This project was focused on creating a high-accuracy Facial Age Estimation model building upon the research of state-of-the-art CNN architectures. We tested different pre-trained models including VGG16, ResNet and DenseNet as well as building our own CNN model. We fine-tuned and trained the models on the VGGFace dataset and performed data cleaning and pre-processing to ensure fair and balanced training between different age groups, races and sex.
Our results showed that due to lack of data among certain age and racial groups, the models were more accurate at predicting ages for certain demographics over others.
We believe that to ensure equality in new Facial Image Recognition technology, a balanced dataset must be collected among all demographic groups.
Please checkout the Presentation folder to see our Final Project Presentation.
To see our code please checkout the Training folder.