📑 Gender Image Classification

🔹 Project Overview

This project aims to classify images of men and women using Deep Learning techniques. By leveraging the power of Convolutional Neural Networks (CNNs) and the pre-trained VGG16 architecture (originally trained on the ImageNet dataset), the model extracts meaningful features from input images. The model is then fine-tuned on a custom dataset to achieve high accuracy in distinguishing between male and female images.

📂 Dataset

The project uses the Men-Women Classification dataset from Kaggle. To download the dataset in your Google Colab environment, run the following command:

!kaggle datasets download -d saadpd/menwomen-classification

!unzip menwomen-classification.zip -d ./data

After downloading, unzip the dataset and organize it into appropriate folders for training, validation, and testing.

The dataset is organized into three subsets:

• Training Set: 1598 images
• Validation Set: 400 images
• Test Set: 800 images

Folder structure:

men_vs_women_small/
├── train/
│ ├── men/
│ └── women/
├── validation/
│ ├── men/
│ └── women/
└── test/
├── men/
└── women/

⚙️ Prerequisites

Before running this project, ensure you have the following installed:

• Python 3.x
• TensorFlow (>= 2.0)
• Keras
• NumPy
• pandas
• Matplotlib
• Kaggle API (optional, for downloading the dataset)

You can install these dependencies using pip:

pip install tensorflow numpy matplotlib pandas

🏗️ Model Design

The model is based on VGG16 (pre-trained on ImageNet). It consists of the following components:

• Frozen convolutional base for feature extraction

• Dense layer (256 units, ReLU) + Dropout (0.5)

• Final sigmoid output for binary classification

🔄 Training Phases

1. Feature Extraction

• VGG16 convolutional base frozen

• Only top classifier layers trained

2. Fine-Tuning

• Last 4 convolutional blocks of VGG16 unfrozen

• Optimizer: Adam (learning_rate = 1e-5)

• Callbacks used to save the best-performing model

📌 Key Learnings

• Transfer learning with pre-trained CNNs significantly improves accuracy on small datasets.

• Fine-tuning only a subset of convolutional layers avoids overfitting.

• Data augmentation is essential for robustness.

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

📬 Contact

Developed by Zahra Sahranavard
For inquiries: zahra.sahranavard7622@iau.ir