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
|
<!doctype html>
<html>
<head>
<title>MegaPixels: Transnational Flows of Face Recognition Image Training Data</title>
<meta charset="utf-8" />
<meta name="author" content="Adam Harvey" />
<meta name="description" content="Analyzing Transnational Flows of Face Recognition Image Training Data" />
<meta property="og:title" content="MegaPixels: Transnational Flows of Face Recognition Image Training Data"/>
<meta property="og:type" content="website"/>
<meta property="og:summary" content="MegaPixels is an art and research project about face recognition datasets created \"in the wild\"/>
<meta property="og:image" content="https://nyc3.digitaloceanspaces.com/megapixels/v1/research/munich_security_conference/assets/background.jpg" />
<meta property="og:url" content="https://megapixels.cc/research/munich_security_conference/"/>
<meta property="og:site_name" content="MegaPixels" />
<meta name="referrer" content="no-referrer" />
<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=no"/>
<meta name="apple-mobile-web-app-status-bar-style" content="black">
<meta name="apple-mobile-web-app-capable" content="yes">
<link rel="apple-touch-icon" sizes="57x57" href="/assets/img/favicon/apple-icon-57x57.png">
<link rel="apple-touch-icon" sizes="60x60" href="/assets/img/favicon/apple-icon-60x60.png">
<link rel="apple-touch-icon" sizes="72x72" href="/assets/img/favicon/apple-icon-72x72.png">
<link rel="apple-touch-icon" sizes="76x76" href="/assets/img/favicon/apple-icon-76x76.png">
<link rel="apple-touch-icon" sizes="114x114" href="/assets/img/favicon/apple-icon-114x114.png">
<link rel="apple-touch-icon" sizes="120x120" href="/assets/img/favicon/apple-icon-120x120.png">
<link rel="apple-touch-icon" sizes="144x144" href="/assets/img/favicon/apple-icon-144x144.png">
<link rel="apple-touch-icon" sizes="152x152" href="/assets/img/favicon/apple-icon-152x152.png">
<link rel="apple-touch-icon" sizes="180x180" href="/assets/img/favicon/apple-icon-180x180.png">
<link rel="icon" type="image/png" sizes="192x192" href="/assets/img/favicon/android-icon-192x192.png">
<link rel="icon" type="image/png" sizes="32x32" href="/assets/img/favicon/favicon-32x32.png">
<link rel="icon" type="image/png" sizes="96x96" href="/assets/img/favicon/favicon-96x96.png">
<link rel="icon" type="image/png" sizes="16x16" href="/assets/img/favicon/favicon-16x16.png">
<link rel="manifest" href="/assets/img/favicon/manifest.json">
<meta name="msapplication-TileColor" content="#ffffff">
<meta name="msapplication-TileImage" content="/ms-icon-144x144.png">
<meta name="theme-color" content="#ffffff">
<link rel='stylesheet' href='/assets/css/fonts.css' />
<link rel='stylesheet' href='/assets/css/css.css' />
<link rel='stylesheet' href='/assets/css/leaflet.css' />
<link rel='stylesheet' href='/assets/css/applets.css' />
<link rel='stylesheet' href='/assets/css/mobile.css' />
</head>
<body>
<header>
<a class='slogan' href="/">
<div class='logo'></div>
<div class='site_name'>MegaPixels</div>
</a>
<div class='links'>
<a href="/datasets/">Datasets</a>
<a href="/about/">About</a>
<a href="/research">Research</a>
</div>
</header>
<div class="content content-dataset">
<section class='intro_section' style='background-image: url(https://nyc3.digitaloceanspaces.com/megapixels/v1/site/research/munich_security_conference/assets/background.jpg)'><div class='inner'><div class='hero_desc'><span class='bgpad'>Analyzing Transnational Flows of Face Recognition Image Training Data</span></div><div class='hero_subdesc'><span class='bgpad'>Where does face data originate and who's using it?
