/vt/ - Virtual Youtubers

>>6107379
By Faction
then Subfaction

>>6107437
No I mean I have a folder every Holomem
the problem would be actually placing the pictures into the folders

Nothing you can really do, unfortunately.

>>6107379
If you figure out a way, please do let us know. I have like a million unsorted images i've hoarded over the course of 15+ years that need sorting

>>6107824
Holy shit, and I thought I was fucked with 100k.

>>6107379
I know this pain.

>>6107379
1. If the collection were really big, you'd train an AI on it, and then you'd just have to sort the training data.
2. Pay people on Amazon Mechanical Turk to do it.
3. Just suck it up and do a bit per day. As long as you decrease the total number of unsorted images per day, you'll get there.

>>6107379
If you ran a Linux disto you could have tagged images booru-style

train a machine learning classifier

>>6107939
>and then you'd just have to sort the training data
You wouldn't even need that, boorus already have classified data sets.

You could train your classifier on that.

>>6107379
https://www.tensorflow.org/tutorials/images/cnn

>>6107437
Based

>>6107824
>hoarded over the course of 15+ years
悲しい…

>>6107379
Whatever you do, use a tagged system instead of folders.

>>6107939
>2. Pay people on Amazon Mechanical Turk to do it.
Half my images folder is loli hentai so this isn't an option I don't think.

>>6108031
I have absolutely no idea what this means.

>>6108161
It's a machine learning library to program neural networks, which can (among other things) classify images into categories.

>>6108141
can you elaborate a bit more as to why?

>>6108220
Because the more shit you look at, the more you realize it never fits cleanly into one folder.

You'll have an image with multiple people in it, or an image with multiple fetishes. Tagging is the only way to properly deal with that.

>>6108270
Fair enough

>>6107379
If someone ever trains a classifier for holomem, please share.

best thing i managed was hydrus with hydrus-dd
https://hydrusnetwork.github.io/hydrus/
https://gitgud.io/koto/hydrus-dd
it identifies and it misses some but it's better than not having anything, dont have to train anything either

>>6108031
You first need to teach the AI the difference between Kiara and Gura to make it work which you need around 500-1,000 to be feasible.

Or just write a program that can identify the image's accent color and group the images accordingly.

>>6107379
Look into unsupervised image classification. This seems pretty doable, and would be a really interesting project. I might take a crack at it, but I don't have a large amount of fanart.

>>6111867
No need for that, boorus give him training sets

>>6107379
Pray to the God Emperor

i bet half of them are ogey

>>6109074
Thanks anon, I'll try my best to make it work

You start today, every new pic you place in a proper folder. Put all the shit you currently have in "other", it's beyond saving.

>>6107379
Create a folder called "Pending" and put everything in it

>>6107437
/thread/

>>6109074
Does DD create its own tag pool or is there a chance of it messing up "My Tags"

>>6107488
You do it little by little everyday. Like ten minutes or so. Eventually you'll get through it.

>>6112495
This

>>6107379
upload it to drive and let anon sort it for you

>>6107379
Try writing a script using DeepDanbooru for tagging and sort it out

>>6107379
Shift+delete the entire folder
In 15 minutes you'll realize nothing of value was lost

>>6107379
Pay someone else to do it for you

>>6109074
>hydrus dd
Relevant to my interests

I've been using hydrus for years but there's some stuff I haven't imported