FiLM: Visual Reasoning with a General Conditioning Layer