Pytorch must be installed seperate from requirements.txt due to different cuda versions. Go to pytorch.org to install your version.
Tested and developed on Windows 11 with Python 3.10.5, CUDA 12.2, and Pytorch CUDA 12.1 Nightly Build.
Press and hold 'q' on your keyboard to stop the main loop.
Fruit Ninja was emulated using BlueStacksX. YOLO-V8 Nano was then used to detect fruits and bombs. Using this information, a path to slice through the fruits while avoiding bombs was found and executed. The model used in the main program is a TensorRT model for fast inference. The the capture screen cordinates are based on a BlueStacks window maximized (not fullscreen) with the right BlueStacks sidebar minimized on a 1920x1080p monitor. Because of this, the screen dimensions in main.py may have to be tested and changed to work with your monitor
YoloV8Testing.py or view_data.py can be used to test inference and if your GPU is being detected. The code for training a model is also included, however the custom FruitNinja dataset is not included due to its size but can be downloaded from Roboflow.