Week 8: Model Refinement & App Development Progress

This week marked a pivotal moment in the project, as I focused on enhancing the model's performance while continuing to build the mobile app. Initially, the model's results were promising, but there were a few areas I knew I could improve. To refine the model, I decided to run multiple training sessions in Google Colab, experimenting with different configurations to optimize the results. These training sessions were essential, as I adjusted parameters such as image resolution, learning rate, and batch sizes to achieve more precise and consistent output. By analyzing key performance metrics like precision, recall, and mAP (mean Average Precision), I was able to fine-tune the model’s ability to detect the ball accurately.

I also explored several augmentation techniques to further enhance the training process. Data augmentation is crucial when working with smaller datasets, as it helps improve model robustness by artificially increasing the diversity of data. I incorporated random flips, rotations, and shifts in the images to simulate various real-world scenarios where ball movement might vary.

Simultaneously, I made significant progress on the app development side. By now, the initial wireframe for the app was in place, so it was time to start turning those designs into a functional app. I worked in Android Studio to lay out the core screens, focusing on creating a user-friendly and clean interface. The goal was to ensure that the user experience would be as smooth as possible, with an intuitive flow from uploading videos, triggering the ball detection model, and displaying the results.

The integration of the ball tracking model into the app is still in progress, but at this point, I had the basic structure of the app working. It was fulfilling to see the two parts of the project; the machine learning model and the mobile app, moving forward in parallel. With each session of model training, I could see the improvements, and the app was starting to feel more interactive and real.