Gender-Classification

A Gender Classifier based on the UKT Image dataset and a custom VGG CNN architecture

UKT Image Dataset

• Consists of 20k+ face images in the wild (only single face in one image)
• Provides the correspondingly aligned and cropped faces
• Provides the corresponding landmarks (68 points)
• Images are labelled by age, gender, and ethnicity.
• The dataset can be downloaded from https://susanqq.github.io/UTKFace/ (Note: Please use the Aligned and Cropped Faces repository)
• The labels of each face image is embedded in the file name, formated like [age][gender][race]_[date&time].jpg
• The images were seggregated by using the Gender code in the lables of each image. Please use the file Preprocessing_File Copy_UKT Dataset for seggregating the dataset into Male and Females
• The dataset is split 80-20 into Train and Test

Custom VGG Architecture

• Input Shape - 128 x 128
• Convolution Layer (32)
• Convolution Layer (32)
• Pooling Layer (2x2)
• Convolution Layer (64)
• Convolution Layer (64)
• Pooling Layer (2x2)
• Convolution Layer (128)
• Convolution Layer (128)
• Pooling (2x2)
• Flatten Layer
• FUC (128) Relu
• Output Later (Output - 1) (Activation = Sigmoid; Loss = Binary Crossentropy; Optimiser = Adam)

Using the Trained Model

• Use the file Model Generator_Gender Prediction_CustomVGG to train the model on the UKT Image dataset and save the model file (.h5) at a desired location
• Use the file Video_Gender Predictor_CustomVGG_CNN by inserting the Model file name and location in order to predict gender with the computer Webcam