Stochastic Gradient Perturbation: An Implicit Regularizer for Person Re-Identification

Article Computer Science, Artificial Intelligence

MHSA-Net: Multihead Self-Attention Network for Occluded Person Re-Identification

Hongchen Tan et al.

Summary: This article presents a novel person reidentification model, MHSA-Net, which prunes unimportant information and captures key local information to improve performance.

IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS (2023)

添加到收藏夹

Article Computer Science, Information Systems

Adversarial and Isotropic Gradient Augmentation for Image Retrieval With Text Feedback

Fuxiang Huang et al.

Summary: This paper introduces the research topic of image retrieval with text feedback (IRTF) and proposes a regularization approach using gradient augmentation to address the issues of model overfitting and low diversity of training data. Experimental results show that the proposed method achieves better performance on two datasets.

IEEE TRANSACTIONS ON MULTIMEDIA (2023)

添加到收藏夹

Article Engineering, Electrical & Electronic

Incomplete Descriptor Mining With Elastic Loss for Person Re-Identification

Hongchen Tan et al.

Summary: In this paper, a novel person Re-ID model called Consecutive Batch DropBlock Network (CBDB-Net) is proposed to capture attentive and robust person descriptors. The CBDB-Net incorporates the Consecutive Batch DropBlock Module (CBDBM) and Elastic Loss (EL) to improve the Re-ID performance. Experimental results show that the CBDB-Net achieves competitive performance on various person Re-ID datasets.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY (2022)

添加到收藏夹

Article Engineering, Electrical & Electronic

Adversarial Decoupling and Modality-Invariant Representation Learning for Visible-Infrared Person Re-Identification

Weipeng Hu et al.

Summary: In this paper, a novel adversarial decoupling and modality-invariant representation learning method is proposed for visible-infrared person re-identification (RGB-IR ReID). By decoupling domain-related features and identity-related features, and orthogonal decorrelation between them, the method effectively separates identity information and domain information for cross-modality pedestrians, improving the accuracy of re-identification.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY (2022)

添加到收藏夹

Article Computer Science, Artificial Intelligence

Collaborative Refining for Person Re-Identification With Label Noise

Mang Ye et al.

Summary: Existing person re-identification methods rely on large-scale annotated training data, but label noise is unavoidable. This paper proposes a method to learn a robust Re-ID model by jointly optimizing labels and networks, using a large learning rate prefatory model and self-label refining strategy. An online co-refining framework is introduced for dynamic mutual learning, along with a favorable selective consistency strategy to reduce the negative impact of noisy labels.

IEEE TRANSACTIONS ON IMAGE PROCESSING (2022)

添加到收藏夹

Article Computer Science, Theory & Methods

Person Re-Identification by Context-Aware Part Attention and Multi-Head Collaborative Learning

Dongming Wu et al.

Summary: In this study, a novel multi-level Context-aware Part Attention (CPA) model is proposed to address the video-based person re-identification problem. By capturing the global relationship and considering multi-level features, the model is able to extract discriminative and robust local part features. Additionally, a multi-head collaborative training scheme is introduced to improve the performance by considering consistency at both multi-head and multi-frame levels.

IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY (2022)

添加到收藏夹

Article Computer Science, Theory & Methods

Dynamic Tri-Level Relation Mining With Attentive Graph for Visible Infrared Re-Identification

Mang Ye et al.

Summary: In this paper, a novel dynamic tri-level relation mining (DTRM) framework is proposed to improve the performance of visible infrared person re-identification (VI-ReID) by simultaneously exploring channel-level, part-level intra-modality, and graph-level cross-modality relation cues. The paper also introduces an intra-modality weighted-part attention (IWPA) and a cross-modality graph structured attention (CGSA) to address misalignment and noise problems in person images. A dynamic aggregation strategy is designed to seamlessly integrate the two components.

IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY (2022)

添加到收藏夹

Article Computer Science, Artificial Intelligence

Augmentation Invariant and Instance Spreading Feature for Softmax Embedding

Mang Ye et al.

Summary: This paper investigates the problem of unsupervised embedding learning and proposes a novel instance-wise softmax embedding method for learning feature representations without using category labels. The research faces challenges in mining reliable positive supervision from highly similar classes and generalizing to unseen testing categories.

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE (2022)

添加到收藏夹

Article Computer Science, Artificial Intelligence

Deep Learning for Person Re-Identification: A Survey and Outlook

Mang Ye et al.

