Age and Gender Detection using Deep Learning

This project predicts age and gender from facial images using deep learning models such as VGG16 and ResNet with transfer learning. It is based on the Adience Benchmark Dataset and demonstrates how CNNs can accurately classify gender and estimate age groups even under challenging lighting and pose conditions.

---

📚 Overview

• Techniques used: Convolutional Neural Networks (CNNs), Transfer Learning, Data Augmentation
• Dataset: Adience benchmark dataset (publicly available)
• Accuracy achieved:
  • Gender detection: 96.59%
  • Age estimation: 62.73%

---

🧩 Project Structure

age-gender-detection/

├── project.ipynb

├── requirements.txt # Python dependencies

├── .gitignore # Files and folders to ignore in Git

├── LICENSE # License for reuse

├── Data/adience

---

⚙️ Installation and Usage

1️⃣ Clone the Repository

git clone https://github.com/<Bharathyalagi>/<age-and-gender-detection-using-deep-learning>.git
cd <age-and-gender-detection-using-deep-learning>

2️⃣ Clone the Repository

pip install -r requirements.txt

3️⃣ Download the Dataset

This project uses the Adience Benchmark Dataset for age and gender classification.

Download link:
[https://talhassner.github.io/home/projects/Adience/Adience-data.html]

[https://www.kaggle.com/datasets/ttungl/adience-benchmark-gender-and-age-classification]

After downloading, place the dataset in:

data/adience/

4️⃣ Run the Notebook

Open project.ipynb in Jupyter/kaggle/collab Notebook and run all cells step-by-step.

Optional

Environment and How I ran experiments

All experiments and results were produced in a Kaggle notebook environment (GPU-enabled). If you want to reproduce the results locally, use Python 3.8+ and install the dependencies from requirements.txt. Some functionality (GPU training) will be faster on a CUDA-enabled GPU.
If you ran this project in a different environment, please update the environment details here.

Evaluation and Metrics

The main evaluation metrics reported in the paper and produced by the notebook are:

Gender Detection

• Training accuracy: 96.59%
• Validation accuracy: 86.14%
• Epochs: 50

Gender Detection (Augmented)

• Training accuracy: 79.32%
• Validation accuracy: 81.25%
• Epochs: 100

Age Detection

• Training accuracy: 62.73%
• Validation accuracy: 58.42%
• Epochs: 200

Overall, the system performed strongly in gender detection due to its binary nature and the use of the Sigmoid function.
Age prediction achieved moderate accuracy because of overlapping facial features among age groups, but still demonstrated solid generalization with proper data augmentation.
Future improvements can focus on balanced age classes, larger datasets, and custom activation or ensemble models to further improve age prediction accuracy.

Confusion matrix and classification report

The notebook includes code to compute and display a confusion matrix, precision, recall and F1 scores for gender classification. See the Evaluation section of project.ipynb to reproduce these plots and tables.

Thank You