🐱🐶 Cats vs Dogs Image Classification 📌 Project Overview

This project implements an image classification model that distinguishes between cats and dogs using Convolutional Neural Networks (CNNs) and transfer learning with MobileNetV2. It is designed for both experimentation in Jupyter/Colab notebooks and production-style training via Python scripts.

⚙️ Installation

Clone the repository:

git clone https://github.com/Oghuz20/cats-vs-dogs.git cd cats-vs-dogs

Create and activate a virtual environment:

python -m venv .venv

Linux/Mac

source .venv/bin/activate

Windows

.venv\Scripts\activate

Install dependencies:

pip install -r requirements.txt

🚀 Usage 🔹 Training

Train the model with transfer learning and fine-tuning:

python src/train.py --epochs 5 --finetune_epochs 3

🔹 Prediction (Local CLI)

Prediction (Local CLI)

You can directly use the pre-trained model included in models/ without retraining.

Run prediction on an image:

python src/predict_local.py --image path/to/your_image.jpg 🔹 Notebook

For interactive exploration and step-by-step workflow, open:

notebooks/cats_vs_dogs_colab.ipynb

📂 Project Structure cats-vs-dogs/ ├─ notebooks/ │ └─ cats_vs_dogs_colab.ipynb # Interactive notebook (Colab-ready) ├─ artifacts/ │ ├─ confusion_matrix.png # Model evaluation plots │ ├─ accuracy_curve.png │ ├─ loss_curve.png │ └─ classification_report.txt ├─ models/ │ └─ cats_vs_dogs_mobilenetv2.keras # ✅ Pre-trained model (included) ├─ src/ │ ├─ train.py # Training script │ ├─ predict_local.py # Local prediction script │ └─ utils.py # Helper functions ├─ data/.gitkeep # Placeholder for raw data ├─ requirements.txt ├─ .gitignore └─ README.md

📊 Results

The model was trained using MobileNetV2 with transfer learning and fine-tuning. It achieves state-of-the-art accuracy on the Cats vs Dogs dataset.

✅ Overall Performance

Validation Accuracy: ~98%

Loss & Accuracy Curves: available in artifacts/

Saved Model: models/cats_vs_dogs_mobilenetv2.keras

🔹 Confusion Matrix

Cats correctly classified: 2220 / 2280

Dogs correctly classified: 2337 / 2372

🔹 Classification Report precision recall f1-score support

     Cat       0.98      0.97      0.98      2280
     Dog       0.97      0.99      0.98      2372

accuracy                           0.98      4652

macro avg 0.98 0.98 0.98 4652 weighted avg 0.98 0.98 0.98 4652

🔎 Interpretation

Both precision and recall are very high (>97%), proving reliability for both classes.

F1-score = 0.98 → excellent balance between precision and recall.

Very few misclassifications (~95 total errors out of 4652 images).

🖼️ Sample Predictions

Here are some example predictions made by the model:

Input Image	Prediction	True Label
	🐱 Cat	🐱 Cat
	🐶 Dog	🐶 Dog

👉 To reproduce, place test images inside artifacts/ and run:

python src/predict_local.py --image path/to/your_image.jpg

✅ Key Features

Transfer learning with MobileNetV2

Efficient preprocessing pipeline with tf.data

Separate training + fine-tuning phases

Clear evaluation metrics and visualizations

Easy-to-use CLI for local predictions

📜 License

This project is licensed under the MIT License.