🐱🐶 Cat vs Dog Classification System

A complete end-to-end Modular Deep Learning project that performs binary image classification (Cat vs Dog) using multiple CNN architectures.
The project compares custom-built and pre-trained models, selects the best-performing model, and based on classification accuracy, and provides an interactive Streamlit-based user interface for model inference.

🏆 Best Model: ResNet-18
📈 Test Accuracy: 99%
🧠 Models Used: LeNet-5 (from scratch), AlexNet, ResNet-18

🎥 Demo

---

📌 Table of Contents

• Why This Project Exists
• Project Highlights
• Models Used
• Workflow
• Technologies Used
• Project Structure
• Setup Instructions
• Deployment

---

Why This Project Exists

Most beginner-level Cat vs Dog projects:

• Use only one pre-trained model
• Treat CNNs as a black box
• Lack modular code structure

This project was built to go beyond that.

🎯 Goals of This Project

• Understand CNN architectures deeply by implementing LeNet-5 from scratch
• Compare multiple CNN models fairly under the same training conditions
• Apply software engineering best practices (modularity, reusability)

📌 Focus: Learning-oriented, scalable, and resume-ready Deep Learning project.

---

🚀 Project Highlights

🧠 Deep Learning & Model Comparison

• Implemented LeNet-5 CNN architecture from scratch
• Used AlexNet and ResNet-18 for comparison
• All models trained for 2 epochs only to ensure fair comparison
• Best model selected based on validation accuracy

🏆 Best Model Selection

• ResNet-18 achieved 99% accuracy
• Automatically selected as the final inference model

⚙️ Modular Codebase

• Separate modules for:
  • Model definitions
  • Training logic
  • Evaluation
  • Inference
• Clean, readable, and scalable structure

🌐 Streamlit Web Application

• Interactive UI for image upload
• Real-time prediction (Cat or Dog)
• Lightweight and user-friendly interface

---

🧠 Models Used

1️⃣ LeNet-5 (From Scratch)

• Fully implemented CNN architecture
• Custom convolutional, pooling, and fully connected layers
• Built for learning and architectural understanding

2️⃣ AlexNet

• Deeper CNN with higher representational capacity
• Used for performance comparison

3️⃣ ResNet-18 (Best Model)

• Residual connections to solve vanishing gradient problem
• Achieved 99% accuracy
• Selected as final production model

---

🔁 Workflow

1. Data Loading

   • Images loaded from train/test directories
   • Standard preprocessing applied

2. Model Training

   • Train LeNet-5, AlexNet, and ResNet-18
   • Fixed training epochs (2) for fair comparison

3. Evaluation

   • Accuracy calculated for each model
   • Best model selected automatically

4. Inference

   • Saved best model weights
   • Used for real-time prediction in Streamlit app

5. Deployment

   • Streamlit UI deployed on Streamlit Cloud

---

🛠️ Technologies Used

Deep Learning & Programming

• Python
• PyTorch
• Torchvision
• NumPy

Model Architectures

• LeNet-5 (Custom implementation)
• AlexNet
• ResNet-18

Web & Deployment

• Streamlit – UI development
• Streamlit Cloud – Deployment platform

Tools

• Jupyter Notebook – Experiments & training
• Git & GitHub – Version control

---

📁 Project Structure

CAT VS DOG CLASSIFIER/
│
├── configs/
│
├── models/
│ ├── init.py
│ ├── lenet5.py
│ └── model_factory.py ← Factory of models (LeNet, alexnet, ResNet)
│
├── NoteBook/
│ ├── Cat_vs_Dog_Classifier.ipynb
│
├── training/
│ ├── init.py
│ ├── evaluate.py
│ ├── train_all_models.py
│ └── train_utils.py
│
│
├── .gitignore
├── app.py ← Application entry point
├── inference.py
├── README.md
└── requirements.txt

---

Setup Instructions

Prerequisites

To run this project locally, ensure you have the following installed:

• Python 3.8+
• Git
• Virtual environemnt (recommended)

Steps

1. Clone the Repository:

   git clone https://github.com/Umer2900/Cat-vs-Dog-Classifier
   cd Cat-vs-Dog-Classifier

2. Install Dependencies:

   python -m venv venv
   venv\Scripts\activate       # Windows

3. Install Streamlit App:

   pip install -r requirements.txt

4. Run the Application:

   streamlit run app.py

   The app will be available at 🌐 http://localhost:8501.