Instance-level Adversarial Source-free Domain Adaptive Person Re-identification

Source-free Style-diversity Adversarial Domain Adaptation with Privacy-preservation for person re-identification

Domain adaptation assumes that samples from source and target domains are freely accessible during a training phase. However, such an assumption is rarely plausible in the real world and possibly causes data privacy issues, especially when the label of the source domain can be a sensitive attribute as an identifier. To avoid accessing source data that could contain sensitive information, we introduce source data free domain adaptation (SFDA). Our key idea is to leverage a pretrained model from the source domain and progressively update the target model in a self-learning manner. We observe that target samples with lower self-entropy measured by the pretrained source model are more likely to be classified correctly. From this, we select the reliable samples with the self-entropy criterion and define these as class prototypes. We then assign pseudolabels for every target sample based on the similarity score with class prototypes. We further propose point-to-set distance-based filtering, which does not require any tunable hyperparameters to reduce uncertainty from the pseudolabeling process. Finally, we train the target model with the filtered pseudolabels with regularization from the pretrained source model. Surprisingly, without direct usage of labeled source samples, our SFDA outperforms conventional domain adaptation methods on benchmark datasets. <p xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><b><i>Impact Statement</i></b>—This study addresses the data privacy issue, especially in unsupervised domain adaptation. Based on our privacy-preserving domain adaptation, various stakeholders, including enterprises and government organizations, can be free of concern about privacy issues with their labeled source dataset. Furthermore, the proposed data-free approach can contribute to creating a positive social impact, especially in large-scale datasets. Recently, since the size of data across various fields has been scaling up, it is almost incapable for individual researchers to directly utilize a large scale of data during training. For this reason, a new social trend of sharing pretrained models, e.g., EfficientNet and BERT, led by global enterprises with their huge amount of resources has been rising up. From this viewpoint, our approach thus enables more people to participate in the domain adaptation field specifically when the source data are large scale and contain sensitive attributes.

Domain adaptation has achieved promising results in many areas, such as image classification and object recognition. Although a lot of algorithms have been proposed to solve the task with different domain distributions, it remains a challenge for multi-source unsupervised domain adaptation. In addition, most of the existing algorithms learn a classifier on the source domain and predict the labels for the target data, which indicates that only the knowledge derived from the hyperplane is transferred to the target domain and the structure information is ignored. In light of this, we propose a novel algorithm for multi-source unsupervised domain adaptation. Generally speaking, we aim to preserve the whole structure from source domains and transfer it to serve the task on the target domain. The source and target data are put together for clustering, which simultaneously explores the structures of the source and target domains. The structure-preserved information from source domain further guides the clustering process on the target domain. Extensive experiments on two widely used databases on object recognition and face identification show the substantial improvement of our proposed approach over several state-of-the-art methods. Especially, our algorithm can take use of multi-source domains and achieve robust and better performance compared with the single source domain adaptation methods.

Domain adaptation (DA), which leverages labeled data from related source domains, comes in handy when the label information of the target domain is scarce or unavailable. However, as the source data do not come from the same origin as that of the target domain, the predictive distributions of the source and target domains are likely to differ in reality. At the extreme, the predictive distributions of the source domains can differ completely from that of the target domain. In such case, using the learned source classifier to assist in the prediction of target data can result in prediction performance that is poorer than that with the omission of the source data. This phenomenon is established as negative transfer with impact known to be more severe in the multiclass context. To combat negative transfer due to differing predictive distributions across domains, we first introduce the notion of positive transferability for the assessment of synergy between the source and target domains in their prediction models, and we also propose a criterion to measure the positive transferability between sample pairs of different domains in terms of their prediction distributions. With the new measure, a predictive distribution matching (PDM) regularizer and a PDM framework learn the target classifier by favoring source data with large positive transferability while inferring the labels of target unlabeled data. Extensive experiments are conducted to validate the performance efficacy of the proposed PDM framework using several commonly used multidomain benchmark data sets, including Sentiment, Reuters, and Newsgroup, in the context of both binary-class and multiclass domains. Subsequently, the PDM framework is put to work on a real-world scenario pertaining to water cluster molecule identification. The experimental results illustrate the adverse impact of negative transfer on several state-of-the-art DA methods, whereas the proposed framework exhibits excellent and robust predictive performances.

