Unlabeled Target Domain Research Articles

Agile Multi-Source-Free Domain Adaptation

Efficiently utilizing rich knowledge in pretrained models has become a critical topic in the era of large models. This work focuses on adaptively utilize knowledge from multiple source-pretrained models to an unlabeled target domain without accessing the source data. Despite being a practically useful setting, existing methods require extensive parameter tuning over each source model, which is computationally expensive when facing abundant source domains or larger source models. To address this challenge, we propose a novel approach which is free of the parameter tuning over source backbones. Our technical contribution lies in the Bi-level ATtention ENsemble (Bi-ATEN) module, which learns both intra-domain weights and inter-domain ensemble weights to achieve a fine balance between instance specificity and domain consistency. By slightly tuning source bottlenecks, we achieve comparable or even superior performance on a challenging benchmark DomainNet with less than 3% trained parameters and 8 times of throughput compared with SOTA method. Furthermore, with minor modifications, the proposed module can be easily equipped to existing methods and gain more than 4% performance boost. Code is available at https://github.com/TL-UESTC/Bi-ATEN.

Reviewing the Forgotten Classes for Domain Adaptation of Black-Box Predictors

For addressing the data privacy and portability issues of domain adaptation, Domain Adaptation of Black-box Predictors (DABP) aims to adapt a black-box source model to an unlabeled target domain without accessing both the source-domain data and details of the source model. Although existing DABP approaches based on knowledge distillation (KD) have achieved promising results, we experimentally find that these methods all have the minority class forgetting issue, which refers that the trained model completely forgets some minority classes. To address this issue, we propose a method called Reviewing the Forgotten Classes (RFC), which including two main modules. Firstly, we propose a simple but effective component called selection training (ST). ST selects classes that the model tends to forget according to the learning status of the model and obtains clean samples of the selected classes with the small-loss criterion for enhanced training. ST is orthogonal to previous methods and can effectively alleviate their minority class forgetting issue. Secondly, we find that neighborhood clustering (NC) can help the model learn more balanced than KD so that further alleviate the minority class forgetting issue. However, NC is based on the fact that target features from the source model already form some semantic structure, while DABP is unable to obtain the source model. Thus, we use KD and ST to warm up the target model to form a certain semantic structure. Overall, our method inherits the merits of both ST and NC, and achieves state of the art on three DABP benchmarks.

Cycle Self-Refinement for Multi-Source Domain Adaptation

Multi-source domain adaptation (MSDA) aims to transfer knowledge from multiple source domains to the unlabeled target domain. In this paper, we propose a cycle self-refinement domain adaptation method, which progressively attempts to learn the dominant transferable knowledge in each source domain in a cycle manner. Specifically, several source-specific networks and a domain-ensemble network are adopted in the proposed method. The source-specific networks are adopted to provide the dominant transferable knowledge in each source domain for instance-level ensemble on predictions of the samples in target domain. Then these samples with high-confidence ensemble predictions are adopted to refine the domain-ensemble network. Meanwhile, to guide each source-specific network to learn more dominant transferable knowledge, we force the features of the target domain from the domain-ensemble network and the features of each source domain from the corresponding source-specific network to be aligned with their predictions from the corresponding networks. Thus the adaptation ability of source-specific networks and the domain-ensemble network can be improved progressively. Extensive experiments on Office-31, Office-Home and DomainNet show that the proposed method outperforms the state-of-the-art methods for most tasks.

Enhancing open-set domain adaptation through unknown-filtering multi-classifier adversarial network

Domain adaptation is a fundamental research problem that aims to address the domain shift issue during the transfer of a model from a labeled source domain to an unlabeled target domain. In traditional domain adaptation scenarios, it is assumed that the class spaces of both the source and target domains are identical. However, real-world applications often entail situations where the target domain contains private classes that are absent in the source domain. Forcing the alignment of these two domains may result in negative transfer. This specific concern is addressed by the emerging field of Open-set Domain Adaptation (OSDA). Previous OSDA methods attempted to align the known classes between the source and target domains while separating the unknown samples. However, these methods are inadequate in effectively discerning instances from the unknown classes. Therefore, we propose Enhancing Open-Set Domain Adaptation through Unknown-Filtering Multi-Classifier Adversarial Network (UFMCAN), which leverages multiple classifiers including a weighted auxiliary classifier, an open-set recognizer, a primary classifier and a three-way domain discriminator through adversarial learning to effectively filter instances from the unknown classes present in the target domain while concurrently aligning the source and target-known distribution. Experiments on extensive benchmarks (Office-31, Office-Home, VisDA-2017 and Office-Mix) demonstrate the superior performance of UFMCAN compared to existing state-of-the-art methods.