Summary: Person re-identification (Re-ID) has gained significant interest in the computer vision community, with the advancement of deep neural networks. It is categorized into closed-world and open-world settings. While closed-world setting has achieved inspiring success, the research focus has shifted to the more challenging open-world setting. We summarize the open-world Re-ID in five different aspects and introduce a new evaluation metric. This metric provides an additional criteria for evaluating Re-ID systems in real applications.

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE (2022)

添加到收藏夹

Article Engineering, Electrical & Electronic

Illumination Unification for Person Re-Identification

Guoqing Zhang et al.

Summary: In this paper, a novel method for person re-identification under different illumination conditions is proposed. The method estimates the illumination scale of testing images and restores them to the illumination scale of training images to reduce disparities, achieving promising results on illumination-adaptive datasets.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY (2022)

添加到收藏夹

Article Engineering, Electrical & Electronic

Cross-Modal Cross-Domain Dual Alignment Network for RGB-Infrared Person Re-Identification

Xiaowei Fu et al.

Summary: This paper presents a dual alignment network for solving the RGB-Infrared cross-modal cross-domain person Re-ID problem. The network consists of three components that improve the model performance by learning domain-invariant and modality-invariant person representations.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY (2022)

添加到收藏夹

Proceedings Paper Computer Science, Artificial Intelligence

Modality Synergy Complement Learning with Cascaded Aggregation for Visible-Infrared Person Re-Identification

Yiyuan Zhang et al.

Summary: Visible-Infrared Re-Identification is a challenging task due to the modality discrepancy. This paper proposes a novel framework, MSCLNet, that synergizes two modalities to construct diverse representations and complements them to improve performance. The Cascaded Aggregation strategy is used to optimize the feature distribution. Experimental results demonstrate the superiority of MSCLNet over existing methods.

COMPUTER VISION - ECCV 2022, PT XIV (2022)

添加到收藏夹

Proceedings Paper Computer Science, Artificial Intelligence

Enhancing Adversarial Training with Second-Order Statistics of Weights

Gaojie Jin et al.

Summary: This paper proposes an enhanced adversarial training method by treating model weights as random variables and optimizing the second-order statistics of weights. Theoretical analysis and experimental results demonstrate that this method can effectively improve the robustness and generalization performance of neural networks.

2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2022) (2022)

添加到收藏夹

Proceedings Paper Computer Science, Artificial Intelligence

Bounded Adversarial Attack on Deep Content Features

Qiuling Xu et al.

Summary: This paper proposes a novel adversarial attack method that achieves fine-grained regulation of content feature mutations with bounded perceptual variations through personalized distribution quantile bounds and polynomial barrier loss function for individual neurons in deep layers. The evaluation on ImageNet dataset and five different model architectures demonstrates the effectiveness of the attack, achieving the state-of-the-art trade-off between attack success rate and imperceptibility.

2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2022) (2022)

添加到收藏夹

Proceedings Paper Computer Science, Artificial Intelligence

Enhancing Adversarial Robustness for Deep Metric Learning

Mo Zhou et al.

Summary: This paper introduces a hardness manipulation method for improving the adversarial robustness of deep metric learning models by setting a specified level of hardness. The method is flexible and can gradually increase the hardness level during training to balance performance and robustness. Additionally, an intra-class structure loss term is used to further enhance model efficiency and robustness.

2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2022) (2022)

添加到收藏夹

Proceedings Paper Computer Science, Artificial Intelligence

Self-Distillation from the Last Mini-Batch for Consistency Regularization

Yiqing Shen et al.

Summary: The paper introduces an efficient and reliable self-distillation framework named DLB, which not only improves generalization ability, but also exhibits robustness and stability, easy implementation, and no additional modifications required. Experimental results show that the proposed method outperforms state-of-the-art approaches.

2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR) (2022)

添加到收藏夹

Proceedings Paper Computer Science, Artificial Intelligence

Hyperspherical Consistency Regularization

Cheng Tan et al.

Summary: This study explores the relationship between self-supervised learning and supervised learning and proposes a new method to avoid classifier bias. Experimental results demonstrate the superior performance of this method in semi-supervised and weakly-supervised learning.

2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR) (2022)

添加到收藏夹

Article Computer Science, Artificial Intelligence

Deeply Supervised Discriminative Learning for Adversarial Defense

Aamir Mustafa et al.