A review on video person re-identification based on deep learning

Two-phase self-supervised pretraining for object re-identification

Domain adaptation has become an attractive learning paradigm, as it can leverage source domains with rich labels to deal with classification tasks in an unlabeled target domain. A few recent studies develop domain adaptation approaches for graph-structured data. In the case of node classification task, current domain adaptation methods only focus on the closed-set setting, where source and target domains share the same label space. A more practical assumption is that the target domain may contain new classes that are not included in the source domain. Therefore, in this paper, we introduce a novel and challenging problem for graphs, i.e., open-set domain adaptive node classification, and propose a new approach to solve it. Specifically, we develop an algorithm for efficient knowledge transfer from a labeled source graph to an unlabeled target graph under a separate domain alignment (SDA) strategy, in order to learn discriminative feature representations for the target graph. Our goal is to not only correctly classify target nodes into the known classes, but also classify unseen types of nodes into an unknown class. Experimental results on real-world datasets show that our method outperforms existing methods on graph domain adaptation.

Abstract. Person Re-Identification (Re-ID) is the task of matching the same person in multiple images captured by different cameras. Recently deep learning-based Re-ID algorithms demonstrated an exciting progress for terrestrial-based cameras, still, person Re-ID in aerial images poses multiple challenges including occlusion of human feature parts, image distortion, and dynamic camera location. In this paper, we propose a new Person Aerial Re-ID framework Robust to Occlusion and Thermal imagery (ParrotGAN). Our model is focused on cross-modality person Re-ID in aerial images. Furthermore, we collected a new large-scale synthetic multimodal AerialReID dataset with 30k images and 137 person ID. Our ParrotGAN model leverages two strategies to achieve robust performance in the task of person Re-ID in thermal and visible range. Firstly, we use a latent space of StyleGAN2 model to estimate the distance between two images of a person. Specifically, we project each real image into the latent space with a correspondent latent vector z. We use the distance between latent vectors to provide a Re-ID similarity metric. Secondly, we use a generative-adversarial network to translate a color image to a synthetic thermal image. We use synthetic image for a cross-modality Re-ID. We evaluate our ParrotGAN model and basleines on our AerialReID and PRAI-1581 datasets. The results of the evaluation are encouraging and demonstrate that our ParrotGAN model competes with baselines in visible range aerial person Re-ID and outperforms them in the cross-modality setting. We made our code and dataset publicly available.

In the absence of the labeled samples in a domain referred to as target domain, Domain Adaptation (DA) techniques come in handy. Generally, DA techniques assume there are available source domains that share similar predictive function with the target domain. Two core challenges of DA typically arise, variance that exists between source and target domains, and the inherent source hypothesis bias. In this paper, we first propose a Stability Transfer criterion for selecting relevant source domains with low variance. With this criterion, we introduce a TARget learning Assisted by Source Classifier Adaptation (TARASCA) method to address the two core challenges that have impeded the performances of DA techniques. To verify the robustness of TARASCA, extensive experimental studies are carried out with comparison to several state-of-the-art DA methods on the real-world Sentiment and Newsgroups datasets, where various settings for the class ratios of the source and target domains are considered.

To address the challenges of data labeling difficulties, data privacy, and necessary large amount of labeled data for deep learning methods in diabetic retinopathy (DR) identification, the aim of this study is to develop a source-free domain adaptation (SFDA) method for efficient and effective DR identification from unlabeled data. A multi-SFDA method was proposed for DR identification. This method integrates multiple source models, which are trained from the same source domain, to generate synthetic pseudo labels for the unlabeled target domain. Besides, a softmax-consistence minimization term is utilized to minimize the intra-class distances between the source and target domains and maximize the inter-class distances. Validation is performed using three color fundus photograph datasets (APTOS2019, DDR, and EyePACS). The proposed model was evaluated and provided promising results with respectively 0.8917 and 0.9795 F1-scores on referable and normal/abnormal DR identification tasks. It demonstrated effective DR identification through minimizing intra-class distances and maximizing inter-class distances between source and target domains. The multi-SFDA method provides an effective approach to overcome the challenges in DR identification. The method not only addresses difficulties in data labeling and privacy issues, but also reduces the need for large amounts of labeled data required by deep learning methods, making it a practical tool for early detection and preservation of vision in diabetic patients.