Domain adaptive remote sensing image semantic segmentation with prototype guidance

Current unsupervised domain adaptation (UDA) techniques in semantic segmentation effectively decrease the domain discrepancy between the labeled source domain and unlabeled target domain, thereby enhancing the model’s pixel-wise discriminative capability for target domain data. However, in remote sensing images (RSIs), our study uncovers that these approaches perform poorly in the presence of class distribution inconsistencies between the source and target domains. In this work, we propose a one-stage mean teacher framework with a novel auxiliary prototype classifier, named MTA, to address this issue. Specifically, the teacher model assigns pseudo labels at pixel level for target samples and captures knowledge from the student model via exponential moving average (EMA). With labeled source samples and target samples that have pseudo labels, the student model can alleviate the divergence between the source and target domains. In addition, the auxiliary prototype classifier (APC) reduces the performance degradation in the parametric softmax classifier of the student model caused by class distribution divergence. We also propose a prototype computation scheme to obtain each class prototype in the APC. Specifically, we build a memory bank for each class of the two domains to store feature embeddings dynamically. Then, we compute the class prototype by applying the clustering algorithm on memory banks corresponding to the class. Meanwhile, the APC reduces the intra-class domain discrepancy by optimizing the cross-entropy loss, which brings each class feature distribution of the two domains closer to the class prototype. The experimental results on RSIs UDA semantic segmentation tasks show the superiority of our approach over comparative methods.

Multi-class center dynamic contrastive learning for unsupervised domain adaptation person re-identification

Unsupervised domain adaptation person re-identification (UDA Re-ID) aims to leverage the pedestrian knowledge learned from labeled source domain to assist in learning the pedestrian knowledge in the unlabeled target domain. Most of existing investigations typically utilize single-class center clustering algorithms to group unlabeled target domain instances. Unfortunately, single-class center clustering algorithms tend to cluster pedestrian pictures from different identities into the same cluster, leading to inaccurate labels. Training with these noisy labels can undesirably deteriorate the accuracy of UDA Re-ID. Responding to the problem, we propose a multi-class center dynamic contrastive learning (MCC-DCL) for UDA Re-ID, which includes three main parts: multi-center clustering (MCC), dynamic pseudo-labeling (DPL), and dynamic contrastive learning (DCL). In order to reduce noisy labels generated during clustering, we introduce MCC method to generates reliable pseudo-labels for instances. Furthermore, to fully utilize the knowledge learned by the network during each iteration, we propose DPL method to optimizes the pseudo-labels of instances. Finally, for improving the discriminative performance of model and its tolerance to noisy labels, we propose DCL method that utilizes dynamic pseudo-labels and dynamic contrastive loss for supervised training. Comprehensive experiments and analyses demonstrate that MCC-DCL significantly outperforms existing approaches in UDA Re-ID tasks.

Deep Joint Semantic Adaptation Network for Multi-source Unsupervised Domain Adaptation

Multi-source Unsupervised Domain Adaptation (MUDA) transfers knowledge learned from multiple labeled source domains to an unlabeled target domain by minimizing the domain shift between multiple source domains and the target domain. Recent studies on MUDA have focused on aligning the distribution of each pair of source and target domains in separate feature spaces to reduce their domain shift. However, these approaches suffer from two main shortcomings. First, they usually focus on the global domain shift which lacks consideration of the joint distribution of category-corresponded subdomains. Second, out-of-distribution samples far from the sample center are hard to align by the global domain alignment. Therefore, we propose a novel Deep Joint Semantic Adaptive Network (DJSAN) for MUDA. Specifically, a new maximum mean discrepancy-based metric, Joint Semantic Maximum Mean Discrepancy (JSMMD), is proposed, which can uniformly optimize the cross-domain joint distribution of category-corresponded subdomains on multiple task-specific layers. Moreover, to deal with the out-of-distribution hard samples, we propose an across-domain data augmentation method called Source-Target Domain Mixing (STDMix) to enhance the robustness of the model, which synthesizes the source domain and target domain into a new domain at a fixed ratio and utilizes information entropy to provide reliable pseudo-labels for samples in the target domain. Experimental results on three public datasets, i.e., Office-31, Digits-five, and Office-Home, show that our proposed method achieves improvements of 0.3%, 1.8%, and 2.7% in average accuracy, respectively.

