I’m a pervert and have to learn AI, so I’m trying to detect cum in videos. I’m also not perfectly sure where the perfect place is to post this, so please advise.
The goal is to detect the type of cumshot (Cum-On-Pussy, etc.) but that is difficult, so I have to take it step by step. First detect the cum only (also makes sense to quickly capture the cum frames in the video), then detect the body part/pose and finally try to calculate the cumshot type.
I’m on the first step and need more training data (if you want to help, I’ve included a labelling package below). The training data will be CC-BY-SA so others can re-use the data if needed.
The current cum detect is ok-ish, but still misses a lot of cum stains (needs more training data!) Example:
You can try it yourself. I packaged the current model and a small script that extracts the frames that contain cum with the detection boxes (like the images above):
To run the script, you need python installed and have python and pip ready in a shell. Then you can run the script like uv run scan.py <path-to-video>. It will output the result in ./detection.
If you want to help labelling (ty!) here is the package and the guide
PS: YOLO is the name of the technology, You Only Look Once.