Disparity between the region caption based queries in our experiments and the user queries in real scenario

First, we address some common concerns. Disparity between the region caption based queries in our experiments and the user queries in real scenario . Compared with the simulated caption data, the real user queries we collected show distinct patterns, e.g. That's why we observe smaller performance We envision the proposed method could generally help with natural image search. We confirm in our experiments that more placeholders result in better performance but also lead to scalability difficulties.