</span></div></div></section><section><h2>Face Datasets and Information Supply Chains</h2>
</section><section><div class='right-sidebar'><div class='meta'><div class='gray'>Images Analyzed</div><div>24,302,637</div></div><div class='meta'><div class='gray'>Datasets Analyzed</div><div>30</div></div><div class='meta'><div class='gray'>Years</div><div>2006 - 2018</div></div><div class='meta'><div class='gray'>Status</div><div>Ongoing Investigation</div></div><div class='meta'><div class='gray'>Last Updated</div><div>June 28, 2019</div></div></div><p>National AI strategies often rely on transnational data sources to capitalize on recent advancements in deep learning and neural networks. Researchers benefiting from these transnational data flows can yield quick and significant gains across diverse sectors from health care to biometrics. But new challenges emerge when national AI strategies collide with national interests.</p>
<p>Our <a href="https://www.ft.com/content/cf19b956-60a2-11e9-b285-3acd5d43599e">earlier research</a> on the <a href="/datasets/msceleb">MS Celeb</a> and <a href="/datasets/duke_mtmc">Duke</a> datasets published with the Financial Times revealed that several computer vision image datasets created by US companies and universities were unexpectedly also used for research by the National University of Defense Technology in China, along with top Chinese surveillance firms including SenseTime, SenseNets, CloudWalk, Hikvision, and Megvii/Face++ which have all been linked to the oppressive surveillance of Uighur Muslims in Xinjiang.</p>
<p>In this new research for the <a href="https://tsr.securityconference.de">Munich Security Conference's Transnational Security Report</a> we provide summary statistics about the origins and endpoints of facial recognition information supply chains. To make it more personal, we gathered additional data on the number of public photos from embassies that are currently being used in facial recognition datasets.</p>
<h3>24 Million Non-Cooperative Faces</h3>
<p>In total, we analyzed 30 publicly available face recognition and face analysis datasets that collectively include over 24 million non-cooperative images. Of these 24 million images, over 15 million face images are from Internet search engines, over 5.8 million from Flickr.com, over 2.5 million from the Internet Movie Database (IMDb.com), and nearly 500,000 from CCTV footage. All 24 million images were collected without any explicit consent, a type of face image that researchers call "in the wild".</p>
<p>Next we manually verified 1,134 publicly available research papers that cite these datasets to determine who was using the data and where it was being used. Even though the vast majority of the images originated in the United States, the publicly available research citations show that only about 25% citations are from the country of the origin while the majority of citations are from China.</p>
</section><section><div class='columns columns-2'><section class='applet_container'><div class='applet' data-payload='{"command": "single_pie_chart /site/research/munich_security_conference/assets/megapixels_origins_top.csv", "fields": ["Caption: Sources of Publicly Available Non-Cooperative Face Image Training Data 2006 - 2018", "Top: 10", "OtherLabel: Other"]}'></div></section><section class='applet_container'><div class='applet' data-payload='{"command": "single_pie_chart /site/research/munich_security_conference/assets/summary_countries.csv", "fields": ["Caption: Locations Where Face Data Is Used Based on Public Research Citations", "Top: 14", "OtherLabel: Other"]}'></div></section></div></section><section><h3>6,000 Embassy Photos Being Used To Train Facial Recognition</h3>
<p>Of the 5.8 million Flickr images we found over 6,000 public photos from Embassy Flickr accounts were used to train facial recognition technologies. These images were used in the MegaFace and IBM Diversity in Faces datasets. Over 2,000 more images were included in the Who Goes There dataset, used for facial ethnicity analysis research. A few of the embassy images found in facial recognition datasets are shown below.</p>
</section><section><div class='columns columns-2'><section class='applet_container'><div class='applet' data-payload='{"command": "single_pie_chart /site/research/munich_security_conference/assets/country_counts.csv", "fields": ["Caption: Photos from these embassies are being used to train face recognition software", "Top: 4", "OtherLabel: Other", "Colors: categoryRainbow"]}'></div></section><section class='applet_container'><div class='applet' data-payload='{"command": "single_pie_chart /site/research/munich_security_conference/assets/embassy_counts_summary_dataset.csv", "fields": ["Caption: Embassy images were found in these datasets", "Top: 4", "OtherLabel: Other", "Colors: categoryRainbow"]}'></div></section></div></section><section class='images'><div class='image'><img src='https://nyc3.digitaloceanspaces.com/megapixels/v1/site/research/munich_security_conference/assets/4606260362.jpg' alt=' An image in the MegaFace dataset obtained from United Kingdoms Embassy in Italy'><div class='caption'> An image in the MegaFace dataset obtained from United Kingdom's Embassy in Italy</div></div>
