Application
Rocket League Object Detection
The project involved creating an image dataset by capturing screenshots from a specific window using Python libraries like numpy, win32gui, and PIL, then labeling the dataset using Make Sense AI for manual annotations. After organizing and labeling the images, the dataset was used to train a YOLOv4-tiny object detection model on Google Colab, leveraging pre-trained weights and a custom configuration. Post-training, the model was tested in real-time on a live window of the "Rocket League" game, detecting objects like the ball and boost. While the model showed reasonable accuracy, particularly with static or slow-moving objects, it struggled with fast-moving scenes and varied lighting conditions. The project demonstrated the practical implementation of object detection, including dataset preparation, model training, and real-time application, highlighting challenges in dynamic environments and emphasizing the iterative nature of machine learning development.