Uses YOLOv5 object detection to power an aimbot
You will have to generate your own models and training data. https://github.com/ultralytics/yolov5 for information on training.
Training data consisted of 500 images captured from splitgate using a simple triggerbot that captures an image every 4 seconds if there is change in reticle color. Please see my colorbot for an example of how you can easily do this in splitgate. Next the data are given two labels: head and body. The body did not include any limbs (ie arms, legs or head) to ensure that the center of the bounding box always included some part of the enemy. Only enemies were labeled. 10% of the training data were negative samples of images with just teammates that remained unlabeled to ensure that the bot did not attempt to target teammates The data was labeled using https://www.makesense.ai/ . The models were trained with 640 img size and a small batch size due to limited memory with pretrained model weights from the YOLOv5 small model and with 200 epochs
Place models in model folder and ensure that testing.py file in src reads the correct model.
use python testing.py from within the src directory to run the the aimbot. A window of the recorded area of the game screen will appear with settings you can change.
The model runs much better on Linux for some reason. Possibly due to faster screenshot system. For some games you may need to remove the Triggerbot from testing.py in order to prevent problems with clicking the mouse.