Unsupervised Multitarget Domain Adaptation With Dictionary-Bridged Knowledge Exploitation.

Unsupervised domain adaptation (UDA) is an emerging learning paradigm that models on unlabeled datasets by leveraging model knowledge built on other labeled datasets, in which the statistical distributions of these datasets are usually not identical. Formally, UDA is to leverage knowledge from a labeled source domain to promote an unlabeled target domain. Although there have been a variety of methods proposed to address the UDA problem, most of them are dedicated to single-source-to-single-target domain, while the works on single-source-to-multitarget domain are relatively rare. Compared to the single-source domain with single-target domain scenario, the UDA from single-source domain to multitarget domain is more challenging since it needs to consider not only the relationships between the source and the target domains but also those among the target domains. To this end, this article proposes a kind of dictionary learning-based unsupervised multitarget domain adaptation method (DL-UMTDA). In DL-UMTDA, a common dictionary is constructed to correlate the single-source and multitarget domains, while individual dictionaries are designed to exploit the private knowledge for the target domains. Through learning the corresponding dictionary representation coefficients in the UDA process, the correlations from the source to the target domains as well as these potential relationships between the target domains can be effectively exploited. In addition, we design an alternating algorithm to solve the DL-UMTDA model with theoretical convergence guarantee. Finally, extensive experiments on benchmark (Office + Caltech) and real datasets (AgeDB, Morph, and CACD) validate the superiority of the proposed method.

Two-stage adversarial learning based unsupervised domain adaptation for retinal OCT segmentation.

Deep learning based optical coherence tomography (OCT) segmentation methods have achieved excellent results, allowing quantitative analysis of large-scale data. However, OCT images are often acquired by different devices or under different imaging protocols, which leads to serious domain shift problem. This in turn results in performance degradation of segmentation models. Aiming at the domain shift problem, we propose a two-stage adversarial learning based network (TSANet) that accomplishes unsupervised cross-domain OCT segmentation. In the first stage, a Fourier transform based approach is adopted to reduce image style differences from the image level. Then, adversarial learning networks, including a segmenter and a discriminator, are designed to achieve inter-domain consistency in the segmentation output. In the second stage, pseudo labels of selected unlabeled target domain training data are used to fine-tune the segmenter, which further improves its generalization capability. The proposed method was tested on cross-domain datasets for choroid or retinoschisis segmentation tasks. For choroid segmentation, the model was trained on 400 images and validated on 100 images from the source domain, and then trained on 1320 unlabeled images and tested on 330 images from target domain I, and also trained on 400 unlabeled images and tested on 200 images from target domain II. For retinoschisis segmentation, the model was trained on 1284 images and validated on 312 images from the source domain, and then trained on 1024 unlabeled images and tested on 200 images from the target domain. The proposed method achieved significantly improved results over that without domain adaptation, with improvement of 8.34%, 55.82% and 3.53% in intersection over union (IoU) respectively for the three test sets. The performance is better than some state-of-the-art domain adaptation methods. The proposed TSANet, with image level adaptation, feature level adaptation and pseudo-label based fine-tuning, achieved excellent cross-domain generalization. This alleviates the burden of obtaining additional manual labels when adapting the deep learning model to new OCT data.

Multiscale Change Detection Domain Adaptation Model Based on Illumination–Reflection Decoupling

In the change detection (CD) task, the substantial variation in feature distributions across different CD datasets significantly limits the reusability of supervised CD models. To alleviate this problem, we propose an illumination–reflection decoupled change detection multi-scale unsupervised domain adaptation model, referred to as IRD-CD-UDA. IRD-CD-UDA maintains its performance on the original dataset (source domain) and improves its performance on unlabeled datasets (target domain) through a novel CD-UDA structure and methodology. IRD-CD-UDA synergizes mid-level global feature marginal distribution domain alignment, classifier layer feature conditional distribution domain alignment, and an easy-to-hard sample selection strategy to increase the generalization performance of CD models on cross-domain datasets. Extensive experiments conducted on the LEVIR, SYSU, and GZ optical remote sensing image datasets demonstrate that the IRD-CD-UDA model effectively mitigates feature distribution discrepancies between source and target CD data, thereby achieving optimal recognition performance on unlabeled target domain datasets.

