Cat vs. Dog Recognition 🐾

A Deep Learning project built with TensorFlow and Keras to classify images of cats and dogs using a custom-built Convolutional Neural Network (CNN).

🚀 Overview

This repository contains a manual CNN architecture designed to identify pets. To ensure high generalization and prevent overfitting, the model utilizes Data Augmentation and Dropout layers, achieving a final training accuracy of 84.94%.

---

🧠 Model Architecture

• 4 Convolutional Layers: Extracts features from simple edges to complex shapes.
• MaxPooling: Reduces dimensionality while retaining spatial features.
• Dropout (0.5): Prevents memorization by randomly deactivating neurons.
• Data Augmentation: Flips, rotates, and zooms images to improve real-world performance.

---

📊 Performance Metrics

Metric	Value
Final Training Accuracy	84.94%
Validation Accuracy	~82.10%
Epochs	15
Optimizer	Adam

---

🛠️ Project Structure

Cat-vs-Dog-recognition/
├── data/               # Raw Kaggle images (Git Ignored)
├── dataset_final/      # Organized train/test split (Git Ignored)
├── models/             # Saved .keras model files
├── src/
│   ├── setup_data.py   # Data organization script
│   ├── train.py        # Deep Manual training script
│   └── predict.py      # Inference script for testing
└── requirements.txt

🚦 Getting Started

1. Clone the Repository

git clone [https://github.com/dhruvil-1207/Cat-vs-Dog-recognition.git](https://github.com/dhruvil-1207/Cat-vs-Dog-recognition.git)
cd Cat-vs-Dog-recognition

2. Install Dependencies

pip install -r requirements.txt

3. Setup Dataset (Optional) If you wish to re-train the model, place the Kaggle "Dogs vs Cats" images in the data/ folder and run:

python src/setup_data.py

🔍 Training & Prediction Phase 1: Training (Optional) The repository includes pre-trained weights in the models/ folder. To re-train the model from scratch:

python src/train.py

Phase 2: Prediction To test the model on any image, run the inference script:

python src/predict.py

Instructions after execution: The script will prompt you for an Image Path. You can provide:

A Relative Path: ../my_pet.jpg

An Absolute Path: D:\Images\dog_test.jpg

A Test Image: ../dataset_final/test/cats/cat.1001.jpg

Output: The script returns the predicted animal and a Confidence Score (e.g., RESULT: DOG 🐶 (Confidence: 92.45%)).