Unsupervised multimodal semantic discovery aims to learn discriminative representations from multimodal data. However, existing methods suffer from two key limitations. First, they only align instances across modalities without modeling semantic-level consistency, which fails to mitigate semantic bias caused by the gaps among feature distributions of multiple modalities. Second, they inevitably generate incorrect negative pairs during contrastive learning, pushing semantically similar samples apart. To address these challenges, we propose GLAD (Global and Local semantic Alignment for unsupervised multimodal semantic Discovery), which aligns multimodal data at both global and local semantic levels. At the global level, GSA integrates multi-modal features into a shared space and employs joint clustering via optimal transport to capture common semantic patterns while mitigating cross-modality semantic bias. At the local level, LSA adaptively weights samples within each cluster based on their semantic importance, alleviating the effect of incorrect negative pairs. Through the joint optimization of GSA and LSA, GLAD effectively captures both the global semantic structure and the local semantic nuances of multimodal data. Extensive experiments on three benchmark datasets demonstrate GLAD significantly outperforms state-of-the-art methods, with an average improvement of 3.22%.

Multi-relational graph clustering aims to uncover complex node interactions by leveraging multiple relational views, yet existing methods often suffer from two key limitations: they assume equal importance across views and decouple representation learning from clustering, both of which hinder overall performance. To address these issues, we propose OMC-DVM, a novel end-to-end Online Multi-Relational Graph Clustering With Dominant View Mining framework. OMC-DVM introduces two core innovations: (1) A unsupervised dominant view mining module that dynamically identifies the dominant view using Maximum Mean Discrepancy (MMD) and adaptively aligns other views to it, mitigating view imbalance. (2) An online ,multi-relational clustering process that unifies representation learning and clustering into a single stage. By performing clustering-level contrastive learning , OMC-DVM directly generates cluster assignments in an end-to-end manner. Extensive experiments on both real-world and synthetic benchmark datasets demonstrate that OMC-DVM not only achieves state-of-the-art clustering performance but also effectively alleviates the view imbalance problem in multi-relational graphs.

Contrastive multi-view clustering is constrained by simplistic positive-pair construction and vulnerability to false negatives. Existing methods mechanically align pre-defined pairs without exploring semantic neighborhoods or generating challenging augmentations. To address these limitations, we propose Adversarial Neighborhood Contrastive Learning (AdvNCL), which leverages K-nearest neighbor structures in the representation space to generate semantically consistent adversarial augmentations. Specifically, AdvNCL employs Mamba networks for dynamic view fusion and introduces a neighborhood-constrained adversarial loss that identifies perturbations maximally disrupting instance-level discrimination while preserving cluster consistency. Experiments on four benchmark datasets (Hdigit, Cifar100, Prokaryotic, Wiki) demonstrate that AdvNCL outperforms six state-of-the-art methods, achieving average improvements of +3.2% in ACC and + 2.8% in NMI. Notably, on the Prokaryotic dataset, AdvNCL improves ACC by 12.5% over the best baseline.

Multi-label text classification (MLTC) is a challenging task where each document can be associated with multiple interdependent labels. This task is complicated by two key issues: the intricate correlations among labels and the partial overlap between labels and text relevance. Existing methods often fail to capture the semantic dependencies between labels or struggle to handle the ambiguities caused by partial overlaps, resulting in suboptimal representation learning. To address these challenges, we propose the Unified Contextual and Label-Aware Framework (UCLAF), which integrates a Label Attention Aware Network(LAN) and Correlation Aware Contrastive Learning (CACL) in a synergistic design. The Label Attention Aware Network explicitly models label dependencies by embedding labels and texts into a shared semantic space, aligning text representations with label semantics. Meanwhile, Correlation Aware Contrastive Learning refines these representations by dynamically modeling sample-level relationships, leveraging a contrastive loss function that accounts for the proportional overlap of labels between samples. This complementary approach enables UCLAF to jointly address complex label correlations and partial label overlaps. Extensive experiments on benchmark datasets demonstrate that UCLAF significantly outperforms state-of-the-art methods, showcasing its effectiveness in improving both representation learning and classification performance in MLTC tasks. We will release our code after the paper is accepted.

The limited accuracy of small object detection significantly constrains the applicability of popular object detection methods within complex industrial contexts. Current research aims to improve the model’s attention to small objects and improve their detection accuracy by incorporating attention mechanisms such as spatial and channel attention. However, these approaches overlook the issues of sparsity and loss of small object features during the feature extraction phase, primarily due to their low pixel ratio and susceptibility to occlusion. Moreover, existing attention mechanisms are designed to emphasize important features, with limited focus on small objects. To address these challenges, we propose EAS-YOLOv5 to enhance the attention of YOLOv5 on small objects, which uses a cross-level feature fusion method guided by high-level features that leverages the relationships between large and small objects to generate the model’s attention on small objects. Additionally, we enhance small object features by integrating detail texture extraction module and context relation extraction module. Using the combined effects of these methods, our approach substantially improves the small object detection performance. In a dataset derived from the real industrial gas cylinder deployment environment, compared to the vanilla YOLOv5s, which achieved mAP and mAPs of 0.517 and 0.371, respectively, our proposed EAS-YOLOv5 achieved mAP and mAPs of 0.531 and 0.505, indicating an approximately 36% improvement in mAPs and markedly enhancing small object detection capabilities.

With the rapid development of deep learning, convolutional neural network (CNN)-based defocus depth recovery techniques have gained significant attention. However, a critical issue observed in the recovered depth maps is edge blurring. This problem arises because standard convolution operations fail to account for depth variations in 3D space, treating physically adjacent pixels in 2D images as having consistent depth. As convolutional kernels extract features within their receptive fields, they inadvertently introduce correlations between pixels with depth disparities, leading to blurred regions and degraded edge quality. To address this limitation, we propose an adaptive convolution (AdpConv) mechanism that dynamically adjusts kernel weights based on local depth discrepancies. By minimizing the interference from pixels with depth disparities, our method enhances the accuracy of defocus measurement and improves depth recovery performance. The proposed approach is generalizable, applicable to both 2D and 3D convolution operations, and can be seamlessly integrated into existing models as a plug-and-play component. Extensive experiments demonstrate that our method achieves superior quantitative metrics and significantly improves depth prediction, particularly in challenging textured regions. This work advances the state-of-the-art in defocus depth recovery and offers a versatile solution for related computer vision tasks involving depth estimation and image restoration.