Multisource unsupervised domain adaptation (MUDA) is an important and challenging topic for target classification with the assistance of labeled data in source domains. When we have several labeled source domains, it is difficult to map all source domains and target domain into a common feature space for classifying the targets well. In this article, a new progressive multisource domain adaptation network (PMSDAN) is proposed to further improve the classification performance. PMSDAN mainly consists of two steps for distribution alignment. First, the multiple source domains are integrated as one auxiliary domain to match the distribution with the target domain. By doing this, we can generally reduce the distribution discrepancy between each source and target domains, as well as the discrepancy between different source domains. It can efficiently explore useful knowledge from the integrated source domain. Second, to mine assistance knowledge from each source domain as much as possible, the distribution of the target domain is separately aligned with that of each source domain. A weighted fusion method is employed to combine the multiple classification results for making the final decision. In the optimization of domain adaption, weighted hybrid maximum mean discrepancy (WHMMD) is proposed, and it considers both the interclass and intraclass discrepancies. The effectiveness of the proposed PMSDAN is demonstrated in the experiments comparing with some state-of-the-art methods.

Most existing person re-identification(Re-ID) approaches achieve superior results based on the assumption that a large amount of pre-labelled data is usually available and can be put into training phrase all at once. However, this assumption is not applicable to most real-world deployment of the Re-ID task. In this work, we propose an alternative reinforcement learning based human-in-the-loop model which releases the restriction of pre-labelling and keeps model upgrading with progressively collected data. The goal is to minimize human annotation efforts while maximizing Re-ID performance. It works in an iteratively updating framework by refining the RL policy and CNN parameters alternately. In particular, we formulate a Deep Reinforcement Active Learning (DRAL) method to guide an agent (a model in a reinforcement learning process) in selecting training samples on-the-fly by a human user/annotator. The reinforcement learning reward is the uncertainty value of each human selected sample. A binary feedback (positive or negative) labelled by the human annotator is used to select the samples of which are used to fine-tune a pre-trained CNN Re-ID model. Extensive experiments demonstrate the superiority of our DRAL method for deep reinforcement learning based human-in-the-loop person Re-ID when compared to existing unsupervised and transfer learning models as well as active learning models.

In real-life scenarios, the accuracy of person re-identification (Re-ID) is subject to the limitation of camera hardware conditions and the change of image resolution caused by factors such as camera focusing errors. People call this problem cross-resolution person Re-ID. In this paper, we improve the recognition accuracy of cross-resolution person Re-ID by enhancing the image enhancement network and feature extraction network. Specifically, we treat cross-resolution person Re-ID as a two-stage task: the first stage is the image enhancement stage, and we propose a Super-Resolution Dual-Stream Feature Fusion sub-network, named SR-DSFF, which contains SR module and DSFF module. The SR-DSFF utilizes the SR module recovers the resolution of the low-resolution (LR) images and then obtains the feature maps of the LR images and super-resolution (SR) images, respectively, through the dual-stream feature fusion with learned weights extracts and fuses feature maps from LR and SR images in the DSFF module. At the end of SR-DSFF, we set a transposed convolution to visualize the feature maps into images. The second stage is the feature acquisition stage. We design a global-local feature extraction network guided by human pose estimation, named FENet-ReID. The FENet-ReID obtains the final features through multistage feature extraction and multiscale feature fusion for the Re-ID task. The two stages complement each other, making the final pedestrian feature representation has the advantage of accurate identification compared with other methods. Experimental results show that our method improves significantly compared with some state-of-the-art methods.

