OmniArt Eye Classifier

This package is an (painted) eye colour classifier. It is trained on a dataset of 2600 different painted eyes that were extracted from the OmniArt dataset. The model achieved an accuracy of 83% for following ten classes: amber, blue, brown, gray, grayscale, green, hazel, irisless, negative, red. The negative is used to classify images that are not eyes, or where, for example, the eye lid is closed.

Project origin

This package is part of a Master's thesis at the University of Amsterdam.