About the project

This project was made just for the fun of it. I've always been interested in cheats for video games (only how they work, not personally using them), so when I found a object-detection dataset on Counter-Strike players, I just knew that I had to make a model that can detect them. I started the project by training a YOLOv5 model on two datasets, first an old CSGO dataset, and then a newer CS2 dataset. Then I evaluated the model and saw that it was doing fine, so I then built a program that would do both live inference using the trained model, and have an overlay that would draw the boxes that the model was predicting. Since I'm using Linux and Wayland, I build an overlay using the GTK API, but this is really simple to replace with one that works for Windows or other Linux compositors.

Tools used

The tools that were used for this project was, YOLO from Ultralytics to create and train the object detection model, Grim & OpenCV to capture the screen and feed the frames to the model, and then the GTK4 API in order to create the overlay for the program.