Object Detection in Red Dead Redemption 2 using YOLOv8

This project made by Serhat Güler implements object detection in the game Red Dead Redemption 2 (short: RDR2) from Rockstar Games. In this project, a custom dataset specially for RDR2 has been made, so YOLO can detect more obstacles in the virtual world.

Dataset on roboflow universe

Gather your own version at roboflow universie to test this project: https://universe.roboflow.com/yolov8-zpkhg/rdr2-object-detection

Prerequisites and Installation

1. This project requires python to be installed on your computer. Get python here. After installing python, the command python's package manager pip comes alongside with python.
2. Run this command to download all required packages: pip install -r requirements.txt
3. Clone my roboflow universe workspace and generate yourself an API Key to generate yourself a version
4. Use the arguments to enter your credentinals into the python script
5. Start project by using py rdr2_obj_detection.py (and supply arguments if necessary)

With those previous two steps, you have installed everything, which is required to run this project.

Usage

This project can be started by using simply py rdr2_obj_detection.py. If py is not found as a command on your system, simply use python. It's the same.

Customization

To start this project with different behavior, here is a list of all arguments available.

Argument	Type	Default Value	Description
--train-only	Boolean	False	Specifies the used dataset version
--epochs	Integer	250	Ability to supply your own API key without leaving it permanently in file
--dataset-version	Integer	3	Specifies the used dataset version
--rf-api-key	String	None	Ability to supply your own API key without leaving it permanently in file
--rf-workspace	String	None	Ability to supply your workspace name
--rf-project	String	None	Ability to supply your project name
--fps	Integer	60	Frames per second
--interval	Integer	25	Interval between frames in milliseconds
--model	String	best.pt	Path to the YOLO model
--ignore-game-not-active	Boolean	False	Skip the verification if the game window is active or not

Note that you only need to use initally --rf-api-key, --rfworkspace and --rf-project, so the script can download simply the desired dataset version for you from roboflow. You can also simply download it in YOLOv8 format and place the files in dataset/RDR2-Object-Detection-<YOUR_DESIRED_VERSION>. Create the subfolders, if you don't have them already. <YOUR_DESIRED_VERSION> must match the version youre using in the argument, if you dont want to change something just name it simply "3".

Windows Troubleshooting

When the model won't start, check if in your files the absolute path is being used and not the relative. This fixed many issues for me.

Author

This project has been made for the lecture Applied Machine Learning with the lecturer Mr. Prof. Dr. Sebastian Leuoth at Hof University by Serhat Güler (matriculation number: 00342920).