Deep Hashing Using N-pair Loss for Image Retrieval

Deep hashing algorithm is one of the most effective techniques for the approximate nearest neighbor search for large-scale image retrieval. Existing deep hash algorithms are based on paired labels and triple ordering loss, they usually only interact with one negative class, and the convergence speed is too slow. In this paper, we propose a novel deep hashing algorithm called N-pair loss deep hashing (NPLDH), which optimization based on the N-pair loss function can help deep hash models to train more effectively. Experimental results show that our NPLDH algorithm achieves higher performance in image retrieval algorithms on the CIFAR-10 and NUS-WIDE datasets.