The loss in accuracy between us and a custom hardware setup for eye-tracking is primarily due to the fact that we're using a very different eye-tracking method.

The loss in accuracy between us and a custom hardware setup for eye-tracking is primarily due to the fact that we're using a very different eye-tracking method. The "gold standard" for eye-tracking involves shining an infrared LED at the user's eye and measuring the reflection. Since we're trying to use no custom hardware at all, we have to use machine learning techniques to extract features from the visible-light image of the user's eye. This is a lot more difficult than locating the bright spot in a reflection, so there's a lot more that can go wrong.