Learning to Assemble Neural Module Tree Networks for Visual Grounding | Read Paper on Bytez