Dogs vs Cats Image Classification using MobileNetV2

A deep learning image classification project that distinguishes between cats and dogs using a MobileNetV2-based convolutional neural network with transfer learning.

---

📋 Table of Contents

• Overview
• Project Objective
• Technologies Used
• Model Architecture
• Dataset
• Project Structure
• Requirements
• Setup
• How to Run
• Model Performance
• Generated Outputs
• Learning Outcomes
• Future Improvements
• Author
• Internship Details

---

🎯 Overview

This project is part of the SkillCraft Technology Machine Learning Internship and focuses on binary image classification between cats and dogs.

Workflow:

• Loading and organizing the dataset
• Splitting data into training and validation sets
• Applying data augmentation
• Using MobileNetV2 pretrained on ImageNet
• Adding a custom binary classification head
• Training the model
• Evaluating performance on test data
• Generating classification report and confusion matrix
• Predicting the class of a single image

---

🎯 Project Objective

Build a CNN model capable of classifying whether an image contains a cat or a dog with high accuracy using transfer learning and image preprocessing.

---

🛠 Technologies Used

• Python 3
• TensorFlow / Keras
• MobileNetV2
• NumPy
• Scikit-learn
• Matplotlib
• Seaborn
• Transfer Learning

---

🏗 Model Architecture

The model uses MobileNetV2 as the feature extractor, with the pretrained base model frozen and a new binary classification layer added on top.

Input Image (160 x 160 x 3)
         |
         v
Data Augmentation
         |
         v
MobileNetV2 (ImageNet pretrained)
         |
         v
Global Average Pooling
         |
         v
Dropout (0.3)
         |
         v
Dense Layer (1 unit, Sigmoid)
         |
         v
CAT / DOG Prediction

---

📊 Dataset

The project uses the standard cats-vs-dogs dataset organized as follows:

dataset/
├── train/
│   ├── cats/
│   └── dogs/
├── test/
│   ├── cats/
│   └── dogs/
└── README.md

Dataset Summary:

Split	Cats	Dogs	Total
Training	5,001	5,001	10,002
Validation	750*	750*	1,500
Testing	450	501	951

*Validation is created automatically from 15% of the training set during training.

---

📁 Project Structure

SCT_ML_Task3/
├── README.md
├── requirements.txt
├── dataset/
│   ├── train/
│   │   ├── cats/
│   │   └── dogs/
│   └── test/
│       ├── cats/
│       └── dogs/
├── models/
│   ├── dogs_cats_mobilenetv2.keras
│   └── dogs_cats_mobilenetv2_final.keras
├── results/
│   ├── cnn_classification_report.txt
│   ├── cnn_confusion_matrix.png
│   ├── cnn_training_accuracy.png
│   └── cnn_training_loss.png
└── src/
    ├── train_cnn.py
    ├── evaluate_cnn.py
    └── predict_cnn.py

---

📦 Requirements

Make sure Python 3.10+ is installed, then install dependencies:

pip install -r requirements.txt

---

⚙️ Setup

1. Clone the repository

git clone https://github.com/VigneshVarsha/SCT_ML-3.git
cd SCT_ML-3

2. Create and activate a virtual environment

Windows PowerShell:

python -m venv .venv
.\.venv\Scripts\Activate.ps1

Windows Command Prompt:

python -m venv .venv
.venv\Scripts\activate.bat

3. Install dependencies

pip install -r requirements.txt

---

▶️ How to Run

Train the model

python src/train_cnn.py

This script:

• Loads the dataset
• Creates the validation split
• Applies augmentation
• Builds the MobileNetV2 model
• Compiles the model
• Trains for 5 epochs
• Saves the trained model in models/

Evaluate the model

python src/evaluate_cnn.py

This generates:

• results/cnn_confusion_matrix.png
• results/cnn_classification_report.txt

Predict a single image

python src/predict_cnn.py "dataset/test/cats/cat_test_00010.jpg"

Example output:

Prediction: CAT
Confidence: 84.93%

Another example:

python src/predict_cnn.py "dataset/test/dogs/dog_test_00019.jpg"

---

📈 Model Performance

The trained model achieved strong results on the test set.

Metric	Value
Test Accuracy	98.00%
Test Loss	0.0674

Classification Report

              precision    recall  f1-score   support

      cats       0.98      0.98      0.98       450
      dogs       0.98      0.98      0.98       501

      accuracy                           0.98       951
     macro avg       0.98      0.98      0.98       951
  weighted avg       0.98      0.98      0.98       951

Confusion Matrix

                  Predicted
                  Cat    Dog
Actual Cat       440     10
Actual Dog         9    492

Interpretation:

• ✅ 440 cats correctly classified as cats
• ✅ 492 dogs correctly classified as dogs
• 📊 932 correct predictions out of 951 total images

---

📂 Generated Outputs

The project creates visual and textual artifacts in the results/ folder:

File	Description
cnn_training_accuracy.png	Training accuracy visualization
cnn_training_loss.png	Training loss visualization
cnn_confusion_matrix.png	Confusion matrix heatmap
cnn_classification_report.txt	Detailed classification metrics

---

🎓 Learning Outcomes

This project demonstrates:

• Image classification with deep learning
• CNN architecture design
• Transfer learning using MobileNetV2
• Data augmentation for image data
• Binary classification with sigmoid output
• Model evaluation using precision, recall, F1-score, and accuracy
• Prediction on unseen images
• Saving and loading TensorFlow/Keras models

---

🚀 Future Improvements

• Fine-tuning deeper MobileNetV2 layers
• Trying additional pretrained architectures (EfficientNet, ResNet, etc.)
• Increasing image resolution
• Experimenting with more augmentation techniques
• Building a web or GUI-based prediction interface
• Deploying the model as an application (Flask/FastAPI/Streamlit)

---

👨‍💻 Author

Devarakonda Vignesh Varsha

---

📌 Internship Details

Detail	Information
Internship	SkillCraft Technology
Domain	Machine Learning
Task	Task 03 - Dogs vs Cats Image Classification
Final Test Accuracy	98.00%

---

Built with ❤️ during SkillCraft Technology ML Internship