I have a massive collection of mostly anime images all with unix timestamp filenames and all in one directory

I have a massive collection of mostly anime images all with unix timestamp filenames and all in one directory.
I want to tag them all, or at least >75% of them.

I tried Hydrus, and not only was it slow as fuck, but it could only tag ~17% of my collection (from multiple smaller samples, importing my entire collection would take waaay too long just to test it), so that was a disappointment.

Illustration2vec is the only option I have left, and it's pretty good at tagging anime images, so it does what I want.
Even better is that I can run it locally for maximum throughput.
There is just one problem - it's way too slow, even when running locally.

Is there any other hope for me or should I just give up?

For the tagging to workz you have both to make sure your tag DB to be up to date and understand as how it's based in the picture hash and not "how the picture looks" that the tagging works. If you don't collect pics from the boorus or sources and take slightly changed pictures (like resized or jpg-shittied) you're fscked.

I know, and yes of course I made sure my tag database was up to date.
And yes, I thought it was a problem with the hashes, but I'm not sure why. I haven't modified any of them, but they're also mostly downloaded from Sup Forums, so maybe Sup Forums has changed them somehow, or the people that post them, which I doubt because they're usually posted straight from boorus or whatever.

Also, I assume it uses a hash of the pixel data, not the actual file, so things like changing the headers should not affect it.

Tagging networks like hydrus obviously don't work, so the only option for me is to tag it with neural networks, but I guess I will have to wait until they get a bit faster.

Post your Reisens

I would if I could easily find them.
That's the problem I'm trying to solve.

Prolly picture tags or slight changes in compression for bandwidth.

But yeah, NN tagging seems like a dream. Let's see how it fares.
If you post shit, care to link to the thread?

None of your images are unique or special. Why waste your life categorizing a bunch of redundant junk? Just delete the lot and use google image search.

DESU this also.

For a moment I was interested into sorting my shit. But in the end it was redundant as I had already more or less enough of the pictures I stored for wanks so each time it ended up in simply deleting them.

Wait, is this board filtering the "to be honest" acronym?

Tell me which one isn't

>NN tagging seems like a dream
Not really. i2v does it, it's just that it's too slow for mass tagging right now.
Just have to wait until we get more computing power, or more efficient NN's.

Dunno. Been lurking for years but I haven't got the interest into searching the whole filter list per boards.

Wait ignore that post, I thought you said "pipe dream".

How does it match tags? Learning tags/sets and trying to apply tags by itself or just trying to match the closest tagged picture and slap the original pic's tags?

Because I like collecting images.
I just added your image to my collection.

I mean where have you been during the s m h t b h f a m era?

I don't know, ask someone who specializes in machine learning.
All I can tell you is that the neural network looks for features in the image, and outputs a confidence value for a specific tag. Look up convolutional neural networks, it's probably what i2v uses.

Just post the whole fucking thing

That's going to take ages.
Not doing.

And in case you thought the OP image is 36gb of just reisens, that's my username and /home/reisen is the base directory.
The actual directory is /home/reisen/Images

I'm not posting my entire fucking images folder, and I wouldn't be able to find the reisens that I know are in there.

Wait, so how many Reisen's do you have?

No idea.
But I'm sure there's plenty in there.

>that's my username
Why?

It's generic, and undoxable.
Good luck finding anything out about me or any of my accounts by googling "reisen" or anything similar.

And it's cute name, and character.

Have you found them yet?

I'm not even looking.

Switch to Mac

Calculate the md5 of each image and compare to some booru.

I guess u can do it easily with img-grabber.

Mean! I've been waiting on your Reisens for 3 hours!

See

oh... u are fucked then

Off yourselves dumb animetards. Immediately.

>anime
Retard. Actually kill yourself.

Anime website buddy

Sup Forums

But I never said I was going to look for them or post them!
But as a reward for waiting so long, I went through and found a reisen, just for you.

Wow, thanks user! You're the best!

Host Webserver, Open Port, Point Webserver to your image directory and post ip.
Thanks

No, I don't think I'll do that.

:(