Everyone talking about passive camera surveillance got me thinking that there must be a way to evade automated person-detection through some exploitable weakness in cameras, especially if accepting high visual conspicuousness to humans if a practical necessity. I've hit a wall and wondering if anyone has ideas. I've played around with specular, retroreflective, and lenticular materials, as well as tried to find a workaround with weird RGB combinations. No luck so far.Obviously, if practicality is removed as a constraint, this is possible. If a person were to put a cardboard refrigerator box over their head such that it completely covered their whole body, person detection would degrade completely. There must be something less obtrusive that would be effective.I believe that the currently most researched arm, adversarial patterns designed to confuse the AI/algorithmic data processing to evade person detection, is misguided. These patterns seem to be ineffective in real situations with multiple angles, etc. They also seem easy to quickly train against. Therefore, I believe it would be better to target the data collection side of things rather than data processing.Further, personal identification seems easy to evade. Covering the face and altering gait is relatively straightforward. Tracing a person's movements back to a source would still be effective lacking personal identifiers. Seems like we need to avoid person detection entirely.
>>17060049Too late.
>>17060055Printed adversarial patterns on garments are weak across differing distance, angle, lighting, and camera ISP pipelines. They also leave a moving human-shaped object, so models can be easily retrained against them.
>>17060060Says you. But you offer nothing to demonstrate your incorrect opinion.Again. Too late.How does your funding compare, btw? Too late?
>>17060064I'm not trying to start a business, I just hope someone finds something that actually works reliably. Here's the maker of the kickstarter admitting that the biggest shortcoming is true
>>17060072>I just hopePonies are on another board.You have nothing. STFU&kys.
>>17060072“The biggest shortcoming is that it works too well.”Kek
>>17060049We need to start wearing blackface at all times. Vantablackface
>>17060060They also flag you as explicitly adversarial to the system.It's the same with retards that don't understand how the internet and the world wide web work thinking their "privacy browser" they cargo-cult install makes them more private instead of making their personal behavior and identifier even more unique.
>>17060175How do they know who to flag if they can't ID you?That is not true in a cargo cult situation, only if you are the only one using the "privacy browser", but if a bunch of people are all using the same "privacy browser" and recommended settings, then they will all blend in together and have the same identifiers rather than being unique from each other, that is the whole point of privacy browsers to make everyone who uses the same ones appear the same.
>>17060245Correct, which is the reason I had to start using SmoothPorcupine's theorem of standards deviation for potential crowdsource displagiarism.
>>17060245That value is missing for a reason.There is an implication, which is a vector score and highly crunchable by AI.
>>17061052In fact, nothing scores higher in finding patterns in the missing data than AI.It found all those lost proofs, didn’t it.
>>17060049>surveillance evasiondude, with vertically federated learning, that ship has sailed
>>17060136Fuliginface...
