Facial Landmark Detection using Kaggle Competition Training Data
In the facial landmarks notebook I experimented with using DLib's facial landmark detector as a tool for filling in the missing datapoints from the Kaggle dataset. Dlibs landmarks are shown as blue in the following image, with red denoting the original Kaggle dataset. Note: for eye centres I average the four Dlib landmarks that surround each eye.
This notebook uses the images from both the train and test csv files to train a simple autoencoder, sharing the similar encoder layers from the previous notebook but stopping at a minimum dimension of 12x12.
