Computer vision - Glasses or no glasses

The purpose of this project is to create a Machine Learning model which will be able to classify different kind of images in two groups, glasss and no glasses. In this model it would be applied a pretrained model MobileNet V2, developed by google, and it will be applied fine-tuning with the weight of the pretrained model.

The model is based on Tensorflow/Learn/Tutorials/Images/Transfer learning and fine-tuning - https://www.tensorflow.org/tutorials¶

Technical details about thje project:

📍 Programming language: Python

📍 Library: Tensorflow

📍 Applied algorithm: Convolutional Neural Network

Data sources:

• Main source: https://www.kaggle.com/jeffheaton/glasses-or-no-glasses
• Secondary source: 1 Million Fake Faces - 1 - https://www.kaggle.com/tunguz/1-million-fake-faces, from which I took 311 adittional photos belonging to the class no_glasses

Some figures:

• Some images from the train sample:

• Example of data augmentation:

Results:

• Application of fine tuning (after the green line):