20Estimation%20of%20Aerial%20Video%20Saliency.pdf" target="_blank">
<p class="link">Paper</p>
</a>
</div>
<div class="column">
<a href="https://github.com/iCVTEAM/SKDAVS/" target="_blank">
<p class="link">Code</p>
</a>
</div>
<div class="column"></div>
<div class="column"></div>
</div>
</div>
</div>
</section>
<div style="text-align: center;">
<div class="container" style="max-width:850px">
<div style="text-align: center;">
<img src="assets/SKDAVS/head.png" class="centerImage">
</div>
</div>
<div class="head_cap">
<p style="color:gray;">
System framework of SKD.
</p>
</div>
</div>
<section class="hero">
<div class="hero-body">
<div class="container" style="max-width: 800px" >
<h1 style="">Abstract</h1>
<p style="text-align: justify; font-size: 17px;">
The performance of video saliency estimation techniques
has achieved significant advances along with the rapid
development of Convolutional Neural Networks (CNNs).
However, devices like cameras and drones may have limited
computational capability and storage space so that
the direct deployment of complex deep saliency models
becomes infeasible. To address this problem, this paper
proposes a dynamic saliency estimation approach for aerial
videos via spatiotemporal knowledge distillation. In this
approach, five components are involved, including two
teachers, two students and the desired spatiotemporal model.
The knowledge of spatial and temporal saliency is first
separately transferred from the two complex and redundant
teachers to their simple and compact students, while the
input scenes are also degraded from high-resolution to
low-resolution to remove the probable data redundancy so
as to greatly speed up the feature extraction process.
After that, the desired spatiotemporal model is further
trained by distilling and encoding the spatial and
temporal saliency knowledge of two students into a unified
network. In this manner, the inter-model redundancy can be
removed for the effective estimation of dynamic saliency on
aerial videos. Experimental results show that the proposed
approach is comparable to 11 state-of-the-art models in
estimating visual saliency on aerial videos, while its
speed reaches up to 28,738 FPS and 1,490.5
FPS on the GPU and CPU platforms,respectively.
</p>
</div>
</div>
</section>
<section class="hero is-light" style="background-color:#FFFFFF;">
<div class="hero-body">
<div class="container" style="max-width:800px;margin-bottom:20px;">
<h1>
Qualitative comparisons
</h1>
</div>
<div style="text-align: center;">
<div class="container" style="max-width:850px">
<div style="text-align: center;">
<img src="assets/SKDAVS/comp.png" class="centerImage">
</div>
</div>
<div class="head_cap">
<p style="color:gray;">
Representative frames of the models on AVS1K. (a) Video frame, (b) Ground truth, (c) HFT,
(d) SP, (e) PNSP, (f) SSD, (g) LDS, (h) eDN, (i) iSEEL,(j) DVA, (k) SalNet, (l) STS, (m) SKD.
</p>
</div>
</div>
</div>
</section>
<section class="hero" style="padding-top:0px;">
<div class="hero-body">
<div class="container" style="max-width:800px;">
<div class="card">
<header class="card-header">
<p class="card-header-title">
BibTex Citation
</p>
<a class="card-header-icon button-clipboard" style="border:0px; background: inherit;" data-clipboard-target="#bibtex-info" >
<i class="fa fa-copy" height="20px"></i>
</a>
</header>
<div class="card-content">
<pre style="background-color:inherit;padding: 0px;" id="bibtex-info">@article{li2019spatiotemporal,
title={Spatiotemporal knowledge distillation for efficient estimation of aerial video saliency},
author={Li, Jia and Fu, Kui and Zhao, Shengwei and Ge, Shiming},
journal={IEEE Transactions on Image Processing},
volume={29},
pages={1902--1914},
year={2019},
publisher={IEEE}
}</pre>
</div>
</section>
<script type="text/javascript" src="assets/scripts/clipboard.min.js"></script>
<script>
new ClipboardJS('.button-clipboard');
</script>
</body>
</html>