Three-stage transfer learning for motor imagery EEG recognition.

Motor imagery (MI) paradigms have been widely used in neural rehabilitation and drowsiness state assessment. The progress in brain-computer interface (BCI) technology has emphasized the importance of accurately and efficiently detecting motor imagery intentions from electroencephalogram (EEG). Despite the recent breakthroughs made in developing EEG-based algorithms for decoding MI, the accuracy and efficiency of these models remain limited by technical challenges posed by cross-subject heterogeneity in EEG data processing and the scarcity of EEG data for training. Inspired by the optimal transport theory, this study aims to develop a novel three-stage transfer learning (TSTL) method, which uses the existing labeled data from a source domain to improve classification performance on an unlabeled target domain. Notably, the proposed method comprises three components, namely, the Riemannian tangent space mapping (RTSM), source domain transformer (SDT), and optimal subspace mapping (OSM). The RTSM maps a symmetric positive definite matrix from the Riemannian space to the tangent space to minimize the marginal probability distribution drift. The SDT transforms the source domain to a target domain by finding the optimal transport mapping matrix to reduce the joint probability distribution differences. The OSM finally maps the transformed source domain and original target domain to the same subspace to further mitigate the distribution discrepancy. The performance of the proposed method was validated on two public BCI datasets, and the average accuracy of the algorithm on two datasets was 72.24% and 69.29%. Our results demonstrated the improved performance of EEG-based MI detection in comparison with state-of-the-art algorithms.

AdvMix: Adversarial Mixing Strategy for Unsupervised Domain Adaptive Object Detection

Recent object detection networks suffer from performance degradation when training data and test data are distinct in image styles and content distributions. In this paper, we propose a domain adaptive method, Adversarial Mixing (AdvMix), where the label-rich source domain and unlabeled target domain are jointly trained by the adversarial feature alignment and a self-training strategy. To diminish the style gap, we design the Adversarial Gradient Reversal Layer (AdvGRL), containing a global-level domain discriminator to align the domain features by gradient reversal, and an adversarial weight mapping function to enhance the stability of domain-invariant features by hard example mining. To eliminate the content gap, we introduce a region mixing self-supervised training strategy where a region of the target image with the highest confidence is selected to merge with the source image, and the synthesis image is self-supervised by the consistency loss. To improve the reliability of self-training, we propose a strict confidence metric combining both object and bounding box uncertainty. Extensive experiments conducted on three benchmarks demonstrate that AdvMix achieves prominent performance in terms of detection accuracy, surpassing existing domain adaptive methods by nearly 5% mAP.

Cluster-based Adversarial Decision Boundary for domain-adaptive open set recognition

Domain adaptation has achieved significant progress recently by adapting models trained on a source domain to an unlabeled target domain. Open Set Domain adaptation (OSDA) has drawn much attention nowadays, where the target domain contains some exclusive categories other than the source domain’s known classes. With no label in the target data, existing OSDA methods often suffer from negative transfer. Conventional methods for unknown class rejection require an empirical setting of the confidence threshold, which lacks flexibility since the model confidence may vary during the training process, and our motivation is to omit the effort of setting the rejection threshold manually. Based on the idea that latent features of the same class should be in the same cluster to address this issue, we propose a domain adaptive open set recognition framework: Cluster-based Adversarial Decision Boundary (CADB). Specifically, we design an end-to-end unknown class rejection model consisting of three components: known class prototype estimation under the cluster assumption; known class similarity score estimation; and adaptive unknown class rejection threshold generation with adversarial feature suppression. These three components work as one entity to give a similarity score for each sample. Those samples that are less similar to the cluster prototype compared with the counterfactual features are rejected as the unknown class. Extensive evaluations are conducted to verify the effectiveness and robustness of the proposed boundary generation procedure.

