Search

Deep Hashing has achieved great success in large-scale image retrieval due to binary code's storage and computation efficiency. However, its learning paradigm under real-world environments is less studied, and most existing approaches are developed in the closed scenario, e.g., simple and unchanging semantics. When images of new classes emerge, they have to retrain the model on all history training datasets, but the constant data uploading makes this impractical. This paper proposes a novel method, called continual deep semantic hashing (CDSH), for learning binary codes of multi-label images with increasing classes. The CDSH consists of two hashing networks. One learns to hash the increasing semantics of data, i.e., label, into the semantic codes and accumulates label-code pairs as long-term knowledge, incorporating empirically verified loss and designed special regularization to ensure encoding old labels unchanged. The other learns to map images to the corresponding semantic code from a probabilistic view and solid knowledge via retaining history exemplar samples and projecting model gradients. We theoretically prove this improves the probability of old data's code unchanged after the model is updated. Extensive experiments on four widely used image datasets demonstrate that the CDSH method can continually learn hash functions and yield state-of-the-art retrieval performance. • The code drift is a key problem in continual hash learning for image retrieval. • The two-step hash learning is a useful architecture for continual hash learning. • Keeping label-code pairs can capture the similarities between old and new data. • The combination of different losses can alleviate the code drift of the hashing. • The gradient projection improves the probability of code of old data unchanged. [ABSTRACT FROM AUTHOR]

The goal of cross-modal hashing is to map data from several modalities into a compact Hamming space for efficient and accurate retrieval. Despite the satisfactory performance, existing approaches are reliant on the closed-world assumption. When confronted with real-world retrieval tasks involving out-of-distribution (OOD) semantic data, the similarity relationships of known data retained in hash codes tend to be disrupted by these unknown ones, resulting in retrieval performance degradation. To this end, we present a deep self-enhancing hashing (DSEH) method, simultaneously learning multi-level similarity-preserved hash codes of the known multi-label cross-modal data and robustness to OOD instances. Specifically, we propose to construct pseudo-OOD samples in the feature space using random linear combinations to explore OOD semantics, during the training process. Meanwhile, a prototype-based generative model is incorporated to aggregate batch data to enhance the data representation's differences in known and unknown semantics. Furthermore, we describe a bounded cosine quadrupled loss with distance bound to preserve the multi-level similarity of multi-label data and control the maximum distance between known data and the minimum distance between known and pseudo-OOD data for learning OOD robustness. Extensive experiments show that the DSEH achieves state-of-the-art performance on closed-world tasks and good performance on simulated real-world tasks. • The non-convex random linear combinations of features help construct pseudo-OOD data. • Features produced by the prototype-based auto-encoder enhance the OOD discriminative. • Constraining the maximum distance of dissimilar codes improves the OOD robustness. [ABSTRACT FROM AUTHOR]