NSFW Neural Network Project

Identify:

· Faces and naughty body parts (e.g. Breasts, Genitalia,…) based on the gender.

· Create one network for (real) Porn (Dataset 1)

· Create one network for Hentai (Dataset 2)

Goal

· 5000 images for each dataset

To easily create training data we created a tool which uses a third party API, this API is already able to identify naughty regions but is limited to 1000 requests per month and user (+requires and internet connection). Once we have a working network we will replace the API with our own network and put the tool online.

We aim to then have the following advantages

· Free to use without limits

· Works offline

· Good results

· Privacy (no 3th party involved)

How to contribute & first steps:

1. Download the beta tool at our Discord Server: (talk to 0131 or coda):

2. Start the tool (requires you to have Java installed)

3. Register at https://deepai.org/ (it is free) and generate your own API key. NOT REQUIRED ANYMORE WE GOT OUR OWN NETWORK YAY

4. Copy that key into the tool (also better save it in a text file for next time). NOT REQUIRED ANYMORE WE GOT OUR OWN NETWORK YAY

5. Select files as input.
Please use either Hentai or real Porn.
If you for some reason mixed up a few files, use the Ignore image checkbox on a wrong, non-lewd or bad image.

6. Use the Start button
The tool will now start to detect naughty parts and faces

7. As we can see for our example it did already do a good job but we have a false positive here.
It detected the arm as Female Breast which is obviously wrong.
Right Click -> Delete or use a Double Click on the box and it will be ignored.

8. Let’s try the next image, use the arrow button to switch the to the next image:

Oh well this time it fucked up big, pretty much everything is wrong.
We have to invest a bit more time here, if you think it is not worth it then use the Ignore image checkbox.
If not you can do the following:

1. Move the box to the right place (Drag and Drop not right click)

2. Change the type if wasn’t classified right (Right Click)

3. Make the box bigger or smaller (Scroll Wheel)

4. Delete the box if it is just wrong (Right Click)

5. Create a new box:

If you want to add a new box since there is something missing then just click on a free part of the image and drag the mouse to create a new box of the size you want (Don’t forget to change its type).

Above you can see the corrected result.
As you might have noticed the age and race might be still wrong but that is ok. This data is not used and it is not important if they are correct.

Now the last step!

Use the Save button to generate the files and please send us your generated data:

The files we need can be found inside the ‘source’ folder which is placed inside the ‘output’ folder.
If not changed this output folder appears in the folder with the application.

After sending us a few examples (from the identified folder) we will invite you to our Dropbox folder, where you can upload your data.

You can reach us at:

· https://discord.gg/YkRAjJY

· http://www.cti-community.net/forum/