Recently, part information of pedestrian images has been demonstrated to be effective for person re-identification (ReID), but the part interaction is ignored when using Transformer to learn long-range dependencies. In this paper, we propose a novel transformer network named Completed Part Transformer (CPT) for person ReID, where we design the part transformer layer to learn the completed part interaction. The part transformer layer includes the intra-part layer and the part-global layer, where they consider long-range dependencies from the aspects of the intra-part interaction and the part-global interaction, simultaneously. Furthermore, in order to overcome the limitation of fixed number of the patch tokens in the transformer layer, we propose the Adaptive Refined Tokens (ART) module to focus on learning the interaction between the informative patch tokens in the pedestrian image, which improves the discrimination of the pedestrian representation. Extensive experimental results on four person ReID datasets, i.e., MSMT17, Market1501, DukeMTMC-reID and CUHK03, demonstrate that the proposed method achieves a new state-of-the-art performance, e.g., it achieves 68.0% mAP and 84.6% Rank-1 accuracy on MSMT17.
Power grid operation occurs in complex, dynamic environments where the timely identification of operator violations is essential for safety. Traditional methods often rely on manual supervision and rule-based detection, leading to inefficiencies. Existing deep learning approaches, while powerful, require fully labeled data and long training times, thereby increasing costs. To address these challenges, we propose an active hard sample learning method specifically for the violation action recognition of operators in power grid operation. We design a hard instance sampling module with multi-strategy fusion based on active learning to improve training efficiency. This module identifies hard samples based on the consistency of models or samples, where we develop uncertainty evaluation and the instance discrimination strategy to assess the contributions of samples effectively. We utilize ResNet50 and ViT architectures with Faster-RCNN for detection and recognition, developed using PyTorch 2.0. The dataset comprises 2000 samples, and 30% and 60% labeled data are employed. Experimental results show significant improvements in model performance and training efficiency, demonstrating the method’s effectiveness in complex power grid environments. Our approach enhances safety monitoring and advances active learning and hard sample techniques in practical applications.
15 October 2025 in IEEE Internet of Things Journal
With the proliferation of the visual Internet of Things (VIoT), the intelligent processing of imagery and textual data to accomplish text-to-image person re-identification (Re-ID) constitutes a formidable challenge within VIoT, where text-to-image person Re-ID endeavors to execute image retrieval across disparate visual sensors through a designated text query. The existing works mainly focus on learning feature alignment between texts and images from global and local perspectives. However, these methods exhibit two fundamental limitations: 1) neglecting the relationship between different kinds of positive pairs (i.e., original positive pairs and weak positive pairs) during global feature alignment and 2) employing horizontally partitioned stripe parts of pedestrian images for local feature alignment constrains adaptability. To address these issues, we propose a novel cross-modal alignment enhancement network (AENet), which effectively aligns cross-modal features by modeling the relationships among different positive pairs and enabling flexible local correspondences. Specifically, we design a hybrid positive pair learning (HPPL) strategy to fully exploit the complementary semantic information in original and weak positive pairs through adaptive fusion, thereby enhancing global feature alignment. Regarding local feature alignment, we develop a local aggregation and dynamic matching (LADM) module to learn fine-grained correspondences through generating representative cross-modal local aggregation features and implementing a dynamic matching mechanism. Correspondingly, we propose a token reconstruction loss (TRL) to regulate the local aggregation process, thereby ensuring the quality of local aggregation features. Extensive experiments on three publicly available benchmarks verify the effectiveness and superiority of the proposed method, exemplified by achieving 76.48%, 68.34%, and 66.70% Rank-1 accuracy on CUHK-PEDES, ICFG-PEDES, and RSTPReid, respectively.
28 November 2025 in IEEE Signal Processing Letters
The rapid development and malicious use of deepfakes pose a significant crisis of trust. To cope with the evolving deepfake technologies, an increasing number of detection methods adopt the continual learning paradigm, but they often suffer from catastrophic forgetting. Although replay-based methods mitigate this issue by storing a portion of samples from historical tasks, their sample selection strategies usually rely on a single metric, which may lead to the omission of critical samples and consequently hinder the construction of a robust instance memory bank. In this letter, we propose a novel Multi-perspective Sample Selection Mechanism (MSSM) for continual deepfake detection, which jointly evaluates prediction error, temporal instability, and sample diversity to preserve informative and challenging samples in the instance memory bank. Furthermore, we design a Hierarchical Prototype Generation Mechanism (HPGM) that constructs prototypes at both the category and task levels, which are stored in the prototype memory bank. Extensive experiments under two evaluation protocols demonstrate that the proposed method achieves state-of-the-art performance.