Sports Ball Classifier

Sports Ball Classifier is a deep learning image-classification project built for the Deep Learning course at the University of Catania. The goal was to create an end-to-end computer vision pipeline that can recognize different types of sports balls from uploaded images. The project covers the full machine learning workflow: dataset preparation, train/validation/test splitting, preprocessing, data augmentation, transfer learning, model comparison, evaluation, and deployment. We experimented with multiple convolutional neural network architectures, including ResNet18, ResNet50, and EfficientNet-B0. The final selected model was EfficientNet-B0, which achieved the best test accuracy of 84.01% on the prepared dataset. The system supports 10 sports ball categories and provides an online interface where users can upload an image and receive an instant prediction. The final model was deployed using Hugging Face Spaces with a Gradio interface, then embedded into a custom portfolio-style HTML page with a polished dashboard layout. This project demonstrates practical skills in deep learning, computer vision, transfer learning, model evaluation, frontend integration, and ML deployment. It also shows how an academic machine learning project can be transformed into a real, shareable portfolio product.