https://github.com/haturusinghe/cnn-age-gender

This repository contains implementations of convolutional neural networks (CNNs) for age and gender classification using facial images from the UTKFace dataset.
https://github.com/haturusinghe/cnn-age-gender

age-detection convolutional-neural-networks gender-detection tensorflow

Last synced: 2 months ago
JSON representation

This repository contains implementations of convolutional neural networks (CNNs) for age and gender classification using facial images from the UTKFace dataset.

• Host: GitHub
• URL: https://github.com/haturusinghe/cnn-age-gender
• Owner: haturusinghe
• Created: 2021-09-23T01:05:39.000Z (almost 5 years ago)
• Default Branch: master
• Last Pushed: 2025-01-09T03:47:38.000Z (over 1 year ago)
• Last Synced: 2025-03-31T06:16:31.705Z (over 1 year ago)
• Topics: age-detection, convolutional-neural-networks, gender-detection, tensorflow
• Language: Jupyter Notebook
• Homepage:
• Size: 6.36 MB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Age and Gender Classification using CNNs

This repository contains implementations of convolutional neural networks (CNNs) for age and gender classification using facial images from the UTKFace dataset.

## Models

### Age Classification Model

- Uses a CNN architecture with separable convolutions

- Predicts age as a continuous value using regression

- Features:

  - Input image size: 200x200x3

  - 6 convolutional blocks with batch normalization and LeakyReLU

  - Dense layers with dropout for regularization 

  - Outputs lower and upper age bounds

  - Mean Absolute Error loss function

### Gender Classification Model

- CNN for binary classification of gender (male/female)

- Architecture:

  - Input image size: 200x200x3

  - 5 convolutional blocks with batch normalization

  - Dense layers with dropout

  - Softmax output layer

  - Categorical crossentropy loss

## Dataset

- Uses the [UTKFace dataset](https://susanqq.github.io/UTKFace/)

- Images are preprocessed to 200x200 pixels

- Training/validation split: 70%/30%

- Data augmentation with random flips and rotations

## Training

- Models trained using Adam optimizer

- Early stopping and model checkpointing

- TensorBoard logging for monitoring training

- Learning rate scheduling

## Results

The models achieve:

- Age prediction: MAE of ~5.5 years

- Gender classification: ~90% accuracy

## Requirements

- TensorFlow 2.x

- NumPy

- Matplotlib

- scikit-learn

## Usage

The Jupyter notebooks contain the complete implementation:

- `age_class_model.ipynb`: Age regression model

- `gender_class_model(UTK_Crop).ipynb`: Gender classification model

Models can be exported to TFLite format for mobile/edge deployment.