DAH: Domain Adapted Deep Image Hashing

Abstract

With abundant labeled data, deep hashing methods have shown great success in image retrieval. However, these methods are often less powerful when applied to novel datasets. In this paper, we apply unsupervised domain adaptation techniques to improve a state-of-the-art deep hashing method, used in a cross-domain scenario where the model is trained with labeled source data and is evaluated with target data. Experiments show that the generalization capability of a supervised hashing method can be improved by the applied domain adaptation techniques.