Adversarial learning based intra-modal classifier for cross-modal hashing retrieval

Cross-modal hashing is popular for its high efficiency, low dimensionality, low storage overhead in information retrieval, and its ability to effectively characterize the consistency of high-level features of multi-source heterogeneous data. In this article, we present a fresh method that utilizes an intra-modal classifier that employs discrimination loss to learn intra-modal discriminations based on classification targets of image-text pairs. Meanwhile, the modality classifier serves as the discriminator in a GAN framework, using adversarial loss to reduce heterogeneity between different modality representations. To measure the efficiency of our method, we evaluate our proposed approaches against 8 other methods on three public benchmark datasets and show that our approaches outperform them in both I2T and T2I query.