Omitted Labels

red high­light­ed objects/persons were miss­ing in a dataset cru­cial for autonomous driving

Brad Dwyer found a lot of miss­ing or omit­ted labels in a set that is used for train­ing and test­ing autonomous dryv­ing sys­tems. »We did a hand-check of the 15,000 images in the wide­ly used Udac­i­ty Dataset 2 and found prob­lems with 4,986 (33%) of them.« Since this is a Open Source Dataset used pri­mar­i­ly for edu­ca­tion­al pur­pos­es, but as the author found out obvi­ous­ly also for test cars on pub­lic streets, he pub­lished a cor­rect­ed set at

Besides the hon­or­able work of Dwyer, these omis­sions lead to the larg­er ques­tion of the reli­abli­ty of many data sets which are being used for training.

Face-recognition respirator masks

Danielle Baskin cre­at­ed a web­site for com­pu­ta­tion­al map­ping to con­vert facial fea­tures into an image print­ed onto the sur­face of N95 sur­gi­cal masks with­out dis­tor­tion. It is a reac­tion to the Coro­na virus epi­dem­ic and allows to unlock (aka trick) face id tech­niques of smart phones.

Tricking OpenCV

KodyK­inzie: »Con­firm­ing crit­i­cal facial recog­ni­tion research by @tahkion regard­ing #jug­ga­lo make­up defeat­ing detec­tion and recog­ni­tion using the #esp32 and #face_recognition/#openCV Python libraries. Results seem conclusive.«

Make sure to read the thread:

Obfuscation of data through using group accounts

Teenagers have come up with elab­o­rat­ed schemes to share insta­gram accounts and pro­duce obfus­cat­ing data, in order to look at what­ev­er they want to look at with­out being tracked individually. 

»Each time she refreshed the Explore tab, it was a com­plete­ly dif­fer­ent top­ic, none of which she was inter­est­ed in. That’s because Mosley was­n’t the only per­son using this account — it belonged to a group of her friends, at least five of whom could be on at any giv­en time. Maybe they could­n’t hide their data foot­prints, but they could at least leave hun­dreds behind to con­fuse track­ers.« Alfred Ng on

Read Full arti­cle here:

Paint Your Face Away workshop

Paint Your Face Away is a drop-in dig­i­tal face paint­ing work­shop by Shin­ji Toya. The devel­op­ment of the dig­i­tal face paint­ing tool for this ses­sion has been inspired by Frank Bowling’s paint­ings. Par­tic­i­pants use the painter to cre­ate their pro­file pic­tures while run­ning a real-time face detec­tion on the image of a face being paint­ed so that at one point the pro­file pic­ture stops being detect­ed by the com­put­er vision through the paint­ing process. In this way, the dig­i­tal paint acts as a type of dis­rup­tive noise for the machine.

Read fur­ther at

Google Maps Traffic Jam

Artist Simon Weck­ert gen­er­ates poi­son data by trans­port­ing 99 sec­ond hand smart­phones in a hand­cart and gen­er­ates a vir­tu­al traf­fic jam in Google Maps. Through this activ­i­ty he shows that it is pos­si­ble to turn a green street red. This in turn has an impact in the phys­i­cal world by nav­i­gat­ing cars on anoth­er route to avoid being stuck in traf­fic. Simon U Rock!


Umbrel­las are prac­ti­cal when it comes to avoid auto­mat­ed face recog­ni­tion from CCTV et cetera, since they are every­day items and can’t be effec­tive­ly banned by authorities.

Pro­test­ers spray paint a sur­veil­lance cam­era in Hongkong in July 2019 using umbrellas

Generated Faces

Icons8 prod­uct design­er Kon­stan­tin Zhabin­skiy worked on a project of gen­er­at­ing 100k faces (using GANs) from a total of 29.000 pho­tographs that they pho­tographed in-house. This has the advan­tage of con­sis­tent light­en­ing and being able to pho­to­graph dif­fer­ent angles of the same face. 

For the time being they have open sourced a large data-set hop­ing for trak­tion. It can be used for avatar images and such – so if you ever want­ed to pre­tend you look like a mod­el, no wrin­kles, per­fect light­en­ing, sym­met­ric eyes and such, only a few GAN-glitch­es, go ahead and use them for your account.


“incog­ni­to” is an anti-recog­ni­tion jew­el­ry mask by design stu­dio NOMA, War­saw it revers­es the nose-eye rela­tion and that’s what we like about it. Once could definit­ly go out on street with this.


This cre­ation by Lon­don based design­er Richard Quinn gets you ful­ly cov­ered. It got some trak­tion, since Car­di B appeared at Paris Fash­ion Week in one of his body and face cov­ers. Maybe a motor­cy­cle hel­met would be still obfus­cat­ing enough, but would you want to wear it on fash­ion week?

Anti Recognition Mask

anti-recog­ni­tion mask by design­er col­lec­tive NOMA, War­saw,

Surveillance Detection Scout

»Sur­veil­lance Detec­tion Scout is a hard­ware and soft­ware stack that makes use of your Tes­la’s cam­eras to tell you if you’re being fol­lowed in real-time. The name, as you like­ly gath­ered, pays homage to the ever-effec­tive Sur­veil­lance Detec­tion Route. When parked, Scout makes an excel­lent sta­t­ic sur­veil­lance prac­ti­tion­er as well, allow­ing you to run queries and estab­lish pat­terns-of-life on detect­ed persons.« 

Researcher Tru­man Kain there­fore uses Facenet Image Recog­ni­tion train­ing data and Plugs into Tes­las pub­lic API. For License Plate Recog­ni­tion he uses ALPR. To save the imagery cre­at­ed by the three Tes­la front cam­eras, he uses a soft­ware called Tes­la USB.

Wired-Author Andy Green­berg notes: 

»Kain, a con­sul­tant for the secu­ri­ty firm Tevo­ra, also isn’t obliv­i­ous to his cre­ation’s creep fac­tor. He says the Sur­veil­lance Detec­tion Scout demon­strates the kind of sur­veil­lance the data that self-dri­ving cars already col­lect could enable.«

To this presents a use-case where you want to have adver­sar­i­al patch­es on license plates (if that is not for­bid­den by law, because it presents some kind of obfus­ca­tion) and of course wear an adver­sar­i­al t‑shirt of some kind… This case also reminds me of the spec­u­la­tion, that UBER at some point might make their cars more prof­itable, by using them as data col­lec­tion drones.