I'm looking for a program that can do reverse image search (such as Google images or iqdb) of a image file then tag the image with appropriate tags, whether it is in the images Metadata OR renaming the image file.
I understand that there are some paid services that do that but they use their own image ID software that tags stuff like "tree, smile, face" and the previously stated image searches return appropriate or relevant tags.
Thank you
Other urls found in this thread:
en.wikipedia.org
image.wiki
docs.api.image.wiki
tmsu.org
illustration2vec.net
demo.illustration2vec.net
iqdb.org
github.com
github.com
caffe.berkeleyvision.org
caffe.berkeleyvision.org
github.com
twitter.com
Just a small bump
Maybe I didn't explain correctly.
I want to do a mass tag or file rename of images I downloaded. I've downloaded thousands of images from Sup Forums so having to go through one by one and putting them in their own folder is exhausting and taking too much time.
I want a free program or site that let's me upload an image, does a reverse image search, then based on the results, add the tags to the Metadata or rename the file name.
Does such a thing exist? Does anyone know of such a thing?
>I want a free program or site that let's me upload an image, does a reverse image search, then based on the results, add the tags to the Metadata or rename the file name.
this is clever as fuck, if no onw has done it, then I'm going to do it
Hey, my idea. I'm currently working on it. Just want to know if it's already done so I don't waste my time
sorry user I guess I can't help if you're working on a propietary solution, but I would totally release my solution for free as in freedom or gladly help you if you do it as well
the best I can do is to point you to product hunt, alternativeto or slant
I would not be against a collab
I was believeing OneDrive and Google Photos would do this automatically, but nope. I just want to be able to search for smugfrog and have all my rare pepes show up, I dont want to organize them myself. I think Watson has some kind of free tier for tagging.
I've been counting on that someone would write this. I'll be very grateful if you deliver OP. I specifically care about fetching tags from danbooru + gelbooru.
very clever and specific, doubt its been done before, please do update Sup Forums when its finished, for there will be praise waiting for its arrival
You may want to integrate with something like this: tmsu.org
Also, I think FUSE might be useful for doing something similar to that.
>tfw new program incorrectly labels dozens of your file as cp
knock knock
Common, man up and write your own neuronal network for autotagging.
How hard can it be? I mean there are already good libraries out there AFAIK.
I'll start on the logo.
your idea is pretty interesting. Keep us posted with your progress!
OP here
Wow, I'm pleasantly surprised to see such a reaction to this; I expected either links or telling me how stupid I am for some reason. Thank you all for the response! I'll accelerate my work on this.
My friends got a booru ripper, has nearly every image with all tags from the main boorus. Fun shit.
A md5 is usually enough to get taks from any given image you see on the chans, and basic image search handles the rest.
I like to think that scratch on the table is her navel and that she's not wearing pants and is actually fucking the guy.
The problem is accuracy.
Using Google's reverse image searching and grabbing the data from what they think is related to it won't be accurate at all, it will often fuck up.
If you're in to neural network stuff, you could build and train your own model. There's also existing applications which are getting there, e.g. if you want automatic tagging for anime-style images, you should definitely check out illustration2vec which you may have already seen posted here a few months ago: illustration2vec.net
I swear there was something like this a few months ago
some japanese PHD/masters thesis or research thing
semi accurate if I remember
FOUND IT
...
I remember some guy made a camera that would print out a description of the image it captured, but it used Mechanical Turk to generate that description IIRC.
doesn't Hydrus do this, sorta? if you get it all working right.
There is no such Booru ripper
It puts everything into one folder so it's shit.
Can anyone find this?
iqdb.org
I suggest going to illustration2vec.net
I have not tagged my images yet, but I've played around with it and it does work.
Loading the model (or training the network, I think) takes some time, but once it's done, tagging images doesn't take too long and you could go through your entire collection within a few days. Really depends on your processing power though. I tested mine on an i3 550 @ 3.2GHz.
Also, in response to , the tags generated by i2v are pulled from boorus, and the model is trained on boorus. so it's the ideal way to tag your images if 99% of your collection is anime, like mine is.
I too share your frustration of having all my thousands of images all with unix timestamps and all shoved into on folder
I've tried to do the same, using the reccomended terms or the results of a reverse image search to know what's inside an image. It's really hard to automate, and i've given up on the thing.
If you manage to do it, i'd like to see how you managed to.
It's your mother.
The Hydrus Client does something similar, OP
The images I want to tag are pulled directly from boorus and I want the exact tags that they used, not some NN guesses.
>saving a bunch of anime art from imageboards and not from pixiv or booru's
>not organising as you're saving
Moron.
Use this in the future
github.com
GNU/Linux, Windows and MacOS.
