Sports Ball Classifier Dashboard

Project Overview

This project classifies images of sports balls into 10 categories using a deep learning model. We trained multiple CNN architectures (ResNet18, ResNet50, EfficientNet-B0) and selected the best performer. The final model is deployed online so anyone can upload an image and get a prediction instantly.

Best Model

EfficientNet-B0

Classes

Best Test Accuracy

84.01%

Team Members

Rohma AnwarTeam Member

Muhammad Salman SabirTeam Member

Abid AliTeam Member

Prepared under the supervision of

Prof. Sebastiano Battiato, Ph.D.

Deputy Head of Department
Department of Mathematics and Computer Science
University of Catania

Home Page LinkedIn

Checking demo status…

What we built

ML Pipeline

1) Dataset Preparation Folder-based dataset → DataFrame → train/val/test split (80/10/10).
2) Preprocessing + Augmentation Resize, flip, rotation, color jitter, normalization to improve generalization.
3) Transfer Learning Pretrained CNN backbone + replace final classifier for 10 ball classes.
4) Evaluation Accuracy, precision/recall/F1, confusion matrix to analyze per-class performance.
5) Deployment Model hosted online using Hugging Face Spaces + Gradio interface.

Live Demo (Upload an image)

Open in new tab