What Makes a Face Look like a Hat: Decoupling Low-level and High-level Visual Properties with Image Triplets | Read Paper on Bytez