The Whys and Hows of the Opt-Out Cap.
Oh bother. It’s all a bit silly really.
We all scoffed at selfies, then embraced them. They became common place. We amassed an enormous data set of our faces and inexorably tied them to our digital persona (1). Other than worrying about a haircut sported at the time of capture, we never had a reason to second guess getting our photos taken for a drivers licence (2). We downloaded apps that applied beautification filters and made us look like our parents (3). Maybe it was a stranger approaching us and offering $5 (4). And then the models were trained and the software started getting it right. The machines started to recognize us without prompting, and that’s when the realization hit.
We’d given our faces away for free, and there was no way to get them back.
We’d been scraped, categorized, placed into databases, and used to train machines that could be used against us. We were now part of a system in which we had no control.
That’s why I made the Opt-Out Cap.
I’d had the idea for the cap earlier this year after finishing William Gibson’s novel Zero History (Published 2010), the final book in his Blue Ant trilogy (Spoilers ahead).
In the book, characters are tasked with making their way through London’s streets without being recorded within the CCTV system. The Ugliest T-Shirt Known to Man displays a graphic of a human face printed across its entire exterior, that when worn is then tracked by CCTV systems in place of the wearer. The face on the Ugly T-Shirt is registered within the CCTV facial tracking system, and due to a clandestine agreement is coded to then be erased from the system automatically. You can read more about it here (5).
While science fiction, its not too far off from what’s needed now.
So then why make it now?
Awareness of facial recognition is having a bit of a moment, largely spurred on by the reports coming out of China around the protests in Hong Kong. Protesters face a huge number of repercussions from being identified, from being doxxed (6) to being attacked for organizing (7). The protests are viewed as unlawful assemblies, and thereby anyone in them can be charged with a crime. On top of that, an enaction of emergency powers by Hong Kong’s leaders further banned masks from being used by protesters, If you’re caught wearing a mask at a protest, you can face up to a year in jail and a fine (7). This Enables further negative consequence to those that would attempt to express their wish to protest.
Making It
After mulling over Zero History for a bit, it became pretty apparent that the general concept behind the Ugly T-Shirt could work. Substitute one face with another for a facial recognition system to latch onto and you’d effectively render those systems useless. Even if replacement faces are imperfect, these systems are trained to lock onto them nonetheless. Look up “Face Swap Glitches” for a particularly funny illustration of this concept. Covering for the face with another face on it would do it.
Sketch #1, in all its MS Paint glory.
I started sketching out ideas, thinking of how to build in an automatic deployment and retraction system, how to custom manufacture the cap to keep the panels inside of the bill of the cap, how to spec out the graphics so as to be printed by a dye-sublimation vendor, that sort of stuff. Thankfully the K.I.S.S. principal kicked in and I pared down the design to approximately what became the cap in the plans. 3 panels with applied graphics that was dead simple to make and to hide.
And with that, I got started.
The first order of business was to figure out how to get the graphic prints to work. The Cap needed to be able to be seen through when in use, which precluded making the hanging panels out of anything opaque. The graphics applied to the the hanging panels should be something that could be done at home at fairly high resolution, which ruled out drawing the panels with sharpie or using screen printing to print on mesh. I thought my only option would be to work with a textiles printing service to get a custom run of sublimated mesh produced given those limitations. Thankfully a friend mentioned T-Shirt transfers as an option.
These sorts of transfers work by using common consumer printers to apply ink onto a sheet of heat-activated vinyl. These sorts of transfers are made to work with inkjet and laser printers, and come in varieties that work on light, dark, elastic, and stiffer mediums. They typically come in the form of a US Letter form factor (8.5” x 11”) and are sold at craft stores in the US for cheap ($12 for 5 transfers). These would work well for the graphics. The actual panels they were printed on were a different matter.
I knew I wanted to use a mesh, but as to what type I had no idea. With that in mind I started poking around online fabric stores to get a sense of what was available. What I quickly found was that there were two typical categories for mesh - decorative and functional. While an embroidered paisley would be fun to work with, it would play hell with visibility (Although a FR blocking wedding veil would be an interesting art piece…). I started looking into functional, and ordered a yard each of 4 mesh styles in white to test with.
