Computer vision project for NNDL W995 at Columbia.
Given training data set that is split into 3 categories (dog, lizard, bird) and ~80 subspecies split amongst the super class labels. Test data contains out of distribution samples for subspecies, as well as distribution shifts between train/test data sets.