CNN-for-Gender-Classification

Overview

This assignment focuses on building and training a Convolutional Neural Network (CNN) using the ResNet18 architecture for gender classification. The dataset contains facial images of males and females, structured into separate folders. The core objective was to implement a complete deep learning pipeline using PyTorch to classify images into two gender categories:

• Male → 0
• Female → 1

Dataset

• URL: Kaggle - Adience Gender Dataset
• Contains face images divided into 'm' and 'f' directories.

Major Steps

1. Environment Setup & Libraries Installation

Installed the following Python libraries:

• torch
• torchvision
• matplotlib
• opencv-python
• Pillow

2. Dataset Preparation

• Mounted Google Drive to Google Colab for file access.
• Extracted the dataset archive to a working directory.
• Verified existence of gender-labeled folders: 'm' and 'f'.

3. Custom Dataset Class

• Created GenderDataset class extending torch.utils.data.Dataset.
• Mapped:
  • 'm' → label 0
  • 'f' → label 1
• Applied image transformations:
  • Resizing
  • Normalization
  • Augmentation using torchvision.transforms

4. Data Loading

• Split dataset:
  • 80% Training
  • 10% Validation
  • 10% Testing
• Used torch.utils.data.DataLoader for efficient batching and shuffling.

5. Model Definition

• Loaded a pre-trained ResNet18 model from torchvision.models.
• Replaced the final fully connected layer to output 2 classes for gender classification.

6. Training Loop

• Trained for 10 epochs.
• Used:
  • Loss Function: CrossEntropyLoss
  • Optimizer: Adam
• Tracked:
  • Training loss
  • Training accuracy
  • Validation loss
  • Validation accuracy

7. Evaluation

• Evaluated the model on the test dataset.
• Achieved a test accuracy of approximately 82.35%.

8. Results & Observations

• Observed consistent accuracy improvements across epochs.
• Noted improved generalization due to data augmentation.
• Validation accuracy remained stable, suggesting no significant overfitting.

---

Final Accuracy

Test Accuracy: ~82.35%

Tools Used

• Google Colab
• PyTorch
• torchvision
• OpenCV
• matplotlib
• Pillow

Author

Priteesh Madhav Reddy
Date: April 2025