Unsupervised domain adaptation (UDA) in the task of person re-identification (re-ID) is highly challenging due to large domain divergence and no class overlap between domains. Pseudo-label based self-training is one of the representative techniques to address UDA. However, label noise caused by unsupervised clustering is always a trouble to self-training methods. To depress noises in pseudo-labels, this paper proposes a Noise Resistible Mutual-Training (NRMT) method, which maintains two networks during training to perform collaborative clustering and mutual instance selection. On one hand, collaborative clustering eases the fitting to noisy instances by allowing the two networks to use pseudo-labels provided by each other as an additional supervision. On the other hand, mutual instance selection further selects reliable and informative instances for training according to the peer-confidence and relationship disagreement of the networks. Extensive experiments demonstrate that the proposed method outperforms the state-of-the-art UDA methods for person re-ID.

Congratulations to Professor Helland on the elegant development of optimal partial least squares (PLS) regression, which reconceptualizes model reduction through the geometry of group actions and Hilbert spaces. Within this framework, PLS identifies a lower-dimensional subspace of functions that is provably optimal among all m $$ m $$ -dimensional reductions under specified covariance or spectral conditions. This rigorous notion of optimality naturally raises the question: can an analogous concept be established for domain adaptation (Ganin et al., 2016; He et al., 2024; Luo et al., 2025)., a technique that has recently attracted significant attention in modern machine learning? To make the context of domain adaptation concrete, consider a simple example. Suppose we want to classify images of cats and dogs. In the source domain S $$ S $$ , we observe clean images; in the target domain T $$ T $$ , the same cats and dogs appear, but all images are rotated by arbitrary angles. A classifier trained on S $$ S $$ may fail badly in T $$ T $$ , because the marginal distributions of pixels, say, f S $$ {f}_S $$ and f T $$ {f}_T $$ , differ drastically. Yet the domains share invariant predictive features, such as those extracted by convolutional pooling layers that ignore rotations. If we can find a representation map h : 𝒳 → 𝒵 that captures these invariants, then prediction becomes possible across domains. This rotation example highlights the central challenge of domain adaptation: although f S $$ {f}_S $$ and f T $$ {f}_T $$ differ, a representation h $$ h $$ may capture outcome-relevant invariants across domains. The key question is whether one can rigorously define an optimal representation, balancing predictive accuracy with invariance, analogous to how PLS identifies optimal subspaces in regression. Moreover, what does it mean for a representation h $$ h $$ to be “optimal” for domain adaptation. This matters because, without a formal notion of optimality, existing methods such as adversarial training can only propose candidate solutions without offering a principled benchmark. The challenge arises from the fact that the distributions of x $$ \mathbf{x} $$ may differ arbitrarily across domains, making direct transfer of models infeasible and requiring careful extraction of predictive features that remain invariant. Helland's symmetry-based framework provides precisely the right tools for this task: group actions, Hilbert spaces, and invariance measures that together define optimality in rigorous geometric terms. In practice, adversarial training (Ganin et al., 2016) enforces h # f S ≈ h # f T $$ {h}_{\#}{f}_S\approx {h}_{\#}{f}_T $$ by jointly training a domain discriminator to distinguish between h # f S $$ {h}_{\#}{f}_S $$ and h # f T $$ {h}_{\#}{f}_T $$ and a representation map h $$ h $$ to both confuse the discriminator and minimize prediction loss on the source domain; however, optimality was not addressed. Our formulation (1), grounded in Helland's geometry, provides the missing rigor by defining “optimal” invariant representations as solutions to constrained optimization problems, in the same way that PLS characterizes optimal subspaces. In sum, the challenge in domain adaptation is that source and target domains differ in their raw covariates x $$ \mathbf{x} $$ , making direct transfer infeasible and necessitating extraction of invariant predictive features z $$ \mathbf{z} $$ . By embedding this problem into Helland's symmetry-based framework, we obtain a principled notion of optimality, that is, minimizing prediction risk subject to distributional invariance. This new formulation replaces heuristic criteria with a precise benchmark for optimality, clarifying why some representations are preferable and enabling fair comparison across methods. This also demonstrates the novel applicability of Helland's framework to modern machine learning. The article was supported by National Institutes of Health under grants R01CA249096 and R01CA269398. The authors declare no conflicts of interest. Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.