Diffusion‐UDA: Diffusion‐based unsupervised domain adaptation for submersible fault diagnosis

AbstractDeep learning has demonstrated notable success in mechanical signal processing with a large amount labelled data. However, the systems of the Jiaolong deep‐sea submersible prone to malfunction are typically diverse, due to the high complexity of its structure and operational environment. Consequently, this diversity gives rise to variations in the types of sensor signals and their associated data distributions that require analysis. Unsupervised domain adaptation (UDA) uses transferable knowledge derived from the source domain and applies it to an unlabelled target domain in order to improve the reusability of pre‐existing models and data. Inspired by the diffusion models that have the robust capabilities to transform data distributions across a large gap, we propose a novel diffusion‐based unsupervised domain adaptation (diffusion‐UDA) model, which leverages contrastive learning to alleviate the challenges of cross‐domain analysis for fault diagnosis within different systems of the deep‐sea submersible. Experimental results show the proposed method achieves state‐of‐the‐art performance on various benchmarks.

Wavelet packet decomposition with motif patterns for rolling bearing fault diagnosis under variable working loads

Bearing intelligent diagnosis based on signal processing has been a hot research topic. However, due to the different data distribution caused by the variable working loads, the model learned from source domain has poor performance in target domain. To solve this problem, a feature extraction method named Wavelet Packet Decomposition with Motif Patterns (WPDMP) is proposed. Firstly, multiscale signals are obtained using wavelet packet decomposition; then, the MP features of these multiscale signals and the original signal are extracted; finally, these MPs are combined as input vector of support vector classification (SVC) for fault identification. Compared with other methods, the proposed method has extraordinary superiority for unlabeled target domain fault diagnosis. In addition, the feature visualization results show that the proposed model can extract domain invariant features, so the proposed model has considerable research prospects.

MDLR: A Multi-Task Disentangled Learning Representations for unsupervised time series domain adaptation

Unsupervised Time Series Domain Adaptation (UTSDA) is a method for transferring information from a labeled source domain to an unlabeled target domain. The majority of existing UTSDA approaches focus on learning a domain-invariant feature space by reducing the gap between domains. However, the single-task representation learning methods have limited expressive capability, while ignoring the distinctive season-related and trend-related domain-invariant mechanisms across different domains. To address this, we introduce a novel approach, distinct from existing methods, through a theoretical analysis of UTSDA from the perspective of causal inference. This analysis establishes a solid theoretical foundation for identifying and modeling such consistent domain-invariant mechanisms, which is a significant advancement in the field. As a solution, we introduce MDLR, a multi-task disentangled learning framework designed for UTSDA. MDLR utilizes a dual-tower architecture with a trend feature extractor (TFE) and a season feature extractor (SFE) to extract trend-related and season-related information. This approach ensures that domain-invariant features at different scales can be better represented. Additionally, MDLR is designed with two tasks: a label classifier and a domain classifier, enabling iterative training of the entire model. The experiments conducted on three datasets, namely UCIHAR, WISDM, and HHAR_SA, along with visualization results, have shown the effectiveness of the proposed approach. The source code for our MDLR model is available to the public at https://github.com/MoranCoder95/MDLR/.

Hierarchical alignment network for domain adaptive object detection in aerial images

Domain Adaptive Object Detection (DAOD) alleviates the reliance on labeled data by transferring knowledge learned from labeled source domain to unlabeled target domain. Recent DAOD methods is modeled mainly based on ground-level images. Compared to ground-level images, aerial images suffer from scale variation and viewpoint diversity. This means that domain adaptive object detection in aerial images is a more challenging task. In this work, we address the domain shift in aerial images from two levels: 1) image-level shifts, such as weather, lighting, viewpoint, etc., 2) instance-level shifts, such as object appearance, scale, etc. Specifically, multiple domain-confusion classifiers are designed to learn image-level common knowledge of the source and target domains. Different levels of domain classifiers are further assigned adaptive weights to coordinate the transferability and discriminability of the adaptive detectors. Meanwhile, instance-level alignment is realized by forcing the intrinsic relationship between classes in both domains to be consistent. In addition, we perform instance-level alignment in different semantic-level feature layers to improve the scale awareness of the adaptation model. Extensive experimental results on VisDrone, UAVDT, DIOR and DOTA datasets demonstrate that our method achieves optimal detection performance in four domain adaption scenarios compared to other state-of-the-art methods, e.g., in Daytime → Night (VisDrone), the mAP50 is 23.5 %; in VisDrone → UAVDT, DIOR → UAVDT and DOTA → VisDrone, the AP50 of the car is 63.1 %, 46.6 % and 44.8 % respectively. Code will be available online (https://github.com/MaYou1997/HANet).

