Generative image steganography aims to conceal secret information in generated images without arousing suspicion. However, in practical scenarios involving high-capacity embedding or lossy transmission, existing methods still suffer from limited extraction accuracy. The main challenge lies in accurately recovering the secret-embedded latent vectors from stego images. To address this issue, we propose CIF, a constrained inversion framework designed to achieve accurate message extraction. Specifically, CIF mitigates dynamic structural errors by enforcing path consistency in the latent space and reduces numerical integration errors by adaptively selecting the integration order according to local trajectory stability. Experimental results show that our method reduces latent reconstruction error by more than 35% and achieves higher message extraction accuracy than existing approaches.

In response to the challenges of small object detection in aerial images, such as the small size of targets, low image resolution, and high similarity between the background and targets, an improved aerial small object detection model based on YOLOv10, named SBF-YOLO, is proposed. First, a new pyramid structure, MBSFPN, is applied to the neck layer to optimize its ability to fuse multi-scale features. Next, large-scale feature extraction modules are added at the base of the feature pyramid, combining SPDConv and CSP-OKM, to improve the edge details of small objects. Additionally, dynamic adaptive modules improve upsampling and downsampling via AKConv and DySample, enabling the model to better capture fine-grained features and spatial relationships, further enhancing detection performance. Experiments on the VisDrone2019 dataset show that SBF-YOLO achieves 14.6% and 16.0% improvements in [email protected] and [email protected]: 0.95, respectively, over the baseline algorithm, demonstrating the effectiveness and superiority of this algorithm in aerial small object detection tasks. Our code is available on GitHub11https://github.com/081degrees/MBT-YOLO.git.

Timely and accurate prediction of urban traffic flow is crucial in the development of smart cities. In spite of the significant stride in spatial-temporal modeling, there are the following challenges. Firstly, traffic flow is influenced by the interplay between long-term trends and sudden events. Secondly, traffic flow exhibits distribution heterogeneity across different regions and temporal dynamics over various time periods. To address these challenges, we introduce a novel method, called Dynamic Spatio-Temporal Graph Neural Network with Decoupled Contrasts (D2STCNet). Specifically, our model based on an efficient spatio-temporal encoder with a dynamic graph structure. In the data upstream, we leverage a decoupled input head to process data independently, disentangling complex traffic flow. Following this, we incorporate self-supervised learning to model Heterogeneity within the data. Through comprehensive experiments conducted on four real-world datasets, we demonstrate that our model boasts superior performance while taking efficiency into account. Our code is available on GitHub11https://github.com/21ess/D2STCNet.