To the owners of Jow Forums, or the moderators or whomever it may concern: I am a computer programmer with some background in basic machine learning. I would like to donate some of my time to create a simple neural network to filter out a certain avatar fag. I am serious about this. Who do I contact to get the ball rolling on this?
Neural network will be created and trained in tensorflow, then the connection weights will be exported and at that point I will get the trained network up and running in whatever programming language Jow Forums uses. Free of charge.
Why bother? The tranny will just adapt until they can get around the filter.
Jonathan Peterson
A simple neural net can detect similar images, even with a lot of differences. For example, rotating the image, blurring it, etc all will not work. The freak would have to change to a completely different avatar, and at that point, it has lost the battle.
Hell, considering the consistency between the avatar images currently, machine learning is not even needed. An extremely simple program could detect all the variations of the fag's avatar as it stands right now. But neural nets are so easy to make these days that I might as well resort to that right away.
Brandon Sullivan
Plot Twist: OP is the very avatarfag under discussion, trying a new tactic to remain the center of attention.
Easton Baker
How about you get muted if: -topic is too similar to a previous topic -image is too similar to a previous image
Cameron Wood
Topic of course means comment
Aiden Taylor
To generalize it to that extent would make the solution much more complex. Or maybe it wouldn't, I don't know; Maybe there are already decent, fast algorithms in existence for detecting similarity between arbitrary images. However, I don't know how to do that, and I'm not willing to spend more than ~4 hours on this project, so I would like to just digitally murder this one avatar and leave it at that for now.
He admitted to making "Posters you recognize" threads for this very same reason in the past. I wouldn't be surprised if it did.
In case it's not I think a more efficient method would be shadow banning, but imho from what I know that'd be heavily resource intensive on 4chans backend under its current form. And it'd open the gates to worse forms of dissident silencing.
Julian Roberts
I feel like many reposts and minor edits of images are valuable, though. You wouldn't be able to draw something funny on someone else's picture and upload it again.
Camden Cook
We could try to solve it ourselves by always responding to those threads with upsetting images and material likely to trigger trannies.
Thomas Davis
A purely front-end solution could potentially be hilarious. Imagine if there was javascript in place which polled the server for the user's banned status, and if the user is banned, the javascript generates a fake post and presents it to the user as if it was business as usual. Even someone who is code savvy would probably not realize it for hours, and even then, they would probably have to de-minify the javascript and troubleshoot for a couple hours to confirm for themselves what is going on
Evan Perez
I would love to see some edits of the freak's avatar with a realistically slit throat, gouged-out eyes etc.
It is usually less complicated to abstract it. In fact, I can already visualize some short pseudocode where a ML object just accepts rows of binary data:
# run once imageML.build(imageDB) commentML.build(commentDB)
# run per post bool isOriginalPost() { if (imageML.test(image, imageThreshold = .1)) return false if (commentML.test(comment, commentThreshold = .01)) return false return true }
Henry Baker
That is the purpose of a configurable threshold. If an image is 5% more original, it can be allowed.
It just does not seem like machine learning is the right tool for that job. You would be much better off just doing a direct comparison of the binary data, with an algorithm similar to whatever git uses for diffing files. Furthermore, duplicate images are not a real problem; There is plenty of legitimate reason for different people to post the same image at different times, the obvious example being feels guy or pepe. The problem is specific avatar fags. I really think the most practical, simplest solution is to have neural nets trained on an avatar-by-avatar basis.
Ryan Russell
>doing a direct comparison of the binary data, with an algorithm similar to whatever git uses for diffing files
I should clarify, this is assuming doing it on a thread-by-thread basis. Obviously not practical to compare to every image ever posted. But some kind of model has its impracticalities too, namely having to train it regularly to learn new images.
Michael Morales
>feels guy or pepe I actually designed this because I'm tired of seeing those very images over and over, just like you hate that avatarfag.
No joke, I may just write an Edaha module and host a full instance in Google Cloud just so that I can shitpost by myself with no frogposters.
Ryan Fisher
Devilish. That's actually a good idea user, and sounds simple enough to be done by even Jow Forumsentoomen. Though it'd have to be reserved for special occasions lest the shitposting masses find ways around it.
Jace Reed
I guess we can agree to disagree. I don't see much comparison between an avatarfag and frogposters. pepe and feels guy are actual memes, they express something and have an application. An avatar is purely an attention-seeking thing. But if you have the time and the will to write it, then go ahead, just don't lobby to apply it to Jow Forums as a whole. Hell you don't need a database of Jow Forums's images, just get an image database, add in a variety of pepes and wojaks, and train using that.
