shallow-cnn-image-classification

This repository contains a Jupyter notebook implementing the shallow CNN architecture described in “Shallow Convolutional Neural Network for Image Classification”. I demonstrate its performance on MNIST, Fashion‑MNIST, and CIFAR‑10 datasets.

---

💻 Environment & Installation

1. Clone this repo

   git clone https://github.com/AryaKoureshi/shallow-cnn-image-classification.git
   cd shallow-cnn-image-classification

2. Create a virtual environment

   python3 -m venv venv
   source venv/bin/activate     # On Windows: venv\Scripts\activate

3. Install dependencies

   pip install -r requirements.txt

---

📝 Notebook Overview

The notebook (Shallow_CNN_for_Image_Classification.ipynb) contains:

1. Data Loading & Preprocessing

   • MNIST, Fashion‑MNIST, CIFAR‑10
   • Normalization, one‑hot encoding, optional resizing

2. Model Definition

   • Shallow CNN: 2×Conv2D → MaxPooling → Flatten → Dense
   • BatchNormalization & Dropout for regularization
   • SGD optimizer

3. Training & Evaluation

   • Trained for 10–20 epochs
   • Plots of training vs. validation accuracy & loss
   • Final test accuracy on each dataset

4. Results Summary

   • MNIST: ~98% test accuracy
   • Fashion‑MNIST: ~91% test accuracy
   • CIFAR‑10: ~60% test accuracy

   (See the notebook’s final cells for precise numbers and graphs.)

---

📈 Results

The notebook produces:

• Accuracy curves for training vs. validation
• Loss curves for training vs. validation
• Bar chart comparing final test accuracies across datasets

---

🛠️ How to Run

1. Launch Jupyter:

   jupyter lab

2. Open Shallow_CNN_for_Image_Classification.ipynb.

3. Run all cells sequentially.

All figures and final metrics will appear in‑notebook.