This project uses Darknet to blur the tattoos of people in photographs and videos.
It will be less later because there is not even executable code yet 🤔
The first thing we need is data about a person's tattoo. Data on this is currently available on Instagram and Pinterest. Then labeling with Yolo_mark.
In the labeling process, tattoos are divided into human arms, legs, chest, abdomen and neck. The labeling was done with the exception of a few photos that I think are too many and difficult to distinguish. Then you get weights through Colab.
Some error weights are detected through openCV + Python code. openCV extracts the tattoo and blurs the pixels. After that, the final extraction project through synthesis.
This project is still in progress! So it may not be perfect yet.
This document will continue to be updated :)
x64 is a folder that mainly contains data about AI in Yolo_mark.
.
└── Release
├── data
│ ├── img
│ │ ├── {SKIP - SO MANY FILES ON THIS DIRECTORY...}
│ ├── obj.data
│ ├── obj.names
│ └── train.txt
├── train_obj.cmd
├── yolo-obj.cfg
└── yolo_mark.cmd
3 directories, 625 files
The data was made, with Yolo_mark where he wanted to judge the authenticity of the label people. , Install Yolo_mark to see the labeling deleted x64 from Yolo_mark. Put it in the x64 folder.
Since it has not been released yet, we will post a download link after all Colab weight calculations have been completed.
The reason why our project was suspended is very simple. This is why artificial intelligence did not provide adequate data for learning. It's too difficult. I had a lot of data to study, but I didn't have time to label it. It also used Google's Colab because it didn't have such a good computer. Colab was so screwed up because runtime initialization paid off and his work blew away.
If you wanna ask something then add issue on this reporisty. And also all images from Pinterest.