🧠 UTKFace Ethnicity Classification with CNN (PyTorch)

This project focuses on building a Convolutional Neural Network (CNN) using PyTorch to classify facial images into 5 ethnicity categories using the UTKFace dataset. The model is trained on image data with labels extracted directly from the filenames.

---

📂 Dataset: UTKFace

• 20,000+ in-the-wild facial images
• Labels include: Age, Gender, and Ethnicity
• File name format: [age]_[gender]_[ethnicity]_[date&time].jpg
• For this project, only ethnicity is used as the target variable
• Link to Data https://www.kaggle.com/datasets/jangedoo/utkface-new/data

Ethnicity Classes:

• 0 → White
• 1 → Black
• 2 → Asian
• 3 → Indian
• 4 → Others

---

💡 Project Objective

To train a CNN to accurately predict the ethnicity of individuals from facial images using:

• Data preprocessing & balancing
• Custom PyTorch Dataset class
• Weighted random sampling for fair training
• Visualization of predictions and performance metrics

---

🏗️ Architecture

• 3 Convolutional Layers (Conv2D + ReLU + MaxPooling)
• Flatten → Fully Connected Layers
• Dropout Layer (0.5) to reduce overfitting
• Final Linear Layer with 5 output classes

---

🧪 Evaluation

• Training loss and accuracy over 30 epochs
• Confusion matrix heatmap for test predictions
• Visual examples: Correct vs Incorrect predictions by ethnicity
• Classification report with precision, recall, F1-score

---

🔧 Setup Instructions

# Clone the repository
git clone https://github.com/yourusername/utkface-ethnicity-cnn.git
cd utkface-ethnicity-cnn

# Install dependencies
pip install -r requirements.txt

# Download UTKFace dataset and place inside /data/UTKFace