You only label once: A self‐adaptive clustering‐based method for source‐free active domain adaptation

AbstractWith the growing significance of data privacy protection, Source‐Free Domain Adaptation (SFDA) has gained attention as a research topic that aims to transfer knowledge from a labeled source domain to an unlabeled target domain without accessing source data. However, the absence of source data often leads to model collapse or restricts the performance improvements of SFDA methods, as there is insufficient true‐labeled knowledge for each category. To tackle this, Source‐Free Active Domain Adaptation (SFADA) has emerged as a new task that aims to improve SFDA by selecting a small set of informative target samples labeled by experts. Nevertheless, existing SFADA methods impose a significant burden on human labelers, requiring them to continuously label a substantial number of samples throughout the training period. In this paper, a novel approach is proposed to alleviate the labeling burden in SFADA by only necessitating the labeling of an extremely small number of samples on a one‐time basis. Moreover, considering the inherent sparsity of these selected samples in the target domain, a Self‐adaptive Clustering‐based Active Learning (SCAL) method is proposed that propagates the labels of selected samples to other datapoints within the same cluster. To further enhance the accuracy of SCAL, a self‐adaptive scale search method is devised that automatically determines the optimal clustering scale, using the entropy of the entire target dataset as a guiding criterion. The experimental evaluation presents compelling evidence of our method's supremacy. Specifically, it outstrips previous SFDA methods, delivering state‐of‐the‐art (SOTA) results on standard benchmarks. Remarkably, it accomplishes this with less than 0.5% annotation cost, in stark contrast to the approximate 5% required by earlier techniques. The approach thus not only sets new performance benchmarks but also offers a markedly more practical and cost‐effective solution for SFADA, making it an attractive choice for real‐world applications where labeling resources are limited.

Subdomain alignment based open-set domain adaptation image classification

Domain adaptation has achieved great success in using labeled source domain samples to identify unlabeled target domain samples. Here, we aim to solve the open-set domain adaptation, which is different from the closed-set domain adaptation in that it contains categories in target domain that do not appear in source domain. To solve this problem, this paper proposes open-set domain adaptation model based on subdomain alignment, which uses variable weights for discriminative training of unknown samples in target domain. Aiming at the distribution differences between domains, the model aligns the distributions of the category subspaces of source and target domains, enhancing the distribution similarity within the subspaces of the same category. Through experiments on different domain adaptation datasets, the results show that the model proposed in this paper effectively improves the accuracy of open-set domain adaptation classification.

Learning to Discover Knowledge: A Weakly-Supervised Partial Domain Adaptation Approach.

Domain adaptation has shown appealing performance by leveraging knowledge from a source domain with rich annotations. However, for a specific target task, it is cumbersome to collect related and high-quality source domains. In real-world scenarios, large-scale datasets corrupted with noisy labels are easy to collect, stimulating a great demand for automatic recognition in a generalized setting, i.e., weakly-supervised partial domain adaptation (WS-PDA), which transfers a classifier from a large source domain with noises in labels to a small unlabeled target domain. As such, the key issues of WS-PDA are: 1) how to sufficiently discover the knowledge from the noisy labeled source domain and the unlabeled target domain, and 2) how to successfully adapt the knowledge across domains. In this paper, we propose a simple yet effective domain adaptation approach, termed as self-paced transfer classifier learning (SP-TCL), to address the above issues, which could be regarded as a well-performing baseline for several generalized domain adaptation tasks. The proposed model is established upon the self-paced learning scheme, seeking a preferable classifier for the target domain. Specifically, SP-TCL learns to discover faithful knowledge via a carefully designed prudent loss function and simultaneously adapts the learned knowledge to the target domain by iteratively excluding source examples from training under the self-paced fashion. Extensive evaluations on several benchmark datasets demonstrate that SP-TCL significantly outperforms state-of-the-art approaches on several generalized domain adaptation tasks. Code is available at https://github.com/mc-lan/SP-TCL.