Produce likelihood for each eye blink predictions

[uetchy]

Because we want to know if webcam actually grasps an image of eyes, the prediction likelihood needs to be added on the output layer.

Solution

• Add random training data containing no eye figure in it and assign them ground-truth likelihood "0".
• Extend the existing dataset and add the likelihood column with value "1".

Refs

• Eye Tracking Update: Neural Network Architecture and Loss Functions