CNN Image Classifier 🖼️

A modular, end-to-end Convolutional Neural Network (CNN) for image classification built with TensorFlow/Keras.
Designed for researchers, students, and developers to train, evaluate, and deploy image classifiers efficiently.

---

📑 Table of Contents

• 🎯 Project Overview
• 🖼️ Project Thumbnail
• 🏗️ Model Architecture
  • Layer-by-Layer Description
  • Model Compilation
  • Summary Table
• 📈 Accuracy & Loss Over Epochs
• ⚙️ Features
• 📁 Project Structure
• 🧰 Getting Started
  • Prerequisites
  • Installation
• 🚀 Quick Start
  • Training the Model
  • Using the Notebook
• 📦 Dataset Sample
• 🔧 Customization
• 🛠️ Configuration
• 📊 Results & Performance
• 🤝 Contributing
• 📜 License
• 🙏 Acknowledgements
• 👤 Author

---

🎯 Project Overview

This repository provides a complete CNN workflow for image classification:

• Load and preprocess images.
• Build a customizable CNN architecture.
• Train, validate, and evaluate the model.
• Save and use the model for predictions on new images.
• Visualize performance metrics and training curves.

It’s designed to be modular, readable, and easy to extend for custom datasets or advanced architectures.

---

📸 Project Thumbnail

Visual representation of the CNN Image Classifier workflow.

---

🏗️ Model Architecture

The CNN model is designed for binary image classification with a simple yet effective architecture. The model is implemented using TensorFlow/Keras.

Layer-by-Layer Description

1. Input Layer

   • Accepts images with shape (height, width, channels) specified by input_shape.

2. Convolutional Block 1

   • Conv2D with 32 filters, kernel size 3x3, activation ReLU.
   • MaxPooling2D with pool size 2x2 to reduce spatial dimensions.

3. Convolutional Block 2

   • Conv2D with 32 filters, kernel size 3x3, activation ReLU.
   • MaxPooling2D with pool size 2x2.

4. Flatten Layer

   • Converts the 2D feature maps into a 1D vector.

5. Fully Connected (Dense) Layer

   • Dense layer with 128 neurons and ReLU activation.

6. Output Layer

   • Dense layer with num_classes neurons and sigmoid activation (binary classification).

Model Compilation

• Optimizer: Adam
• Loss Function: Binary Crossentropy
• Metrics: Accuracy

Summary Table

Layer Type	Output Shape	Parameters
Conv2D (32 filters)	(62, 62, 32)	896
MaxPooling2D (2x2)	(31, 31, 32)	0
Conv2D (32 filters)	(29, 29,32)	9248
MaxPooling2D (2x2)	(14, 14, 32)	0
Flatten	6,272	0
Dense (128 neurons)	128	802,944
Output Dense	1	129

---

📊 Accuracy & Loss Over Epochs

Visualizations generated during training provide insights into model performance:

Training Accuracy vs Epochs

Training & Validation Loss vs Epochs

---

⚡ Features

• Modular CNN implementation (src/model.py)
• Data preprocessing & augmentation utilities (src/data.py)
• Notebook demonstrating full workflow (notebook/cnn_image_classifier_main.ipynb)
• Model saving and loading for inference
• Clear visualizations of metrics and predictions

---

📁 Project Structure

Path	Description
assets/	Thumbnail and performance plots
├── cnn_image_classifier_thumbnail.png	Project thumbnail image
├── accuracy_over_epochs.png	Accuracy plot over training
└── loss_over_epochs.png	Loss plot over training
dataset/	Sample dataset (cats & dogs)
├── training_set/	Training images
└── test_set/	Testing images
notebook/	Main Jupyter notebook
└── cnn_image_classifier_main.ipynb	Contains training and evaluation
src/	Modular code components
└── __init__.py	Marks src/ as a Python package
├── model.py	CNN architecture
└── data.py	Data loading and preprocessing
saved_model/	(Optional) Saved trained model
└── cnn_model.h5	Saved model file
requirements.txt	Python dependencies for reproducibility
.gitignore	Git ignore rules
LICENSE	Project license
README.md	Project documentation

---

🛠️ Getting Started

Prerequisites

• Python ≥ 3.7
• TensorFlow 2.x
• Numpy, Matplotlib, Pandas
• (Optional) GPU for accelerated training

Installation

git clone https://github.com/ArianJr/cnn-image-classifier.git
cd cnn-image-classifier
python -m venv venv
source venv/bin/activate      # Windows: venv\Scripts\activate
pip install -r requirements.txt

---

🚀 Quick Start

Training the Model

# Example: run in the notebook or a script
from src.data import load_training_data, load_test_data
from src.model import build_cnn

training = load_training_data('dataset/training_set', (64, 64))
test = load_test_data('dataset/training_set', (64, 64))
model = build_cnn(input_shape=(64,64,3), num_classes=1)
history = model.fit(x=training, validation_data=test, epochs=25)

Using the Notebook

Launch the Jupyter notebook for an interactive demo:

jupyter notebook notebook/cnn_image_classifier_main.ipynb

---

📦 Dataset Sample

This repository includes a sample subset of the Dogs vs. Cats dataset for demonstration purposes:

• train/: 200 images (100 cats, 100 dogs)
• test/: 50 images (25 cats, 25 dogs)

This subset is ideal for:

• Testing the modular CNN pipeline
• Validating config-driven training and evaluation
• Keeping the repository lightweight

For full-scale training, download the complete dataset from Kaggle.

---

🔧 Customization

• Replace or augment dataset (dataset/ folder).
• Modify CNN architecture (src/model.py).
• Experiment with hyperparameters (learning rate, batch size, optimizer).
• Implement transfer learning with pretrained networks.
• Add evaluation metrics: confusion matrix, precision, recall, F1-score.

---

⚙️ Configuration

This project is designed to be easily customizable. You can modify key training parameters directly in the code files:

Parameter	Location	Default Value
input_shape	model.py	(64, 64, 3)
epochs	cnn_image_classifier.ipynb	25
batch_size	cnn_image_classifier.ipynb	32
train_dir	data.py	dataset/training_set
test_dir	data.py	dataset/test_set
model_path	cnn_image_classifier.ipynb	saved_model/cnn_model.h5

These values can be adjusted to suit your dataset size, image resolution, or training goals.

---

📈 Results & Performance

Metric	Value
Training Accuracy	88.21%
Validation Accuracy	79.10%
Final Training Loss	0.2653
Final Validation Loss	0.5460

---

🤝 Contributing

1. Fork the repository.
2. Create a feature branch (git checkout -b feature/YourFeature).
3. Commit changes with clear messages.
4. Open a pull request.

---

📜 License

MIT License – see the LICENSE file for details.

---

🙏 Acknowledgements

• TensorFlow & Keras communities for excellent tools.
• Open-source datasets used for training.
• Tutorials and guides on CNNs that inspired this workflow.

---

👤 Author

Arian Jr
📧 Contact Me • 🌐 GitHub Profile

---

Made with ❤️ by ArianJr

_{⭐ If you found this project useful, please consider giving it a star! It helps others discover it and supports my work.}

---