monaabdelrazek/ancient-egypt-cnn

🧠 Custom CNN for classifying Ancient Egyptian artifacts — 96% accuracy, built from scratch with live API deployment. # 🧠 Image Classification from Scratch using CNN This deep learning project demonstrates the end-to-end development of an image classification system using a **custom Convolutional Neural Network (CNN)** architecture, built entirely **from scratch** — without relying on pre-trained models. > 📌 **Key Achievements:** > ✅ Built CNN model from zero (no transfer learning) > ✅ Achieved **96% test accuracy** on unseen data > ✅ Designed a robust pipeline including cleaning, augmentation, training, evaluation, and deployment > ✅ Deployed on Hugging Face with live API access --- ## 📁 Project Contents | Notebook | Description | |-------------------------|----------------------------------------------------------| | `01_data_analysis.ipynb` | Initial EDA, class distribution analysis, imbalance insights | | `02_data_augmentation_static.ipynb` | Data augmentation techniques to address class imbalance | | `03_model_attempt1_88acc.ipynb` | First baseline CNN model with 88% accuracy | | `04_model_final_96acc.ipynb` | Final refined CNN model with 96% accuracy | | `class_info.json`| Contains general descriptive information about each class, used to display names during testing and inference | | `deployment/` | Scripts and links for online inference API | | `example_request.ipynb` | Upload an image and send it to the deployed Hugging Face API for testing | > 🗒️ **Note:** The notebooks `01_data_analysis.ipynb` and `02_data_augmentation_static.ipynb` contain some printed messages in Arabic. This **does not affect any code functionality** or the results. --- ## 🧼 Dataset Preparation & Cleaning The dataset was originally based on publicly available resources, including: - Egypt Monuments Dataset on GitHub Additionally, more images were **manually collected** from various online websites to enrich and balance the dataset. - Broken/corrupted images were dete …