Visible-infrared person re-identification (VI-ReID) is a challenging cross-modality image retrieval task. Compared to visible modality that handles only the intra-modality discrepancy, VI-ReID suffers from an additional gap. Most existing methods achieve promising accuracy in supervised setting, but high annotation cost limits their scalability real-world scenarios. Although few unsupervised already exist, they typically rely on initialization and instance selection, despite computational...

RGB (visible), near-infrared (NI), and thermal infrared (TI) imaging modalities are commonly combined for round-the-clock surveillance. We introduce a novel unsupervised multi-modality person re-identification (MM-ReID) task, which, based on an individual's image in any one modality, seeks to identify matches the other two modalities. Compared prior MM-ReID problem formulations, significantly reduces labeling cost constraints. To address we propose inter-modality similarity learning (IMSL)...

Fully unsupervised person re-identification (ReID) methods aim to learn discriminative features without using labeled ReID data. Because these are easily affected by camera discrepancies, similar studies have typically designed optimization enable the model camera-invariant features. However, they often ignore impact of discrepancies on clustering results. Specifically, will reduce intra-class diversity and promote generation noise labels. To solve above problems, we propose a unified...

The person re-identification (ReID) method in a single-domain achieves appealing performance, but its reliance on label information greatly limits extensibility. Therefore, the unsupervised cross-domain ReID has received extensive attention. Its purpose is to optimize model by using labelled source domain and unlabelled target finally make well generalized domain. We propose an based median stable clustering (MSC) global distance classification (GDC). Specifically, measurement used MSC...

<p>Visible-infrared person re-identification (VI-ReID) is a challenging cross-modality image retrieval task. Compared to visible modality that handles only the intra-modality discrepancy, VI-ReID suffers from an additional gap. Most existing methods achieve promising accuracy in supervised setting, but high annotation cost limits their scalability real-world scenarios. Although few unsupervised already exist, they typically rely on initialization and instance selection, despite...

In the cross-domain person re-identification (ReID) method based on clustering, performance of model depends heavily quality information it obtains from clustering. To improve reliability clustering obtained by model, we propose a biclustering collaborative learning (BCL) framework derived an identity disentanglement adaptation network (IDA-Net). IDA-Net encodes and style input image transfers premise maintaining consistency. By comparing results same dataset before after transfer process,...

With the continuous development of deep learning, performance intelligent diagnosis system for ocular fundus diseases has been significantly improved, but during training process, problems like lack samples and uneven sample distribution (the number disease is much smaller than normal samples) have become increasingly prominent. In view previous issues, this paper proposes a method generating images based on “Combined GAN” (Com-GAN), which can generate both with hard exudates, so that be...

Abstract Privacy protection in the computer vision field has attracted increasing attention. Generative adversarial network-based methods have been explored for identity anonymization, but they do not take into consideration semantic information of images, which may result unrealistic or flawed facial results. In this paper, we propose a Semantic-aware De-identification Adversarial Network (SDGAN) model anonymization. To retain expression effectively, extract image using edge-aware graph...

Emerging deep learning (DL) techniques have greatly improved pedestrian reidentification (PRI) performance. However, the existing DL-based PRI methods cannot learn robust feature representations owing to single view of query images and limited number extractable features. Inspired by generative adversarial networks (GANs), this paper proposes a novel method based on multiview GAN (PmGAN) classification recognition network (CRN). The PmGAN consists three generators one multiclass...

The current unsupervised domain adaptation person re-identification (re-ID) method aims to solve the shift problem and applies prior knowledge learned from labelled data in source unlabelled target for re-ID. At present, re-ID based on pseudolabels has obtained state-of-the-art performance. This obtains via a clustering algorithm uses these optimize CNN model. Although it achieves optimal performance, model cannot be further optimized due existence of noisy labels process. In this paper, we...

Active learning aims to select the most valuable unlabelled samples for annotation. In this paper, we propose a redundancy removal adversarial active (RRAAL) method based on norm online uncertainty indicator, which selects their distribution, uncertainty, and redundancy. RRAAL includes representation generator, state discriminator, module (RRM). The purpose of generator is learn feature sample, discriminator predicts vector after concatenation. We added sample improve ability designed...

<p>Visible-infrared person re-identification (VI-ReID) is a challenging cross-modality image retrieval task. Compared to visible modality that handles only the intra-modality discrepancy, VI-ReID suffers from an additional gap. Most existing methods achieve promising accuracy in supervised setting, but high annotation cost limits their scalability real-world scenarios. Although few unsupervised already exist, they typically rely on initialization and instance selection, despite...

Abstract The core idea of active learning is to obtain higher model performance with less annotation cost. This paper proposes an independency‐enhancing adversarial method. Independency‐enhancing different from the previous methods and pays more attention sample independence. Specifically, it believed that informativeness a group samples related independence rather than simple sum each in group. Therefore, independent selection module based on hierarchical clustering designed ensure An...