ImageClassification_NN

A Convolutional Neural Network (CNN) built with TensorFlow / Keras that classifies images into categories. This project covers the full ML workflow — from raw data cleaning and preprocessing, through model training and evaluation, to inference on new images.

Trained on a cats vs dogs dataset with results tracked via TensorBoard and model checkpointing.

---

📊 Results

Metric	Score
Training Accuracy	~95%
Validation Accuracy	~92%
Test Precision	logged via model_precision
Test Recall	logged via model_recall
Test Accuracy	logged via model_accuracy

Training loss and accuracy curves are plotted automatically at the end of training using Matplotlib.

---

📦 Project Structure

ImageClassification_NN/
├── models/                      # Folder where trained models are saved
├── data/                        # Dataset directory — images organized into class folders
├── test_data/                   # Sample images for prediction/testing
├── Train_ImageClassifier.py     # Full training pipeline
├── Predict_ImageClassifier.py   # Loads saved model and runs inference
├── README.md                    # This file
├── pyproject.toml               # Python project config
└── .gitignore

---

🧠 How It Works

The training pipeline (Train_ImageClassifier.py):

1. Scans and cleans the dataset — removes corrupt or unsupported images
2. Loads images into batches using tf.keras.utils.image_dataset_from_directory
3. Normalizes pixel values to [0, 1]
4. Splits data into 70% train / 20% validation / 10% test
5. Builds and trains a CNN with 3 convolutional blocks
6. Uses callbacks: TensorBoard logging, model checkpointing (best model only), early stopping
7. Plots training/validation loss and accuracy curves
8. Evaluates precision, recall, and accuracy on the test set
9. Saves the final model to ./models/imageclassifier.keras

The inference script (Predict_ImageClassifier.py):

• Loads the saved model
• Preprocesses a new image (resize to 256x256, normalize)
• Outputs the predicted class with a matplotlib visualization

---

🏗️ Model Architecture

Input (256x256x3)
→ Conv2D(16, 3x3, relu) → MaxPooling
→ Conv2D(32, 3x3, relu) → MaxPooling
→ Conv2D(16, 3x3, relu) → MaxPooling
→ Flatten
→ Dense(256, relu)
→ Dense(1, sigmoid)  ← binary output

Compiled with:

• Optimizer: Adam
• Loss: BinaryCrossentropy
• Metrics: Accuracy

---

🛠️ Technologies Used

• Python 3
• TensorFlow / Keras
• OpenCV — image reading and preprocessing
• NumPy
• Matplotlib — training visualization
• Pillow — image format validation

---

🚀 Getting Started

1. Clone the Repository

git clone https://github.com/maherhms/ImageClassification_NN.git
cd ImageClassification_NN

2. Install Dependencies

python -m venv venv
source venv/bin/activate      # macOS / Linux
venv\Scripts\activate         # Windows

pip install tensorflow opencv-python matplotlib pillow

---

📁 Prepare Your Dataset

Organize training images by class:

data/
├── cat/
│   ├── img1.jpg
│   └── ...
├── dog/
│   ├── img1.jpg
│   └── ...

Place test images in test_data/.

---

📈 Train the Model

python Train_ImageClassifier.py

This will clean the dataset, train the CNN, save the best model, and plot performance curves automatically.

---

🔍 Run Inference

python Predict_ImageClassifier.py

Modify the test_image_path variable inside the script to point to any image you want to classify.

---

📡 Monitor Training with TensorBoard

tensorboard --logdir logs

Open the URL shown in the terminal to view live loss and accuracy curves during or after training.

---

🔧 Possible Improvements

• Multi-class classification support
• Data augmentation (flipping, rotation, zoom)
• Transfer learning with pretrained models (MobileNet, ResNet, EfficientNet)
• Confusion matrix and per-class metrics
• Simple web interface for drag-and-drop prediction

---

📄 License

Open for learning and modification. See LICENSE for details.