Recent years have witnessed wide applications of deep learning for large-scale image hashing tasks, as deep hashing algorithms can simultaneously learn feature representations and hash codes in an end-to-end way. However, although these methods have obtained promising results to some extent, they seldom take the effect of different training samples into account and treat all samples equally throughout the training procedure. Therefore, in this paper, we propose a novel deep hashing algorithm dubbed "Teach to Hash" (T2H), which introduces a "teacher" to automatically select the most effective samples for the current training period. To be specific, the "teacher" utilizes two criteria to measure the effectivity of all samples, and iteratively update the training set with the most effective ones. Experimental results on two typical image datasets indicate that the introduced "teacher" can significantly improve the performance of deep hashing framework and the proposed method outperforms the state-of-the-art hashing methods.
展开▼
