How far should we get with computer vision and data labeling tools in VR porn?

General
1

We have some really good progress with computer vision and data labeling. We can identify every aspect of what’s happening in the video. Maybe even reconstructing 6DOF to a decent degree with algorithmic tools. Depth maps autofocus will make videos look much better, especially the older ones. Probably machine learning upscale up to 8K.

We should start releasing interactive sex toy scripts and censor videos for Japanese market created with these tools really soon.

3 Likes
2

Your image segmentation algorithm certainly does seem to be working nicely in that example. Since you are able to resolve so much more information, in what format(s) will you be releasing scripts “really soon”?

Most interactive toys currently have only one controllable degree of freedom. Even those that have two controllable elements usually are usually only trying to sync one with the single DOF listed in funscripts (or other script formats). Will you be releasing script files with multiple degrees of freedom. Do you have multiple approaches to compressing the multi-DOF data into a single parameter (which one might want to do very differently to, e.g. control a toy corresponding to particular scene elements.

Anyway, nice work so far!

1 Like
3

These will be the same .funscripts.

I can only think of OSR2 with more than 2 dimensions and it’s already supported by @raser1. We should get it at some point, shouldn’t be a problem there

4

Now the next step would be to decensor those japanese videos with KI :wink:

4 Likes