Animal Detection Project

Overview 🐾🎨🔗

This project aims to detect and classify different animal species using deep learning. Leveraging the ResNet50 architecture, the model is fine-tuned on a dataset of animal images to achieve high accuracy in classification. A Streamlit web application is also included to allow users to upload images and receive predictions in real-time.

Features 🚀🌟🔧

• Deep Learning Model: Utilizes a pre-trained ResNet50 model fine-tuned for animal classification.
• Data Augmentation: Enhances the dataset with various transformations for robust training.
• Streamlit Web App: Provides an intuitive interface for users to interact with the model.
• GPU Support: Optimized for GPU training to speed up computations.

Project Structure 📁🌍🔗

• main.py: Contains the training pipeline for the model.
• streamlit_app.py: Hosts the Streamlit web application for real-time predictions.
• DL_project.keras: The trained model file.
• class_names.npy: File containing the class labels.

Getting Started 🔄🛠️💲

Prerequisites 📊📝🔧

Ensure you have the following libraries installed:

• TensorFlow
• NumPy
• Matplotlib
• PIL (Pillow)
• Streamlit
• patool
• shutil

You can install these dependencies using the following command:

pip install tensorflow numpy matplotlib pillow streamlit patool

Dataset Preparation 🔐📚🌐

1. Place your dataset in a .rar file with the folder structure:

2. Update the path to your .rar file in the train_model function of main.py.

Training the Model 🎓🔢🔄

Run the main.py script to train the model:

python main.py

Streamlit Web Application 📲🌐🌄

To run the web application:

1. Place the trained model (DL_project.keras) and class labels file (class_names.npy) in the project directory.
2. Update the MODEL_PATH and CLASS_NAMES_PATH variables in streamlit_app.py to reflect the correct paths.
3. Start the Streamlit app:

streamlit run streamlit_app.py

4. Access the app in your browser at http://localhost:8501/.

Usage 🌍🔦🔧

• Training: Customize the train_model function in main.py to adapt to different datasets.
• Web App: Upload an image through the Streamlit interface to classify it and view the predicted class with a confidence score.

Team Members 👤🌈🔗

• Ahmed Soudy Tawfik Ahmed
• Mustafa Gaser Mekhemar
• Elsayed Osama Elsayed
• Mahmoud Foad Sleem
• Islam Ragab Ahmed
• Ahmed Reda Farag

Results 🏆🎨🔠

• Model Accuracy: Achieved a high validation accuracy through fine-tuning.
• Loss Metrics: Demonstrated convergence with minimal overfitting as visualized in the training history plots.

Future Work 📊🌐🔧

• Expanding the dataset for broader classification.
• Integrating additional architectures for comparative analysis.
• Deploying the app on cloud platforms for wider accessibility.