<div class='image'><img src='https://nyc3.digitaloceanspaces.com/megapixels/v1/site/research/munich_security_conference/assets/4749096858.jpg' alt=' An image in the MegaFace dataset obtained from the Flickr account of the United States Embassy in Kabul, Afghanistan'><div class='caption'> An image in the MegaFace dataset obtained from the Flickr account of the United States Embassy in Kabul, Afghanistan</div></div></section><section class='images'><div class='image'><img src='https://nyc3.digitaloceanspaces.com/megapixels/v1/site/research/munich_security_conference/assets/4730007024.jpg' alt=' An image in the MegaFace dataset obtained from U.S. Embassy Canberra'><div class='caption'> An image in the MegaFace dataset obtained from U.S. Embassy Canberra</div></div></section><section><p>This brief research aims to shed light on the emerging politics of data. A photo is no longer just a photo when it can also be surveillance training data, and datasets can no longer be separated from the development of software when software is now built with data. "Our relationship to computers has changed", says Geoffrey Hinton, one of the founders of modern day neural networks and deep learning. "Instead of programming them, we now show them and they figure it out."<a class="footnote_shim" name="[^hinton]_1"> </a><a href="#[^hinton]" class="footnote" title="Footnote 1">1</a>.</p>
<p>As data becomes more political, national AI strategies might also want to include transnational dataset strategies.</p>
<p><em>This research post is ongoing and will updated during July and August, 2019.</em></p>
<h3>Further Reading</h3>
<ul>
<li><a href="/datasets/msceleb">MS Celeb Dataset Analysis</a></li>
<li><a href="/datasets/brainwash">Brainwash Dataset Analysis</a></li>
<li><a href="/datasets/duke_mtmc">Duke MTMC Dataset Analysis</a></li>
<li><a href="/datasets/uccs">Unconstrained College Students Dataset Analysis</a></li>
<li><a href="https://www.dukechronicle.com/article/2019/06/duke-university-facial-recognition-data-set-study-surveillance-video-students-china-uyghur">Duke MTMC dataset author apologies to students</a></li>
<li><a href="https://www.bbc.com/news/technology-48555149">BBC coverage of MS Celeb dataset takedown</a></li>
<li><a href="https://www.spiegel.de/netzwelt/web/microsoft-gesichtserkennung-datenbank-mit-zehn-millionen-fotos-geloescht-a-1271221.html">Spiegel coverage of MS Celeb dataset takedown</a></li>
</ul>
</section><section>
<div class="hr-wave-holder">
<div class="hr-wave-line hr-wave-line1"></div>
<div class="hr-wave-line hr-wave-line2"></div>
</div>
<h2>Supplementary Information</h2>
</section><section class='applet_container'><div class='applet' data-payload='{"command": "load_file /site/research/munich_security_conference/assets/embassy_counts_public.csv", "fields": ["Headings: Images, Dataset, Embassy, Flickr ID, URL, Guest, Host"]}'></div></section><section>
<h4>Cite Our Work</h4>
<p>
If you find this analysis helpful, please cite our work:
<pre id="cite-bibtex">
@online{megapixels,
author = {Harvey, Adam. LaPlace, Jules.},
title = {MegaPixels: Origins, Ethics, and Privacy Implications of Publicly Available Face Recognition Image Datasets},
year = 2019,
url = {https://megapixels.cc/},
urldate = {2019-04-18}
}</pre>
</p>
</section><section><h3>References</h3><section><ul class="footnotes"><li>1 <a name="[^hinton]" class="footnote_shim"></a><span class="backlinks"><a href="#[^hinton]_1">a</a></span>"Heroes of Deep Learning: Andrew Ng interviews Geoffrey Hinton". Published on Aug 8, 2017. <a href="https://www.youtube.com/watch?v=-eyhCTvrEtE">https://www.youtube.com/watch?v=-eyhCTvrEtE</a>
</li></ul></section></section>
</div>
<footer>
<ul class="footer-left">
<li><a href="/">MegaPixels.cc</a></li>
<li><a href="/datasets/">Datasets</a></li>
<li><a href="/about/">About</a></li>
<li><a href="/about/news/">News</a></li>
<li><a href="/about/legal/">Legal & Privacy</a></li>
</ul>
<ul class="footer-right">
<li>MegaPixels ©2017-19 <a href="https://ahprojects.com">Adam R. Harvey</a></li>
<li>Made with support from <a href="https://mozilla.org">Mozilla</a></li>
</ul>
</footer>
</body>
<script src="/assets/js/dist/index.js"></script>
</html>
|
