NSFW image tagging via Deep Learning [Need Your Help]

[ski23]

I'm working on a program that uses deep learning to tag images with many tags of roughly 300 possible tags.


Why should I care. What does it do for me?
1. The first big benefit of this tool is to Tease AI and Tease AI Java users. Labeled images are essential to a good experience with either of these applications and currently it is very time consuming to hand label images and then there is the negative that the user will already have seen the images before when they labeled them which can break immersion.
2. I think most people would agree that the biggest problem with porn content is that there is a lot of content out there and it can be very hard to find what you are looking for. I'm sure I'm not alone in searching over an hour to find a certain type of video. The biggest impact on this tool that can help nearly anyone is that this tool will allow you to find porn videos or images customized to your tastes. You will be able to tell the program what type of content you like and what you dislike and it will be able to search the web and download content customized to you.

I'm interested. What can I do to help?
I've made a list of nearly every general image tag I can think of. I'm going to be using this list in creating the new model that will be able to detect all of these tags in images. I need as many people as possible to look at this list and suggest additions that they think need to be added (or none if they think its good now). I want the list to be comprehensive but it can't be anything extremely specific otherwise it will be too hard to get content for to train the model. Here is the list. I've divided all of the labels up into categories to make it easier to view them:

possibletags.xlsx

(11.11 KiB) Downloaded 295 times

[CXZman]

Seems quite complete to me.
Could add in Describing Body :
- Angry
- Blindfolded (unless you want to keep it in clothing)
- Kneeling
- Laying
- Tied up/Restrained
- Shocked (attitude)
- Sleeping
- Sweaty

[ski23]

Seems quite complete to me.
Could add in Describing Body :
- Angry
- Blindfolded (unless you want to keep it in clothing)
- Kneeling
- Laying
- Tied up/Restrained
- Shocked (attitude)
- Sleeping
- Sweaty

Added several of those. Thanks!

[foxyfred]

I would highly recommend you swing by the CTI community, they have a censoring system that has quite a large set of data that you might be able to ask if you could make a fork of to adjust for your needs. It already works very well and has an insane amount of tagged data they built up over at least 6 months.
Here's a link to their description of their system: http://www.cti-community.net/nsfwproj/nsfw_tut.htm

[ski23]

I would highly recommend you swing by the CTI community, they have a censoring system that has quite a large set of data that you might be able to ask if you could make a fork of to adjust for your needs. It already works very well and has an insane amount of tagged data they built up over at least 6 months.
Here's a link to their description of their system: http://www.cti-community.net/nsfwproj/nsfw_tut.htm

I know of that community. I built their current model lol.

[foxyfred]

I would highly recommend you swing by the CTI community, they have a censoring system that has quite a large set of data that you might be able to ask if you could make a fork of to adjust for your needs. It already works very well and has an insane amount of tagged data they built up over at least 6 months.
Here's a link to their description of their system: http://www.cti-community.net/nsfwproj/nsfw_tut.htm

I know of that community. I built their current model lol.

I was wondering if that was you in their discord...

[ski23]

I would highly recommend you swing by the CTI community, they have a censoring system that has quite a large set of data that you might be able to ask if you could make a fork of to adjust for your needs. It already works very well and has an insane amount of tagged data they built up over at least 6 months.
Here's a link to their description of their system: http://www.cti-community.net/nsfwproj/nsfw_tut.htm

I know of that community. I built their current model lol.

I was wondering if that was you in their discord...

Yea I've been bouncing back and forth between there and Tease AI Java a lot lately.

[lockedkeyless]

Out of curiosity, has this been finished?

[ski23]

Out of curiosity, has this been finished?

Havent been working on it a lot recently. This stuff is really tricky to debug and troubleshoot since there isnt much of anyone doing anything similar.

[lockedkeyless]

Out of curiosity, has this been finished?

Havent been working on it a lot recently. This stuff is really tricky to debug and troubleshoot since there isnt much of anyone doing anything similar.

I don't know much about this however I was going to suggest if it wasn't your plan for it to cover the basics (it can always update more later), and maybe autofill what it detects allowing the user to confirm it is correct. This solution may make it more work than you initially planned, however would help the process and reduce consequence of errors.
If there was any way I can help I will.

[ski23]

Out of curiosity, has this been finished?

Havent been working on it a lot recently. This stuff is really tricky to debug and troubleshoot since there isnt much of anyone doing anything similar.

I don't know much about this however I was going to suggest if it wasn't your plan for it to cover the basics (it can always update more later), and maybe autofill what it detects allowing the user to confirm it is correct. This solution may make it more work than you initially planned, however would help the process and reduce consequence of errors.
If there was any way I can help I will.

The real issue lies in the fact that for this work id need to make a large dataset of images that are tagged perfectly for the model to train on. I have part of that but the issue is its very time consuming and i don't have time to go through and tag thousands of images at the moment.

[lockedkeyless]

Havent been working on it a lot recently. This stuff is really tricky to debug and troubleshoot since there isnt much of anyone doing anything similar.

I don't know much about this however I was going to suggest if it wasn't your plan for it to cover the basics (it can always update more later), and maybe autofill what it detects allowing the user to confirm it is correct. This solution may make it more work than you initially planned, however would help the process and reduce consequence of errors.
If there was any way I can help I will.

The real issue lies in the fact that for this work id need to make a large dataset of images that are tagged perfectly for the model to train on. I have part of that but the issue is its very time consuming and i don't have time to go through and tag thousands of images at the moment.

Are the current available databases of images not any use?

[ski23]

I don't know much about this however I was going to suggest if it wasn't your plan for it to cover the basics (it can always update more later), and maybe autofill what it detects allowing the user to confirm it is correct. This solution may make it more work than you initially planned, however would help the process and reduce consequence of errors.
If there was any way I can help I will.

The real issue lies in the fact that for this work id need to make a large dataset of images that are tagged perfectly for the model to train on. I have part of that but the issue is its very time consuming and i don't have time to go through and tag thousands of images at the moment.

Are the current available databases of images not any use?

No, unfortunately they aren't very useful because they are all tagged differently from each other and different people use different labels.

[Spielers]

how is the status is there a program release

[ski23]

how is the status is there a program release

Slow progress due to other projects so no scheduled release at this time.