Adam Harvey – Tisch School of Arts, New York University
CV Dazzle: Open source camouflage
CV Dazzle: Open source camouflage
CV Dazzle is camouflage from face detection. It is a response to the growing prowess of computer vision technology and the resulting phenomenon of shrinking privacy.
Concept + Photography: Adam Harvey – Interactive Telecommunications Program, Tisch School of Arts, New York University Thesis Advisor: Despina Papadopoulos Counsel: Sarah Scaturro, Mark Shepherd, Giana Gonzalez Hair Stylist: Pia Vivas Makeup Artist: Leigh Brown Models: Jen Jaffe, Irina, Jude, Maria, Mickey DIS Magazine (Lauren Boyle, Marco Roso, Solomon Chase) Obinna Izeogu
If one of the key roles of conceptual design is to ask questions that conventions and norms have obscured, then this design piece exemplifies such an interrogation. It partakes in “design-fiction” and projects design behaviors of people in a time where technologies such as face recognition software and monitoring in public spaces is pervasive enough that all privacy is compromised. It picks up the issue of privacy in an increasingly digital and sensed world and shows that design can be used at a personal level to subvert and maintain autonomy. This piece of work is avant-garde, witty, internally consistent, nuanced, and makes us think about the invisible shifts taking place around our civic lives. Making the canvas one’s own face and body – rather than a garment – drives the message of autonomy home even further. The judges felt that the CV Dazzle project scored high on conceptual integrity as well as in design expression.
CV Dazzle: Open source camouflage
1. Summarize the problem you set out to solve. What was the challenge posed to you? Did it get you excited and why?
The overarching challenge in designing CV Dazzle was to thwart face recognition by blocking face detection.
This project arose out my interest in photography and our changing relationship to surveillance and privacy. I was excited by the revolutionizing effect of digital photography in the last decade but also concerned about the repercussions of our increasingly interconnected cultures and sophisticated visual-data mining tools. The impetus for designing a camouflage against face detection was to show how we could adapt to occularcentric, surveillance-societies without retreating into anonymity, and, in doing so, celebrate style and augment privacy.
I was excited to work on this project because I dislike the one-way communication inherent in surveillance. Here, I’m referring to surveillance by unmanned cameras, but also the casual surveillance that occurs when someone inadvertently uploads your photo to the Internet. In both cases, the subject is unaware of being photographed and can be unfairly subjected to anonymous visual scrutiny.
The reason to design a camouflage against face detection and not recognition is influenced by the fact that face detection is the first step in automated face recognition. Therefore, any obstructions of this first criterion should, theoretically, thwart any automated facial recognition system. Without a face, there can be no recognition. This is the underlying principle of how CV Dazzle works.
2. What point of view did you bring to the challenge? Was there anything additional that you wanted to achieve with this project or bring to this project that was not part of the original brief?
Going into this project I had a lot of experience as a photographer and I think that this informed my direction towards a highly stylized solution. During my first few years in New York City I worked nights as a party photographer to hone my skills and supplement income. During the day I would read blogs about other photographer’s party photos and found that some of the most popular images were, of course, the most scandalous. I become a little self-conscious that I, too, was indiscriminately posting incriminating images of others online, permanently archiving them for future generations of data-mining robots. Perhaps I was little bit guilt-ridden or just sympathetic to the victims of casual surveillance, but I really wanted to approach this project from a photographer’s perspective and make the overall experience of surveillance-based photography less authoritarian and more playful. Even though this project is about hiding from cameras and augmenting privacy, it also about creating a dialogue between the subject and the camera, and encouraging more creativity.
3. When designing this project, whose interests did you consider? (Discuss various stakeholders, audiences, retailing, manufacturing, assembly, distribution, etc., for example.)
In the beginning, I saw this project as a something that would only be suitable for partygoers because it was going to require a lot of bold makeup and hair styling. And because the concept carries futuristic overtones of anti-big brother sentiments I tried to balance the need for it to feel this way yet make sure it doesn’t die from being overly conceptual. I figured it to should be flashy enough for any of my friends to wear at a party yet practical enough to wear into a bank without causing a scene.
In terms of accessibility, there were three main criteria. Firstly, it must be legal. Some cities, including NYC, have anti-mask laws that prohibit disguising oneself in public. Second, it should be a low-cost solution so that people can feel free to experiment with it. Third is the scalability. I didn’t want this project to be limited to a current season’s trend. I wanted the final outcome to scale with time, and allow other designers to build on top the work I’ve done with their own vision of how it can look.
An important part of the design is that there is no absolute solution since that could easily be integrated into future face detection algorithms. The strength of the concept lies in the infinite number of ways it can look, which I think is also nice metaphor for the importance of personal expression.
4. Describe the rigor that informed your design. (Research, ethnography, subject matter experts, materials exploration, technology, iteration, testing, etc., as applicable.) If this was a strictly research or strategy project, please provide more detail here.
I began this project by creating a simple Processing sketch that allowed me to draw over a face until it was no longer detected. This proved that there was some potential in this method, but most of the designs required a large portion of the face to be covered so I wasn’t convinced this project would succeed. Next, I began looking for pre-existing makeup styles that might thwart the face detector. I scraped websites for party photos, in particular sites with images from the Boombox party scene in London and Oceanian face decoration, and tested these images for makeup patterns that broke the detector. What I found was that there was no clear pattern that emerged and a lot of the bold tribal face-paint actually improved the detection process. However, there was one image that seemed to work well: a picture of Lady Gaga as Jesus Christ. Ironically, this gave me hope.
I then manually rebuilt the entire face detection algorithm in Java and created visualizations of the detection process, including a 3D model and statistical analysis of the key facial features, such as the mouth, nose, and eyes. This proved insightful and I could begin to see which types of patterns would break the detector.
Throughout these stages I consulted with several computer vision experts and fashion stylists. I also received a lot of important feedback from my fellow students and thesis advisor. Most of the technical information I needed to learn about face detection was available online and in a few research papers.
With an improved knowledge of the face detection process and possibilities for makeup and hair styling, I rebuilt the drawing application and used it to generate a contact sheet of designs that seemed like they could work in real life. With the help of a hair stylist and makeup artist these looks were then applied to a model, photographed and tested. The first CV Dazzle design not only worked again the targeted face detector, OpenCV, it also worked against Face.com, iPhoto, and Picasa.
5. What is the social value of your design? (Gladdening, educational, economic, paradigm-shifting, sustainable, labor-mindful, environmental, cultural, etc.) How does it earn its keep in the world?
I think the value of this project is in proving that the concept will work and that, through creativity, we can adapt to living in surveillance societies without retreating into anonymity. I hope that one day we will change our existing models of compromising privacy for security and permanently archiving our public and private moments on the Internet. But, in reality, this is increasingly less likely to happen with each advance in networking and digital imaging technology. And until that paradigm shift occurs, I hope that this project serves as a model for how we can creatively adapt to authoritarian technologies.
6. If you could have done one thing differently with the project, what would you have changed?
The most enjoyable part of this project was collaborating with others to bring it to life. Through the first three months of developing the concept, my biggest fear was that the designs would have to look ridiculous in order to block face detection. I kept hesitating to bring it out of the computer and into the real world. In retrospect, I wish I had done that earlier. Because when I did, I found that seeing in its physical form facilitated visualizing the subsequent digital prototypes.