diff options
Diffstat (limited to 'site/content/pages/research')
3 files changed, 64 insertions, 81 deletions
diff --git a/site/content/pages/research/munich_security_conference/assets/embassy_counts_summary_dataset.csv b/site/content/pages/research/munich_security_conference/assets/embassy_counts_summary_dataset.csv index 89f3c226..3a439821 100755 --- a/site/content/pages/research/munich_security_conference/assets/embassy_counts_summary_dataset.csv +++ b/site/content/pages/research/munich_security_conference/assets/embassy_counts_summary_dataset.csv @@ -1,5 +1,5 @@ dataset,images -ibm_dif,389 -megaface,5679 -vgg_face,1 -who_goes_there,2372 +IBM Diversity in Faces,389 +MegaFace,5679 +VGG Face,1 +Who Goes There,2372
\ No newline at end of file diff --git a/site/content/pages/research/munich_security_conference/assets/megapixels_origins_top.csv b/site/content/pages/research/munich_security_conference/assets/megapixels_origins_top.csv index 081b4636..ae6e8f11 100755 --- a/site/content/pages/research/munich_security_conference/assets/megapixels_origins_top.csv +++ b/site/content/pages/research/munich_security_conference/assets/megapixels_origins_top.csv @@ -1,9 +1,8 @@ source,images -Search Engines,30127200 -Flickr.com,11783888 -IMDb.com,5251410 -CCTV,959312 -Wikimedia.org,183500 -Mugshots,113268 -YouTube.com,31888 -Other Sources Combined,37044 +Internet Search Engines,15063600 +Flickr.com,5891944 +Internet Movie Database (IMDB.com),2625705 +CCTV,479656 +Wikimedia.org,91750 +Mugshots,56634 +YouTube.com,15944
\ No newline at end of file diff --git a/site/content/pages/research/munich_security_conference/index.md b/site/content/pages/research/munich_security_conference/index.md index 2a97a7c0..0f8a5bda 100644 --- a/site/content/pages/research/munich_security_conference/index.md +++ b/site/content/pages/research/munich_security_conference/index.md @@ -3,7 +3,7 @@ status: published title: MSC slug: munich-security-conference -desc: Analyzing the Transnational Flow of Facial Recognition Training Data +desc: Analyzing Transnational Flows of Face Recognition Image Training Data subdesc: Where does face data originate and who's using it? cssclass: dataset image: assets/background.jpg @@ -13,7 +13,7 @@ authors: Adam Harvey ------------ -## Analysis for the Munich Security Conference Transnational Security Report +## Face Datasets and Information Supply Chains ### sidebar @@ -21,21 +21,30 @@ authors: Adam Harvey + Datasets Analyzed: 30 + Years: 2006 - 2018 + Status: Ongoing Investigation -+ Last Updated: June 27, 2019 ++ Last Updated: June 28, 2019 ### end sidebar +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. -Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum." +Our earlier research on the [MS Celeb](/datsets) and [Duke](/datsets/duke_mtmc) 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. + +In this new research for the Munich Security Conference's Transnational Security Report 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. + + +### 24 Million Non-Cooperative Faces + +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 researchers call "in the wild". + +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 all 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. -Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum." === columns 2 ``` single_pie_chart /site/research/munich_security_conference/assets/megapixels_origins_top.csv -Caption: Sources of Publicly Available Face Training Data 2006 - 2018 +Caption: Sources of Publicly Available Non-Cooperative Face Image Training Data 2006 - 2018 Top: 10 OtherLabel: Other ``` @@ -44,85 +53,32 @@ OtherLabel: Other ``` single_pie_chart /site/research/munich_security_conference/assets/summary_countries.csv -Caption: Locations Where Face Data Is Used +Caption: Locations Where Face Data Is Used Based on Public Research Citations Top: 14 OtherLabel: Other ``` === end columns +### 6,000 Embassy Photos Being Used To Train Facial Recognition -=== columns 2 - -#### Sources of Face Data - -Add text - -| Source | Images | -| --- | --- | -|Search Engines | 30,127,200 | -|Flickr.com | 11,783,888 | -|IMDb.com | 5,251,410 | -|CCTV | 959,312 | -|Wikimedia.org | 183,500 | -|Mugshots | 113,268 | -|Other Sources Combined | 37,044 | -|YouTube.com | 31,888 | - -=== - -#### Where Face Data Is Used - -Add text - -|country | citations| -| --- | --- | -|China | 327| -|United States | 302| -|United Kingdom | 187| -|Australia | 38| -|Germany | 35| -|Singapore | 27| -|Canada | 25| -|Netherlands | 25| -|Italy | 22| -|France | 17| -|India | 14| -|South Korea | 12| -|Spain | 10| -|Switzerland | 9| - -=== end columns - - - -## Over 6,000 Embassy Images on Flickr Found in Face Recognition Datasets - -Including over 2,000 more for racial analysis - - - - - - - - +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, IBM Diversity in Faces datasets. Over 2,000 more images were used in the Who Goes There datasets used for facial ethnicity analysis research. A few of the embassy images found in facial recognition datasets are shown below. === columns 2 ``` -single_pie_chart /site/research/munich_security_conference/assets/megapixels_origins_top.csv -Caption: Sources of Face Training Data -Top: 5 -OtherLabel: Other Countries +single_pie_chart /site/research/munich_security_conference/assets/country_counts.csv +Caption: Photos from these embassies are being used to train face recognition software +Top: 4 +OtherLabel: Other Colors: categoryRainbow ``` -=========== +===== ``` single_pie_chart /site/research/munich_security_conference/assets/embassy_counts_summary_dataset.csv -Caption: Dataset sources +Caption: Embassy images were found in these datasets Top: 4 OtherLabel: Other Colors: categoryRainbow @@ -130,6 +86,29 @@ Colors: categoryRainbow === end columns + + + + + + +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."[^hinton]. + +National AI strategies might also want to include transnational dataset strategies. + +*This research post is going and will updated during July and August, 2019.* + +### Further Reading + +- [MS Celeb Dataset Analysis](/datasets/msceleb) +- [Brainwash Dataset Analysis](/datasets/brainwash) +- [Duke MTMC Dataset Analysis](/datasets/duke_mtmc) +- [Unconstrained College Students Dataset Analysis](/datasets/uccs) +- [Duke MTMC dataset author apologies to students](https://www.dukechronicle.com/article/2019/06/duke-university-facial-recognition-data-set-study-surveillance-video-students-china-uyghur) +- [BBC coverage of MS Celeb dataset takedown](https://www.bbc.com/news/technology-48555149) +- [Spiegel coverage of MS Celeb dataset takdown](https://www.spiegel.de/netzwelt/web/microsoft-gesichtserkennung-datenbank-mit-zehn-millionen-fotos-geloescht-a-1271221.html) + + {% include 'supplementary_header.html' %} ``` @@ -137,5 +116,10 @@ load_file /site/research/munich_security_conference/assets/embassy_counts_public Headings: Images, Dataset, Embassy, Flickr ID, URL, Guest, Host ``` +{% include 'cite_our_work.html' %} + +### Footnotes + +[^hinton]: "Heroes of Deep Learning: Andrew Ng interviews Geoffrey Hinton". Published on Aug 8, 2017. <https://www.youtube.com/watch?v=-eyhCTvrEtE> + -{% include 'cite_our_work.html' %}
\ No newline at end of file |