naosher98/ball-recognizer

Sports Ball Recognition App

Sports Ball Recognizer

Project Objective:

The objective of the project is to develop an image classification model that accurately classifies 20 different types of sports balls, including Soccer Ball, Cricket Ball, Basketball Ball, Field Hockey Ball, Volleyball, Tennis Ball, Rugby Ball, Baseball, Table Tennis Ball (Ping Pong), American Football, Golf Ball, Handball, Pool Ball, Water Polo Ball, Softball, Bocce Ball, Bowling Ball, Squash Ball, Lacrosse Ball, and Wiffleball.
The model should cover the entire process, from data collection, cleaning, training, deployment, and API integration, and provide a comprehensive solution for sports ball classification. The goal is to achieve a high level of accuracy in classifying sports balls, making it useful for various applications such as ball recognition in sports events, product identification in retail, and sports equipment classification.

Data Preparation:

Dataset Preparation is a crucial step in the development of any image classification model. Great care is taken to ensure the highest quality of data for training and testing the model.

Data Collection: The dataset is collected by downloading images from DuckDuckGo using the sport ball's name as the search term.
DataLoader: To set up the DataLoader, the fastai DataBlock API is used which is a powerful and flexible library for loading and preprocessing data.
Data Augmentation: To further improve the model's performance, built-in data augmentation techniques from fastai which operate on the GPU for faster processing is utilized.
For more details on the dataset preparation process, please refer to the notebook notebooks/data_prep.ipynb which provides a step-by-step guide and in-depth explanations of the techniques used.

Model Optimization and Data Processing

Training: A resnet34 model was fine-tuned for 5 epochs, repeated 3 times resulting in an accuracy of approximately 95%.
Data Cleaning: Data cleaning was an integral and time-consuming aspect of the model development process. Since the data was collected from the internet, there were many irrelevant and noisy images. Furthermore, some images contained errors. The fastai ImageClassifierCleaner was used to clean and update the data after each training or fine-tuning iteration, except for the final iteration which was used to train the final model.

Model Deployment:

The model was deployed to the HuggingFace Spaces Gradio App, where it can be accessed and tested by users. The implementation details and code can be found in the deployment folder or by this link.

Deployment of API on GitHub Pages Website

The deployed model API has been integrated into this GitHub Pages Website, allowing users to easily access and test the model's capabilities. The implementation details, usage instructions, and other relevant information can be found in the docs folder for reference.