-
Notifications
You must be signed in to change notification settings - Fork 0
/
index.html
98 lines (88 loc) · 4.48 KB
/
index.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
<!DOCTYPE html>
<html lang="en"><head>
<meta http-equiv="content-type" content="text/html; charset=UTF-8">
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Towards Robust Neural Image Compression</title>
<script type="text/javascript" src="src/latexit.js"></script>
<script type="text/javascript">
LatexIT.add('p',true);
</script>
<!-- CSS includes -->
<link href="src/bootstrap.css" rel="stylesheet">
<link href="src/css.css" rel="stylesheet" type="text/css">
<link href="src/mystyle.css" rel="stylesheet">
</head>
<body>
<div id="header" class="container-fluid">
<div class="row">
<h1>Towards Robust Neural Image Compression: Adversarial Attack and Model Finetuning</h1>
<div class="authors">
<a href="https://tongxyh.github.io", target="_blank">Tong Chen</a> <sup>1</sup>,
<a href="https://vision.nju.edu.cn/fc/d3/c29470a457939/page.htm", target="_blank">Zhan Ma</a> <sup>1</sup>
<div class="affiliations">
<sup>1</sup> Nanjing University
</div>
</div>
<p style="text-align:center;">
<a href="https://www.nju.edu.cn/" target="_blank"><img src="src/NJU_logo.jpg" height="100"></a>
<a href="http://vision.nju.edu.cn/" target="_blank"><img src="src/vision_logo.png" height="100"></a>
</p>
</div>
</div>
<div class="container" id="method">
<p style="text-align:center;">
<img src="src/attack_pipe.svg" height="1100">
</p>
<br>
</div>
<div class="container" id="abstractdiv">
<h2>Abstract</h2>
<p>
Deep neural network-based image compression has been extensively studied. However, the model robustness which is crucial to practical application is largely overlooked. We propose to examine the robustness of prevailing learned image compression models by injecting negligible adversarial perturbation into the original source image. Severe distortion in decoded reconstruction reveals the general vulnerability in existing methods regardless of compression settings (e.g., network architecture, loss function, quality scale). We then explore possible defense strategies against the adversarial attack to improve the model robustness, including geometric self-ensemble based pre-processesing, and adversarial training. Experiments report the effectiveness of various defense strategies. Additional image recompression case study further confirms the substantial improvement of the robustness of compression models in real-life applications. Overall, our methodology is simple, effective, and generalizable, making it attractive for developing robust learned image compression solutions.
</p>
</div>
<div class="container" id="paperdiv">
<h2>Code & Paper</h2>
<div class="row">
<div class="col-sm-3">
</div>
<div class="col-sm-3">
<a href="https://github.com/tongxyh/ImageCompression_Adversarial" target="_blank"><p style="text-align: center;">
<img src="src/github_icon.png">
<br/>
Code (Github)
</p></a></div>
<div class="col-sm-3">
<a href="https://arxiv.org/abs/2112.08691" target="_blank"><p style="text-align: center;">
<img src="src/pdf_icon.png" height="120">
<br/>
Paper (arXiv)
</p></a></div>
<div class="col-sm-3">
</div>
</div>
<a href="" target="_blank">
<div class="thumbs">
</div>
</a>
<div>
<pre class="citation">
@ARTICLE{chen2021robust,
author={Chen, Tong and Ma, Zhan},
journal={IEEE Transactions on Circuits and Systems for Video Technology},
title={Towards Robust Neural Image Compression: Adversarial Attack and Model Finetuning},
year={2023},
volume={},
number={},
pages={1-1},
doi={10.1109/TCSVT.2023.3276442}}
</pre>
</div>
</div>
<div id=footer><br></div>
<!-- Javascript includes -->
<script src="src/jquery-3.5.1.js"></script>
<script src="src/bootstrap.js"></script>
</body></html>