Submissions are evaluated on *classification accuracy* between the predicted class and the ground truth class.