The growing adoption of Internet of Things (IoT) devices has introduced significant challenges to network security due to their heterogeneous nature and temporal pattern vulnerabilities. Among the emerging threats, adversarial attacks targeting IoT network traffic have gained attention for their ability to evade traditional and machine learning-based Network Intrusion Detection Systems (NIDS). While prior work has focused on static adversarial perturbations, these approaches fail to account for the temporal dynamics inherent in IoT traffic. IoT networks exhibit time-dependent patterns driven by device behavior, environmental factors and user interactions, creating an opportunity for more sophisticated adversarial strategies. This paper introduces a co-evolutionary adversarial framework termed dynamic Adversarial Temporal-Camouflaged Perturbation (ATCP) for IoT network attack traffic. ATCP dynamically segments network traffic into temporal intervals and applies targeted adversarial perturbations to each segment. By leveraging the temporal characteristics of IoT traffic, the proposed method generates subtle yet effective adversarial modifications that confuse NIDS by disrupting their ability to model time-dependent traffic patterns. Unlike static perturbation methods, ATCP provides valuable insights into adversarial attack methodologies, lays the foundation for developing more robust IoT security frameworks, and adapts to evolving traffic dynamics, making it a more effective and robust NIDS in real-world scenarios. Extensive experiments conducted on real-world IoT network datasets demonstrate that the proposed method achieves high evasion rates against Machine Learning (ML) based NIDS while preserving the functional integrity of IoT communications. Notably, among the four NIDS evaluated, KitNET experiences the most significant degradation, with its detection rate dropping from 93.07% to 18.55% after applying ATCP. Furthermore, ATCP exhibits strong adaptability across diverse IoT device types and network configurations, highlighting its generalizability.

Given a composite image, image harmonization aims to adjust the foreground illumination to be consistent with background. Previous methods have explored transforming foreground features to achieve competitive performance. In this work, we show that using global information to guide foreground feature transformation could achieve significant improvement. Besides, we propose to transfer the foreground-background relation from real images to composite images, which can provide intermediate supervision for the transformed encoder features. Additionally, considering the drawbacks of existing harmonization datasets, we also contribute a ccHarmony dataset which simulates the natural illumination variation. Extensive experiments on iHarmony4 and our contributed dataset demonstrate the superiority of our method. Our ccHarmony dataset is released at https://github.com/bcmi/Image-HarmonizationDataset-ccHarmony.

Image harmonization is an essential step in image composition that adjusts the appearance of composite foreground to address the inconsistency between foreground and background. Existing methods primarily operate in correlated RGB color space, leading to entangled features and limited representation ability. In contrast, decorrelated color space (e.g., Lab) has decorrelated channels that provide disentangled color and illumination statistics. In this paper, we explore image harmonization in dual color spaces, which supplements entangled RGB features with disentangled L, a, b features to alleviate the workload in harmonization process. The network comprises a RGB harmonization backbone, an Lab encoding module, and an Lab control module. The backbone is a U-Net network translating composite image to harmonized image. Three encoders in Lab encoding module extract three control codes independently from L, a, b channels, which are used to manipulate the decoder features in harmonization backbone via Lab control module. Our code and model are available at https://github.com/bcmi/DucoNet-Image-Harmonization.