Re-trained the edges-to-dogs model with head photos. It’s...