4 ms·
> Sure, you could argue for further abuse scanning like Google foolishly tries with content ID but the real solution is to go after the uploaders. If Pornhub is
by mschuster91 4y ago
> Sure, you could argue for further abuse scanning like Google foolishly tries with content ID but the real solution is to go after the uploaders. If Pornhub is required to ask for special validation so is every single service with user postable content.
Content ID actually isn't so foolish. It works reasonably well to target copyrighted material, the problem rather is that it by nature cannot determine "fair use" exceptions and getting an actual human to review false strikes is an unreasonably excessive effort.
When someone notifies Pornhub that a video they host is distributed without the consent of the persons involved, it should be very very easy for Pornhub to scan for other videos containing the same content, even as snippets, and subsequently take these down. The technology exists, both as SaaS (e.g. Amazon Rekognition) or as open-source solution to self-host.
Side note, I always did wonder how PornHub and friends were able to evade the record-keeping requirements of 18 USC § 2257.
- jeroenhd 4y agoContent ID is extremely sensitive to audio and vague video. Upload half an hour of white noise and you'll have at least three content ID strikes. Actual "songs" get stricken easily as well because someone uploaded their own rendition of a free song, uploaded it to YouTube, and now all renditions of the song are "theirs". I agree that Pornhub should remove and block all related content but finding such content is not a solved problem. The article talks about small snippets being uploaded independently, you're never going to find those. If proper contextual recognition was a solved problem we'd have self driving cars by now. It's very easy to overestimate image recognition technology. We're getting closer every day, but we're still far from an real AI.
- mschuster91 4y ago> Content ID is extremely sensitive to audio and vague video. Upload half an hour of white noise and you'll have at least three content ID strikes. That's mostly because Youtube has allowed the major labels to directly upload stuff to Content ID instead of doing quality control to make sure the content is actually valid and at a quality that allows actual comparisons. > The article talks about small snippets being uploaded independently, you're never going to find those. Take each video, create a visual hash of each frame (or at least for each detected scene). When a video is reported, compare its frame hashes with the database, manually examine matches if they are identical, and do the same for all new uploads. Easy enough to get at least a sensible start on the problem and to crack down on the worst abusers. Identifying all videos of a given person is harder, granted, but seriously, at 97 billion $ of revenue [1] each year it should not be too much of an expectation that Pornhub actually does something beyond requiring victims to scan for their content on their own! [1] https://www.dailymail.co.uk/news/article-10058065/Owner-97bn-Pornhub-named-Pandora-Papers-defends-investment-saying-taxes-paid.html https://www.dailymail.co.uk/news/article-10058065/Owner-97bn...
- jeroenhd 4y agoVisual hashes collide all the time, though. Apple's child porn scanning algorithm hashes were reversed and images could easily be doctored to collide. If anyone has the resources and technical prowess to develop a good algorithm for this stuff, it'll be the Cupertino people bringing home obscene amounts of money for writing computer code. The black magic they use to achieve things like 5G speeds, reliable 3D vision, and even just the way modern flash storage operates, is just baffling. Every article I read about modern computing makes me think we're all just practicing witchcraft and I'm confident the smart people behind the modern smartphone can find a way to make the system work if there is any. However, I don't think the current solutions are feasible yet. Tie someone in a gimp suit to a chair in some sex dungeon and you'll have no idea who's sitting where. In a less extreme example: some people don't like to have their face on camera when filming erotic material. You're working with partial bodies in similar positions making similar noises and movements, I think it'll be one hell of a task to properly identify unique matches in a way that won't lead to content ID abuse like on YouTube. I'm also not so sure about having a company fingerprint the way people have sex "for their protection" to be honest.
- mschuster91 4y agoI agree with you, which is why you always need a human in the loop. The key thing is to tackle the low hanging fruit, which is the utter utter majority: people dumping the videos and photos of their ex girlfriends, tinder hookups or whatever over and over again unaltered. A ContentID-like system could save a lot of pain for the victims at relatively low cost here. Additionally, it could help to rein in the utter absurd amount of theft going on - pretty much everyone with an Onlyfans or whatever account can tell you more than enough stories about people dumping fresh content in a matter of minutes on imgur, Reddit and tube sites, not to mention catfish accounts. The more difficult stuff such as content without faces - I'm actually inclined to default for "doesn't need to be removed until there is something like a tattoo or scars that clearly identify a specific person". But again, the key is to tackle the mass offenders.
- stevenicr 4y agoI would like to explore this: "The technology exists, ... as open-source solution to self-host." further - can you share specifics, my google-fu is not showing me.. re: wonder : I believe phub is not a US corp and so they don't need to follow USC anything if they don't want to.
- mschuster91 4y agoSomeone has written a comprehensive howto: https://pyimagesearch.com/2018/09/24/opencv-face-recognition/ https://pyimagesearch.com/2018/09/24/opencv-face-recognition... Someone else has managed to get face recognition running in real time on a Raspberry Pi, which is mind-blowing on its own given how lackluster performance was before the Pi 4 came along: https://towardsdatascience.com/real-time-face-recognition-an-end-to-end-project-b738bb0f7348 https://towardsdatascience.com/real-time-face-recognition-an... And all of this is from 2018 - hard to imagine a company the scale of Pornhub could not hire a bunch of ten developers and let them google "opencv video face recognition". For something like Content ID, I'd use ffmpeg to get raw stills of the video, normalize them into 720p resolution and then go with a visual hash implementation (again OpenCV has enough howtos on that topic) over these. Yes it's a ton of data to store but it should not be too difficult to implement at the scale of pornhub.