Lei Huang

Large-scale image retrieval increasingly depends on triplet-based deep hashing methods, which typically construct training triplets using class labels. However, existing triplet-based hashing methods relying on sparse and noisy categorical labels suffer from two main issues: no triplet, where sparse annotations limit valid triplet formation, and bad triplet, where noisy or ambiguous labels cause misleading samples, which together hinder the network from learning a well-structured similarity space, causing semantically similar images to scatter and dissimilar ones to collapse, ultimately undermining retrieval accuracy. In this paper, we solve this dilemma with a novel unified deep hashing framework, termed Deep Relational Knowledge Distillation Hashing (DRKDH), which decouples semantic relation modeling from visual feature learning by leveraging a teacher-student paradigm to learn semantically consistent hash codes. Specifically, the teacher model captures comprehensive semantic structural relationships across all instances, providing information-rich and noise-resilient supervision to guide the student model in learning expressive and discriminative visual embeddings, effectively mitigating the impact of low-quality triplet labels. Furthermore, by leveraging similarity weights predicted by the teacher model, a relaxed masking triplet loss is introduced into the student model, which dynamically adjusts the contribution of each triplet based on its informativeness, suppressing invalid or misleading triplets while emphasizing valuable ones to enhance training efficiency. Comprehensive experiments on THINGS, ImageNet, MIRFLICKR-25K, and NUS-WIDE show that DRKDH consistently outperforms state-of-the-art deep hashing methods under sparse and noisy supervision, while additional evaluations suggest competitive performance in selected zero-shot and few-shot settings. Source codes: https://github.com/QinLab-WFU/DRKDH.

Due to the preferable efficiency of storage and computation, hashing algorithms show great potential. With the explosive growth of web data, new and emerging categories of multimedia data continue to increase, and zero-shot hashing has been one of the hot research topics in retrieval research. Nevertheless, existing zero-shot hashing methods focus mainly on single-label image retrieval, using class semantic embeddings as a link between visual features and binary codes to align visual features with corresponding class semantics and simultaneously transfer knowledge from seen classes to unseen classes. Meanwhile, single-label zero-shot learning employs Generative Adversarial Networks (GAN) to synthesize class-specific characteristics from the corresponding class attribute embeddings, achieving encouraging results. However, synthesizing multi-label features from GANs remains unexplored in the context of zero-shot hashing settings. When multiple objects co-appear in an image, a key issue is how to fuse multi-class information effectively. In this paper, by introducing the multi-class information fusion strategy, a novel Generative Zero-Shot Hashing (GZSH) is proposed to transform zero-shot hashing into traditional supervised hashing by generating features of unseen categories. Specifically, this study introduces three different fusion methods (attribute-based fusion, feature-based fusion, and semantic-enhanced fusion) for synthesizing multi-label features from the corresponding multi-label class embeddings. Subsequently, the semantic-enhanced fusion is integrated into the representative generative architecture to learn the feature distribution of unlabeled images in the context of multi-label zero-shot hashing. Besides, by jointly learning the seen/source and unseen/target samples, a pairwise similarity loss is introduced to optimize the deep hashing model. Extensive experimentation on three widely used multi-label datasets illustrates the outstanding performance of our proposed GZSH framework compared to current state-of-the-art methods. The implementation code for our GZSH framework can be found at https://github.com/QinLab-WFU/GZSH.

Deep hashing techniques are widely used in remote sensing image retrieval due to their fast retrieval and low cost. However, existing multi-label retrieval methods encounter two main challenges. First, the input sample pixels commonly include irrelevant backgrounds, degrading semantic feature representation. Second, traditional vision transformer models directly adopt fine-grained feature extraction, which generates redundant long sequences and significantly increases computational costs. To address the mentioned problems, a Deep Semantic-attention Proxy Hashing framework (DSaPH) is proposed for multi-label remote sensing image retrieval. Specifically, to achieve the dynamic feature extraction of images, a coarse-to-fine representation module with two stages of coarse segmentation and fine segmentation is proposed, which identifies the object-related patches in the coarse-grained segmentation through critical region detection and aggregates the global features of the original image in combination with the feature reuse mechanism. To exclude background interference and focus on object information, the object-focus loss highlights the object information through Grad-CAM and generates object-fused image-focus hash codes. Furthermore, class-related proxy learning is introduced to use proxy hash codes to supervise the semantic representation of object-fused hash codes, and embed proxy hash codes and object-fused hash codes into a unified Hamming space to capture the Hamming similarity of neighboring multi-label data. Extensive experimental evaluations on three benchmark datasets show that our proposed DSaPH framework achieves excellent performance in multi-label remote sensing image retrieval. The code for our DSaPH framework is publicly available at https://github.com/QinLab-WFU/DSaPH.git.

Reference-based super-resolution (RefSR) leverages an additional high-resolution (HR) image as supplementary information to super-resolve a low-resolution (LR) image, generally yielding superior results over single image super-resolution (SISR). Nevertheless, existing RefSR methods still struggle with accurate reference texture transfer, especially under geometric transformations between the input LR and reference images. This issue mainly arises from the improper design of current coupled transfer frameworks, which perform alignment and aggregation jointly within a single step. The resulting degradation gap and disproportion gap impair alignment and aggregation, respectively, while their coupling further aggravates the transfer difficulty through mutual interference. This motivates us to propose a novel RefSR framework with geometry-aware transfer, termed GeoRef, which explicitly decouples texture transfer process to borrow realistic texture across geometric transformations. For the degradation gap, we propose a progressive alignment network that leverages more detailed guidance from SISR and HR images to incrementally refine textures in a coarse-to-fine manner, thereby producing faithful details while suppressing incongruous artifacts. For the disproportion gap, we design a selective aggregation network that adaptively transfers multiple correlated textures according to the object scale ratio between the two images, thereby complementing more relevant information. Additionally, extensive experiments on both reference-based and single image datasets demonstrate that our GeoRef framework achieves competitive performance with state-of-the-art methods in both quantitative and qualitative comparisons. The source codes are available at https://github.com/huanglab-research/GeoRef.