This project implements a gender classification system using the CLIP model for feature extraction and a custom neural network for classification. The dataset consists of male and female images, and the model predicts the gender based on the provided image.
This project leverages the OpenAI CLIP model to extract features from images and a custom TensorFlow neural network to classify gender as either male or female. The system is trained on a dataset of images categorized by gender and achieves a high level of accuracy.
The dataset used for this project is the CCTV Gender Classifier Dataset. It contains two folders, FEMALE and MALE, each holding .bmp images representing female and male individuals respectively.
Note: You need to download and place the dataset in your local directory. Update the data_folder variable in the code with the correct path to your dataset.
To run this project, you will need to have the following libraries installed:
tensorflowtorchtransformersnumpyPillowscikit-learn
You can install these dependencies using pip:
pip install tensorflow torch transformers numpy Pillow scikit-learnDownload the dataset and place it in a folder (e.g., CCTV Gender Classifier Dataset). Update the data_folder variable in the code with the path to your dataset. Run the script to train the model
python gender_classification.pyAfter training, you can test the model on a new image by calling the predict_gender(image_path) function.
The model architecture consists of two main parts:
- Feature Extraction with CLIP:
-The CLIP model (openai/clip-vit-base-patch32) is used to extract features from the images.
-The extracted features are passed into a custom neural network for classification.
- Custom Neural Network:
-Input Layer: Receives the extracted features from the CLIP model.
-Dense Layer 1: Fully connected layer with 128 units and ReLU activation.
-Dropout Layer 1: Dropout layer with a rate of 0.3 to prevent overfitting.
-Dense Layer 2: Fully connected layer with 64 units and ReLU activation.
-Dropout Layer 2: Dropout layer with a rate of 0.3 to further reduce overfitting.
-Output Layer: A dense layer with a single unit and a sigmoid activation function for binary classification (male or female).
-The model is compiled with the Adam optimizer and binary cross-entropy as the loss function, which is suitable for binary classification tasks. It also tracks accuracy as a performance metric.
-The training process includes early stopping to prevent overfitting, monitoring the validation loss and restoring the best weights after the training is completed.
The model achieves around 80% accuracy on the test set. Below are the sample results:
Test Loss: 1.384 Test Accuracy: 80.4%
Contributions are welcome! Please feel free to submit a pull request or open an issue to discuss your ideas for improving the project.