Eyes 👀 Speaks ♀️ ♂️

Through this project we will try to classify gender ( Male/ Female) by the Morphology of the Eye

Problem Statement

The anthropometric analysis of the human face is an essential study for performing craniofacial plastic and reconstructive surgeries.

Facial anthropometrics are affected by various factors such as age, gender, ethnicity, socioeconomic status, environment, and region.  

Plastic surgeons who undertake the repair and reconstruction of facial deformities find the anatomical dimensions of the facial structures useful for their surgeries. These dimensions are a result of the Physical or Facial appearance of an individual. Along with factors like culture, personality, ethnic background, age; eye appearance and symmetry contributes majorly to the facial appearance or aesthetics. 

Objective:

Our objective is to build a model to scan the image of an eye of a patient and find if the gender of the patient is male or female.

Dataset:

This is a dataset consisting of about 11k images of male eyes and female eyes. Our main objective here is to determine the gender of a person by training a model on these eyes. This model can make gender prediction an easier task. We could predict the gender of a person even if we don't have complete access to the face of the person.

Modelisation

Simple Model 

Through implementing a simple VGG using TensorFlow, and Adam as an optimize and Binary_crossentrpoy as loss we got an accuracy of 93%.

Transfer Learning

While using TL, and by going under several experiments with many architectures we found a high accuracy using Xception with 99%.

I will be uploading the model into Github repo soon.