CNN Cats and Dogs Classifier

A deep learning image classification project that uses a Convolutional Neural Network (CNN) to distinguish between cats and dogs. Built with TensorFlow and Keras, this model demonstrates fundamental computer vision techniques for binary image classification.

🎯 Project Overview

This project implements a CNN-based image classifier trained on a dataset of cat and dog images. The model achieves high accuracy by employing modern deep learning techniques including batch normalization, dropout regularization, and early stopping.

🏗️ Model Architecture

The CNN model consists of:

• 4 Convolutional Blocks, each containing:

  • Conv2D layer (32 → 64 → 128 → 256 filters)
  • Batch Normalization
  • MaxPooling2D (2×2)
  • Dropout (0.25)

• Fully Connected Layers:

  • Flatten layer
  • Dense layer (512 neurons, ReLU activation)
  • Dropout (0.5)
  • Output layer (2 neurons, Softmax activation)

Total Parameters: Progressive feature extraction from 32 to 256 filters across layers

🔧 Technical Features

• Image Preprocessing: Grayscale conversion and resizing to 50×50 pixels
• Data Augmentation: Shuffled training data for better generalization
• Regularization Techniques:
  • Dropout layers to prevent overfitting
  • Batch normalization for stable training
• Early Stopping: Custom callback stops training at 95% validation accuracy
• Optimization: Adam optimizer with categorical cross-entropy loss
• Batch Processing: 32-image batches with TensorFlow Dataset API

📊 Training Configuration

• Image Size: 50×50 pixels (grayscale)
• Batch Size: 32
• Max Epochs: 50
• Early Stopping Threshold: 95% validation accuracy
• Training/Validation Split: 12,500 images each

🚀 Getting Started

Prerequisites

pip install tensorflow opencv-python numpy pandas tqdm

Dataset Structure

Organize your dataset as follows:

train/
  ├── cat.0.jpg
  ├── cat.1.jpg
  ├── dog.0.jpg
  ├── dog.1.jpg
  └── ...
test1/
  ├── 1.jpg
  ├── 2.jpg
  └── ...

Configuration

Update the directory paths in code.py:

TRAIN_DIR = 'path/to/your/train/folder'
TEST_DIR = 'path/to/your/test/folder'

Training the Model

Run the training script:

python code.py

The model will:

1. Load and preprocess images
2. Train on the dataset with progress bars (tqdm)
3. Stop early if 95% validation accuracy is reached
4. Save the trained model as dog_cat_classifier.h5

📁 Project Files

• code.py - Main training script with model architecture and data processing
• dog_cat_classifier.h5 - Saved trained model (generated after training)

🧠 Key Functions

• label_img(img) - Extracts labels from filenames using one-hot encoding
• create_train_data() - Loads, processes, and labels training images
• process_test_data() - Processes test images without labels
• EarlyStoppingAtValAcc - Custom callback for early stopping at target accuracy

📈 Model Performance

• Target validation accuracy: 95%
• Early stopping prevents overfitting
• Model summary available after training

🛠️ Technologies Used

• TensorFlow/Keras - Deep learning framework
• OpenCV (cv2) - Image processing
• NumPy - Numerical computations
• Pandas - Data manipulation
• tqdm - Progress visualization

📝 Notes

• Images are converted to grayscale to reduce computational complexity
• The model uses categorical cross-entropy for binary classification
• Dataset paths are configured for local Windows environment (update as needed)
• Inspired by GeeksForGeeks image processing tutorials

🔮 Future Improvements

• Add data augmentation (rotation, flipping, zoom)
• Experiment with larger image sizes
• Implement transfer learning with pre-trained models (VGG16, ResNet)
• Add prediction visualization
• Deploy model as a web application

📄 License

This project is open source and available for educational purposes.

👨‍💻 Author

millingtonsully

---

Built as part of an Honors Programming project