Power Mesh - Typically in pantyhose and other undergarments - Nylon / Spandex mix, $11.95 / yard - Link
Standard Mosquito Netting - Used to keep out the bloodsuckers - 100% polyester, $2.99 / yard - Link
‘8838’ Athletic Polyester Mesh - Good for jerseys or bag liners - 100% Polyester, $7.25 / yard - Link
‘1571’ Medium Polyester Mesh - Used in traffic vests and bags - 100% Polyester, $7.50 / yard - Link
From there I started working on how to test the transfer sheets. As I was working with an inkjet printer, I selected a few transfer sheets that were: 1. Made for white or light colored textiles and 2. Were made to be printed on an inkjet printer. The 3 selected transfer types were:
Avery 3271 - US Letter, Non-Stretch, 6 sheets per pack, $15.59 per pack- Link
Avery 3302 - US Letter, Stretchable, 5 sheets per pack, $13.99 per pack- Link
PPD PPD-1 - US Letter, Non-Stretch, 10 sheets per pack, $13.35 per pack- Link
I then quickly whipped up a test graphic to print and devised the my test criteria:
The transparency of the mesh with applied graphic needed to be high, but not show the users face underneath. The mesh needed to respond well to the graphic application without significant warping. The fidelity of the printed graphic on the final mesh needed to be fairly high. Finally, the materials being tested needed to be easy to work with.
Try each transfer type on each mesh type + a control of cotton linen. To control for uneven application by using a household iron that could cause irregularities, I use a heated press set to 380°F to evenly apply pressure and heat for 3 minutes. A1 was the Avery 3271, A2 the Avery 3302, PPD the PPD-1. All pieces were stretched after transfer.
Of the tests, it became quite clear that the Power Mesh was a winner on the mesh base layer side. After the transfer, it remained semi-transparent while still obscuring what was underneath it. It also was warped very little after the initial stretch fresh off the press. On the Power mesh, the Avery 3302 distorted the least and didn’t fractured the least. As the Power Mesh is partially spandex, it makes sense that the transfer made to be stretched would fair the best. Surprisingly the PPD-1 was an easy second place, and distorted the least after a stretch. The winners then were the Power Mesh and Avery 3302 Stretch Transfers.
As an aside, the mosquito mesh performed surprisingly well, and found be interesting to dial in the right transfers to make it work best. Given its abundance and useful original purpose, it could be well suited as an alternative to the Power Mesh.
With that, I moved onto designing an printing the graphics. I’ve worked with wrapping graphics around 3D objects before, and it usually involves the generation of a tool called a UV Map, which is a complicated and computation intensive process. Thankfully my Drawing 101 instruction came back to me, and I remembered that the human face can be divided into 3 planes, loosely contained as vertical rectangles. The plane front contains most of the recognizable features of the face, and the 2 side planes comprise the cheeks and outer edges of the eyes at roughly a 3/4 view. With that in mind I sketched I developed the template to work from.
From that point it was just a matter of printing the full graphic onto the transfer paper in color and then affixing it to the hat. For the initial affixing, I used masking tape to hold the mesh panels to the sweat band of the cap.
And with that, the process was pretty much complete. All that remained was to properly write up and document the instructions. (Which, as it happens, are available here - Link to Blog Post.
As it turns out, that whole process was a scope of work unto itself. Writing up easy to follow plans is no joke, and creating and editing photos and videos is a skill I’ve been trying to improve for a while. Most of it isn’t too interesting.
However, to get this one video , I had to set up a well lit photo studio in my space. Thankfully I’d been designing lighting mounts and had been getting better with my DSLR, but this still posed a challenge.
to get this one shot, I had to shoot 12 takes of the same action. I had to not look at the camera, figure out a the ways to move my head to show off the face on the panels best, and make sure to not move out of frame when I was moving.
On top of that, I then had to take the footage and then color correct it, crop in from 16:9 to 1:1, track my head, and figure out how to export the video in GIF form in such a way where it wouldn’t balloon into an enormous file size.
So that’s how it came together. It’s now published and available for all to use, and has been seen by quite a few people so far - Reddit Post.
I’ve seen a lot of excellent feedback on it so far and am thinking about how to incorporate it into further revisions. In particular, adding in some way to secure the panels to under the chin so that they don’t move in the wind.
But that’s all for now, and I’m sure I’ll be working on this more shortly.
Cheers,
Mac Pierce