This document summarizes a research paper on query-adaptive image search using hash codes. It introduces an approach that learns bitwise weights for hash codes offline to represent semantic concept classes. At query time, weights are computed based on the query's proximity to concepts. This allows ranking images by a weighted Hamming distance at a finer-grained level than the original Hamming distance. The paper shows this approach provides clearer improvements over methods that use a single hash code weight set for all queries.