TL;DR
A test-time adaptation framework that requires no target-domain supervision that introduces two key innovations: a learnable dynamic augmentation module for better capturing target-domain characteristics, and a self-supervised adaptation strategy for leveraging unlabeled target-domain images by enforcing structural priors and cross-augmentation consistency, thereby improving prediction consistency under domain shift.
Source-free domain adaptation (SFDA) aims to adapt pre-trained source models to new target domains without requiring access to any source domain data, thereby addressing privacy and efficiency concerns. Existing SFDA methods for object detection primarily follow a teacher–student self-training paradigm; however, their performance is often limited by noisy pseudo-labels. To address this issue, this paper proposes an SFDA object detection framework for the YOLO family of single-stage detectors. First, a weak–strong pseudo-label consistency filtering strategy is designed to remove unreliable pseudo-labels by exploiting the prediction consistency across different augmented views. Second, a multiscale object-level contrastive learning mechanism is introduced to extract object-level features at multiple feature scales, thereby enhancing the consistency and discriminability of object representations across different views and scales through supervised contrastive constraints. Experimental results show that the proposed method consistently outperforms the baseline on multiple cross-domain detection tasks, demonstrating its effectiveness and good generalization ability under the source-free setting.
Domain shifts caused by varying remote sensing sensors severely degrade the model performance of object detection when deployed in target domains. While unsupervised domain adaptation provides a promising solution, existing methods tend to suffer from pseudolabel noise accumulation, leading to unstable and unreliable detection in complex remote sensing scenarios. To address both domain discrepancy and pseudolabel degradation, we propose ICPNet, an Iterative Consistency Pseudo-Labeling Network that integrates self-supervised learning with pseudolabel refinement to enhance domain robustness and label reliability. Specifically, we design an adversarial teacher-student framework enhanced with an Auxiliary Masked Autoencoder to learn domain-invariant representations by randomly masking feature regions and enforcing cross-domain reconstruction. In addition, we introduce an Iterative Consistency Pseudolabel Filter (ICPF) that leverages a memory of historical predictions. By filtering labels based on their stability across multiple training iterations, the ICPF generates a high-quality and stable supervisory information. A confidence decay mechanism further enhances stability by reducing the influence of outdated predictions. Extensive experiments on diverse remote sensing datasets demonstrate that our ICPNet outperforms state-of-the-art unsupervised domain adaptation (UDA) approaches in cross-domain object detection.
Data augmentation, which simulates diverse visual variations to expand the source distribution and induce synthetic domain shifts, is a simple yet effective strategy for mitigating severe domain shifts and limited labeled target data in cross-domain few-shot object detection (CD-FSOD). Existing approaches rely on conventional data augmentation, such as Color-Jitter, Mosaic, and background-centric adaptation (e.g., Domain-RAG), which are limited in modeling complex domain shifts and often lead to suboptimal performance. In this paper, we propose PSP-FSOD, a principled framework that integrates prompt-driven domain simulation with feature perturbation regularization to improve generalization in CD-FSOD. To enable controllable domain synthesis, we design a prompt-driven strategy that leverages the visual grounding capability of large VLMs to jointly model foreground and background variations, generating semantically consistent yet domain-diverse training samples. Moreover, we adopt a grounding-aware generation scheme that guides object placement and alleviates semantic-spatial misalignment, thereby improving foreground adaptation. To ensure training stability and robustness, we further introduce a noise-induced feature perturbation mechanism that injects Gaussian noise into multi-scale intermediate features with distribution correction, encouraging consistent predictions under perturbations and reducing reliance on domain-specific cues. Extensive experiments demonstrate that PSP-FSOD produces high-quality domain-diverse supervision and learns domain-invariant representations, consistently improving performance across CD-FSOD benchmarks.
Deep learning models have achieved state-of-the-art performance in several computer vision tasks. However, they experience severe performance degradation when applied to real-world scenarios due to unanticipated distribution shifts. Test-Time Adaptation (TTA) attempts to solve this problem by using unlabeled data from the target domain to dynamically adapt to the test distribution at inference time, without access to the source data. However, TTA remains a challenging problem when adapting to continuous, temporally correlated data, such as videos, and in scenarios where the target domain contains severe domain shifts. For this reason, few works in the literature explore TTA for videos under such extreme conditions. To overcome these limitations, we propose Test-time Adaptation via Dual Distillation (TADD), an online adaptation framework that relies on a lightweight projection adapter to bridge the domain gap. The adapter module is pre-trained on the source domain and then adapted to the target using our proposed complementary losses: (i) zero-shot distillation, which encourages alignment with the domain-agnostic features from a pre-trained vision-language model (VLM); and (ii) target distillation, which retains the source domain discriminative knowledge encoded in the pre-trained adapter. Built upon a frozen CLIP backbone, our method introduces this lightweight projection adapter as the sole updatable component during inference. We conducted extensive evaluations on three well-known video action recognition benchmarks: UCF-HMDB, Daily-DA, and Sports-DA. Our experiments in the closed-set scenario demonstrate that our method consistently outperforms state-of-the-art TTA baselines. Notably, our TTA approach improves upon previous methods by up to +3.81% on UCF-HMDB, +2.63% on Daily-DA, and +3.03% on Sports-DA.
Efficient Unsupervised Domain Adaptation (EUDA) is proposed, a parameter-efficient framework that leverages a frozen DINOv2 backbone as a feature extractor and updates only a lightweight bottleneck and classification head to promote both discriminative learning and cross-domain alignment.
This article addresses the problem of cross-view heterogeneous change detection using satellite images and autonomous aerial vehicle imagery, a setting characterized by severe viewpoint differences, scale variations, and sensing modality discrepancies, as well as the absence of reliable labels. To overcome these challenges, we propose a self-supervised contrastive and predictive learning framework that learns discriminative representations directly from unlabeled cross-view image pairs. The framework jointly exploits patch-level and pixel-level objectives to capture both fine-grained local changes and higher-level semantic consistency across views. In addition, a CutMix-based data augmentation strategy is introduced to improve representation diversity. We further enhance Swin Transformer with cross-attention and channel attention mechanisms to facilitate effective multiscale and cross-modal feature interaction. Experimental results on cross-view and bitemporal change detection datasets demonstrate that the proposed approach achieves robust and competitive performance without relying on precise geometric alignment or manual annotations, highlighting its practical relevance for real-world Earth observation applications involving heterogeneous and rapidly acquired data.