This looks good but it needs more usage information on the readme and maybe some screenshots. I can't figure out how it works without compiling.
If you do make something like this OP, please share. I actually could use it for something.
The NN guesses are pretty good though.
Try the demo on the site, seriously.
I've tried it. It's impressive, but still makes a lot of mistakes that fetching the tags directly from boorus won't have.
the problem is tags change every day on boorus.
what needs to be done is to download the entire fucking tag database on danbooru and let it automatically tag your shit based on the filename because boorus never change the filename
/h/ has a steady general for this sort of thing.
>1girl
lmbo
Better than UNIX timestamps.
Selenium is your friend.
# -*- coding: utf-8 -*-
from selenium import selenium
import unittest, time, re
class Image Search Automator(unittest.TestCase):
def setUp(self):
self.verificationErrors = []
self.selenium = selenium("localhost", 4444, "*chrome", "yande.re
self.selenium.start()
def test_image _search _automator(self):
sel = self.selenium
sel.open("/post/similar")
sel.type("id=login-popup-email", "XXX")
sel.type("id=login-popup-password", "XXX")
sel.type("id=file", "/XXX/XXX/XXX/XXX/XXX/picture.jpg")
sel.click("css=td > input[type=\"submit\"]")
sel.wait_for_page_to_load("30000")
sel.click("link=Use all services")
sel.wait_for_page_to_load("30000")
sel.type("id=login-popup-email", "XXX")
sel.type("id=login-popup-password", "XXX")
sel.click("//li[@id='p1']Sup Forumsspan[3]")
sel.wait_for_page_to_load("30000")
def tearDown(self):
self.selenium.stop()
self.assertEqual([], self.verificationErrors)
if __name__ == "__main__":
unittest.main()
Use Requests and Beautiful Soup to get the tags.
Bumping because more people need to know about this.
So compile then.
Its easy as pie to use.
ive also struggled with this for years, trying to catalog images i have/had around. i finally decided to abandon the project, since the images, even with the proper tags input manually, hold no value. its the service that allows to extrapolate tags from image inputs that holds all the value.
maybe create a captcha service, like google has, and let other folks tag your pictures for free? gl
This but by md5
I will, but the point is that I shouldn't have to. Yeah, yeah, the person who made the software doesn't own others anything, I'm just saying that it would be really helpful if more people did that. I'm tired of having to figure out stuff that should be straightforward. Compiling shit is often a pain too, so I want to know beforehand if it's worth trying.
>maybe create a captcha service, like google has, and let other folks tag your pictures for free? gl
That's brilliant.
What does unix timestamps have to do with anything?
>tfw on Windows and never worry about things like that
not him but what about setting up some repo just so it would be easier for work, I might add some code too
I've been wanting the same thing forever.
will try all the suggestions itt when I get home
You also don't do the things that require that kind of thing. Don't kid yourself, this is a problem on Windows just as much as on Linux. Yeah, sometimes there's a Windows binary available when the same isn't in your Linux repos, but a lot of stuff is a pain to setup on Windows because it's a second-class system.
Heck, the software we're discussing in this very thread is harder to setup on Windows.
Check the archive for Sup Forums, I believe there was a project doing just this a while back that would tag images just as you said. Booru style tags using machine learning I believe was the topic, probably last fall 2015
...
ty senpai
>a lot of stuff is a pain to setup on Windows because it's a second-class system
While that's true, it doesn't better me because I don't do any Sup Forums related things other than fucking around with hardware.
I'll start working on the logo
All logo related jokes aside, I'd be happy to pitch in dev time as well.
>labeling a file as CP makes it illegal to possess
that's not how the law works
isn't it missing sanic and hyperscat tags?
He's talking about pictures downloaded from Sup Forums you dolt
In case this is your first day here, Sup Forums pictures' filenames when downloaded are the UNIX timestamp when uploaded.
Isn't “difficult compiling” primarily a Windows-thing? What with all the MinGW, MSYS, cygwin etc. nonsense floating around of which half is completely broken.
Ah, duh. My filenames are all md5 hashes though, and I don't want to change them, I want to add the tags in some form of metadata.
Apparently I'm tech-literate
How do I use the .caffemodle from the illustration2vec.net
Same for me, sounds like a fun project.
Also I just finished my graphic design degree, so I could actually make a logo. :^)
This is incredible. I haven't had much practice with python but wow, a simple search and I find many that discuss selenium to save as well as dealing with tags.
Thank you for this!
It's kinda sad init
Bump because I need this in my life
My bad, he wrote it I should have said.
He's done it twice so far, once in C, then in python. He's too fucking busy studying so he hasn't been maintaining it, so I can't rip shit. Proper sad.