Similarity-preserving hashing is a commonly used method for nearest neighboursearch in large-scale image retrieval. For image retrieval, deep-networks-basedhashing methods are appealing since they can simultaneously learn effectiveimage representations and compact hash codes. This paper focuses ondeep-networks-based hashing for multi-label images, each of which may containobjects of multiple categories. In most existing hashing methods, each image isrepresented by one piece of hash code, which is referred to as semantichashing. This setting may be suboptimal for multi-label image retrieval. Tosolve this problem, we propose a deep architecture that learns\textbf{instance-aware} image representations for multi-label image data, whichare organized in multiple groups, with each group containing the features forone category. The instance-aware representations not only bring advantages tosemantic hashing, but also can be used in category-aware hashing, in which animage is represented by multiple pieces of hash codes and each piece of codecorresponds to a category. Extensive evaluations conducted on several benchmarkdatasets demonstrate that, for both semantic hashing and category-awarehashing, the proposed method shows substantial improvement over thestate-of-the-art supervised and unsupervised hashing methods.
展开▼