Summary: Deep neural networks are vulnerable to attacks with small perturbations, but can be made more robust by disentangling feature representations by class. This approach improves model resilience against adversarial attacks without sacrificing classification performance.

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE (2021)

添加到收藏夹

Article Computer Science, Artificial Intelligence

Adversarial Metric Attack and Defense for Person Re-Identification

Song Bai et al.

Summary: Recent research has shown that current distance metrics are highly vulnerable to adversarial examples, which may increase security risks in commercial re-identification systems used in video surveillance. Adversarial examples have been rarely studied in metric analysis, possibly due to the natural gap between training and testing. The proposed Adversarial Metric Attack method demonstrates adversarial effects in re-ID systems and presents an early attempt at training a metric-preserving network.

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE (2021)

添加到收藏夹

Proceedings Paper Computer Science, Artificial Intelligence

IDM: An Intermediate Domain Module for Domain Adaptive Person Re-ID

Yongxing Dai et al.

Summary: This study introduces an Intermediate Domain Module (IDM) to tackle the task of unsupervised domain adaptive person re-identification. By incorporating bridge losses and diversity loss, the method ensures that the generated intermediate domains effectively bridge the source and target domains while preventing overfitting.

2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2021) (2021)

添加到收藏夹

Proceedings Paper Computer Science, Artificial Intelligence

Cross-Modality Person Re-Identification via Modality Confusion and Center Aggregation

Xin Hao et al.

Summary: This paper proposes a novel Modality Confusion Learning Network (MCLNet) that learns modality-invariant features by confusing modalities and maximizing cross-modality similarity in a single framework. By introducing identity-aware marginal center aggregation strategy and camera-aware learning scheme, the performance is further enhanced. Extensive experiments show that MCLNet outperforms the state-of-the-art significantly.

2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2021) (2021)

添加到收藏夹

Proceedings Paper Computer Science, Artificial Intelligence

Dual Attention Suppression Attack: Generate Adversarial Camouflage in Physical World

Jiakai Wang et al.

Summary: This study introduces the Dual Attention Suppression (DAS) attack to generate visually-natural physical adversarial camouflages with strong transferability by suppressing both model and human attention.

2021 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, CVPR 2021 (2021)

添加到收藏夹

Proceedings Paper Computer Science, Artificial Intelligence

Generalizable Person Re-identification with Relevance-aware Mixture of Experts

Yongxing Dai et al.

Summary: The study introduces a novel method, RaMoE, for domain generalizable person re-identification using a relevance-aware mixture of experts. By dynamically leveraging the diverse characteristics of source domains, the model's generalization performance is enhanced.

2021 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, CVPR 2021 (2021)

添加到收藏夹

Article Computer Science, Artificial Intelligence

Deep High-Resolution Representation Learning for Cross-Resolution Person Re-Identification

Guoqing Zhang et al.

Summary: In this paper, a Deep High-Resolution Pseudo-Siamese Framework (PS-HRNet) is proposed to tackle the Cross-Resolution Person Re-identification problem by improving VDSR, designing HRNet-ReID, and introducing a pseudo-siamese framework. Experimental results on five cross-resolution person datasets demonstrate the effectiveness of the proposed approach in improving Rank-1 accuracy on different datasets.

IEEE TRANSACTIONS ON IMAGE PROCESSING (2021)

添加到收藏夹

Article Computer Science, Theory & Methods

Visible-Infrared Person Re-Identification via Homogeneous Augmented Tri-Modal Learning

Mang Ye et al.

Summary: The paper proposes a Homogeneous Augmented Tri-Modal (HAT) learning method for matching person images between daytime visible modality and nighttime infrared modality, which significantly outperforms the current state-of-the-art methods by generating a grayscale auxiliary modality to enforce structure relations across multiple modalities.

IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY (2021)

添加到收藏夹

Article Computer Science, Artificial Intelligence

Dual-Refinement: Joint Label and Feature Refinement for Unsupervised Domain Adaptive Person Re-Identification

Yongxing Dai et al.

Summary: In this study, a novel approach called Dual-Refinement is proposed to jointly refine pseudo labels and features in both offline clustering and online training phases, improving the label purity and feature discriminability in the target domain for more reliable re-ID. This method effectively reduces the influence of noisy labels and refines learned features within the alternative training process, outperforming state-of-the-art methods by a large margin according to experiments.

IEEE TRANSACTIONS ON IMAGE PROCESSING (2021)

添加到收藏夹

Article Engineering, Electrical & Electronic