VVVVVStéphanie Vilayphiou

Articles/papers to read

Books

Unpleasant Design (2013), by Selena Savić and Gordan Savičić is a website and 2 books on listing designs which are thought to be unpleasant: for example benches for avoid people staying too long or sleeping on it, anti-climb paintings…

Unpleasant Design

Design and the Elastic Mind: Design and artistic proposals for speculative uses of technology.

Paola Antonelli, *Design and the Elastic Mind*, 2008

Contributors of Wikipedia, «Camouflage»

The majority of camouflage methods aim for crypsis, often through a general resemblance to the background, high contrast disruptive coloration, eliminating shadow, and countershading.

Steven Porter, «Can this clothing defeat face recognition software? Tech-savvy artists experiment», The Christian Science Monitor, 2017

«That’s a problem,» Dr. Sellinger and Dr. Hartzog wrote. «The government should not use people’s faces as a way of tagging them with life-altering labels. The technology isn’t even accurate. Faception’s own estimate for certain traits is a 20 error rate. Even if those optimistic numbers hold, that means that for every 100 people, the best-case scenario is that 20 get wrongly branded as a terrorist.»


[𪀦] «I think camouflage is often misunderstood as a Harry Potter invisibly cloak, when camouflage actually is about optimizing the way you appear and reducing visibility.»

Kate Mothes, «Trick Facial Recognition Software into Thinking You’re a Zebra or Giraffe with These Pyschedelic Garments», Colossal, 2023

«Choosing what to wear is the first act of communication we perform every day. (It’s) a choice that can be the vehicles of our values, » says co-founder and CEO Rachel Didero. Likening the commodification of data to that of oil and its ability to be sold and traded by corporations for enormous sums—often without our knowledge—Didero describes mission of Cap_able as «opening the discussion on the importance of protecting against the misuse of biometric recognition cameras.» When a person dons a sweater, dress, or trousers woven with an adversial image, their is no longer detectable, and it tricks the software into categorizing them as an animal rather than a human.

Adam Harvey, «On computer vision», * UMBAU: Political Bodies*, 2021

Photography has become a nineteenth-century way of looking at a twenty-first century world. In its place emerged a new optical regime: computer vision.


Computer vision, unlike photography, does not mirror reality but instead interprets and misinterprets it, overlaying statistical assumptions of meaning. There is no truth in the output of computer vision algorithms, only statistical probabilities clipped into Boolean states masquerading as truthy outcomes with meaning added in post-production.


Face detection algorithms, for example, do not actually detect faces, though they claim to. Face detection merely detects face-like regions, assigning each with a confident score.


Algorithms are rule sets, and these rules are limited by the perceptual capacities of sensing technologies. This creates «perceptual topologies» that reflect how technology can or cannot see the world. In the first widely used face detection algorithm, developed in 2001 by Viola and Jones, the definition of a face relied on the available imagery of the time for training data. This comprised blurry, low resolution, grayscale CCTV imagery. The Viola-Jones face detection algorithm mirrored back the perceptual biases of low-resolution CCTV systems from the early 2000’s by encoding a blurry, noisy, grayscale definition of the human face. Understanding this perceptual topology can also help discover perceptual vulnerabilities. In my research for CV Dazzle (2010) and HyperFace (2016) I showed that the Viola-Jones Haar Cascade algorithm is vulnerable to presentation attacks using low-cost makeup and hair hacks that obscure the expected low resolution face features, primarily the nose-bridge area. By simply inverting the blocky features of their Haar Cascade algorithm with long hair or bold makeup patterns, faces could effectively disappear from security systems. Another vulnerability of the Haar Cascade algorithm is its reliance on open-source face detection profiles, which can be reverse-engineered to produce the most face-like face. In 2016, I exploited this vulnerability for the HyperFace project to fool (now outdated) face detection systems into thinking dozens of human faces existed in a pink, pixellated graphic on a fashion accessory.


In Paglen’s ImageNet Roulette he excavates the flawed taxonomies that persisted in the WordNet labeling system that was used to label ImageNet, then purposefully trained a flawed image classification algorithm to demonstrate the dangers of racist and misogynistic classification structures.


Becoming training data is political, especially when that data is biometric. But resistance to militarized face recognition and citywide mass surveillance can only happen at a collective level. At a personal level, the dynamics and attacks that were once possible to defeat the Viola-Jones Haar Cascade algorithm are no longer relevant. Neural networks are anti-fragile. Attacking makes them stronger. So-called adversarial attacks are rarely adversarial in nature. Most often they are used to fortify a neural network. In the new optical regime of computer vision every image is a weight, every face is a bias, and every body is a commodity in a global information supply chain.

Adam Harvey, «Origins and endpoints of image training datasets created “in the wild”», 2020

The new logic is not better algorithms; it is better data, and more data.


In 2016, a researcher at Duke University in North Carolina created a dataset of student images called Duke MTMC, or multi-targeted multi-camera. The Duke MTMC dataset contains over 14 hours of synchronized surveillance video from 8 cameras at 1080p and 60FPS, with over 2 million frames of 2,000 students walking to and from classes. The 8 surveillance cameras deployed on campus were specifically setup to capture students «during periods between lectures, when pedestrian traffic is heavy». The dataset became widely popular and over 100 publicly available research papers were discovered that used the dataset. These papers were analyzed according to methodology described earlier to understand endpoints: who is using the dataset, and how it is being used. The results show that the Duke MTMC dataset spread far beyond its origins and intentions in academic research projects at Duke University. Since its publication in 2016, more than twice as many research citations originated in China as in the United States. Among these citations were papers linked to the Chinese military and several companies known to provide Chinese authorities with the oppressive surveillance technology used to monitor millions of Uighur Muslims.


From one perspective, «in the wild» is an ideal characteristic for training data because it can provide a closer match to an unknown deployment environment. Theoretically, this can improve real-world performance by reducing disparity and bias. In reality, data collected from sources «in the wild» inherit new problems including the systemic inequalities within society and are never «natural» or «wild». Representing datasets as unconstrained or «wild» simplifies complexities in the real world where nothing is free from bias. Further, collecting data without consent forces people to unknowingly participate in experiments which may violate human rights.


It is advisable to stop using Creative Commons for all images containing people.

Adam Harvey, «What is a Face?», 2021

Computer vision requires strict definitions. Face detection algorithms define faces with exactness, although each algorithm may define these parameters in different ways. For example, in 2001, Paul Viola and Michael Jones introduced the first widely-used face detection algorithm that defined a frontal face within a square region using a 24 × 24 pixel grayscale definition. The next widely used face detection algorithm, based on Dalal and Trigg’s Histogram of Oriented Gradients (HoG) algorithm, was later implemented in dlib and looked for faces at 80 × 80 opixels in grayscale. Though in both cases images could be upscaled or downscaled, neither performed well at resolutions below 40 × 40 pixels. Recently, convolutional neural network research has redefined the technical meaning of face. Algorithms can now reliably detect faces smaller than 20 pixels in height, while new face recognition datasets, such as TinyFace, aim to develop low-resolution face recognition algorithm that can recognize an individual at around 20 × 16 pixels.


As an image resolution decreases so too does the dimensionality of identity.