Shuxue Ding

Segmenting unknown domains using a model trained in the source domain still faces challenges. Although some approaches tried to resolve the problem through various data generation and network architecture designs, they cannot achieve satisfactory segmentation results compared with single domain segmentation of consistent data distribution. Therefore, we propose a data augmentation method based on amplitude perturbation to expand the distribution of data types, thereby covering target data. A feature suppression strategy is proposed to reduce the network’s over-reliance on important features of the source domain data to improve generalization performance. In addition, we design a luminance contrast consistency (LCC) learning module to harmonize the data styles between different domains and a multiscale convolutional attention (MSCA) module to enhance the network’s perception of small target objects and improve the segmentation performance of the model, which further improves segmentation performance. Our method achieves the state-of-the-art (SOTA) results on two public datasets of ATLAS2.0 and Prostate. The code is available at https://github.com/butterflyGN/DGSFTAFS

Received signal strength (RSS) based device-free localization (DFL) is commonly used in the Internet of Things (IoT) field. However, the current DFL algorithms have limitations in terms of stability and accuracy, which hinders the widespread application of DFL. Current research on DFL predominantly revolves around sparse representation and deep learning. The sparse representation method requires building a suitable dictionary to achieve higher accuracy, while the deep learning method is affected by data volume and computational complexity. In contrast to traditional localization methods that rely on raw data features, this paper suggests using the deep dictionary learning (DDL) framework to extract depth features. Then, the extracted low-level and high-level features are not only used to construct a dictionary but also to reconstruct the testing data for DFL using sparse representation classification. This approach leverages the advantages of sparse representation and deep learning to achieve highly accurate localization. The proposed deep dictionary learning model involves learning multiple dictionaries with varying descriptive capabilities to extract deep features from the observed signal through a layer-by-layer deep dictionary learning process. For better dictionary learning, we introduce the minimax-concave penalty (MCP) for each layer of dictionary learning. Utilizing the difference of convex (DC) programming, the formulated nonconvex problems are efficiently optimized. Furthermore, to enhance localization accuracy, the data is expanded to reinforce the essential features of deep dictionary learning. The performance of the DC deep dictionary learning (DCDDL) algorithm was assessed using collected laboratory datasets and public datasets, demonstrating its superiority over existing localization algorithms.

Metasurfaces, due to their electromagnetic wave manipulation capabilities, provide a crucial technological approach for designing multi-band absorbers. Given the broad application potential of absorbers in various scenarios, their inverse design has increasingly become a research focus. Current deep learning–based inverse design methods for absorbers face two major challenges: the original absorption spectrum data sequences are excessively long, making effective feature extraction difficult and prone to dimensionality explosion. The Transformer for Inversed Metasurface Design (TIMD) method overcomes the sequence length limitation by introducing Multi-Head Self-Attention (MHSA) to effectively capture contextual dependencies in long sequences. However, the absorption peaks in the spectrum are jointly determined by the unit shape and material properties of the absorber, containing rich structural information. TIMD’s global attention mechanism, while computing global correlations, tends to weaken the focused emphasis on these critical peak features, which limits model performance improvement. To further enhance model performance, this paper proposes TIMD with Pyramid Attention and Learnable Feature Fusion (TIMD-PALFF). Based on TIMD, a pyramid attention mechanism is employed to progressively reduce the feature dimensionality in the attention process, guiding the model from global perception to local focus and extracting multi-level features. Meanwhile, a Learnable Feature Fusion (LFF) module is introduced to adaptively adjust the contribution of each level’s features through a set of learnable weights, thereby enhancing the weight of peak region features and improving overall model performance. The final model achieves a loss value converging to 6.14 × 10⁻⁵ on the validation set, with the predicted absorber structures producing absorption spectra highly consistent with the target spectra.

This paper introduces ProFiC-VLP, a fine-grained vision-language pre-training (VLP) model featuring progressive freezing and feedback-controlled cropping. Our ProFiC-VLP model addresses the challenge of fine-grained image-text semantic relationship alignment in VLP, specifically in aligning visual regions with their corresponding textual phrases. ProFiC-VLP integrates three levels of image-text semantic alignment: global alignment, object alignment, and relationship alignment. It employs vision transformer for image encoding and a semantic parser to decompose text into multi-level semantic structures, which are then encoded into text semantics using lightweight querying transformers. To prevent overfitting during pre-training, ProFiC-VLP introduces a progressive freezing strategy that aligns multi-level textual representations with the corresponding image concepts. Furthermore, a novel language generation approach with a feedback-controlled cropping loss function is proposed for relationship alignment pre-training, which effectively localizes image regions corresponding to specific textual phrases. Experimental results show that ProFiC-VLP outperforms existing models across various vision-language tasks, especially in cross-modal reasoning scenarios such as visual grounding and image caption generation. Visualization experiments further underscore ProFiC-VLP’s ability to achieve fine-grained alignment of semantic relationships, significantly enhancing its capacity to address complex vision-language challenges.

Building footprint extraction holds immense significance in remote sensing image analysis and has great value in urban planning, land use, environmental protection and disaster assessment. Despite the progress made by conventional and deep learning approaches in this field, they continue to encounter significant challenges. This paper introduces a novel plug-and-play attention module, Split Coordinate Attention (SCA), which ingeniously captures spatially remote interactions by employing two spatial range of pooling kernels, strategically encoding each channel along x and y planes, and separately performs a series of split operations for each feature group, thus enabling more efficient semantic feature extraction. By inserting into a 2D CNN to form an effective SCANet, our SCANet outperforms recent SOTA methods on the public Wuhan University (WHU) Building Dataset and Massachusetts Building Dataset in terms of various metrics. Particularly SCANet achieves the best IoU, 91.61% and 75.49% for the two datasets. Our code is available at https://github.com/AiEson/SCANet

Gaze estimation is an important indicator of human behavior that can be used for human assistance. Recent gaze estimation methods are primarily based on convolutional neural networks (CNNs) or attention Transformers. However, CNNs extract a limited local context while losing important global information, whereas attention mechanisms exhibit low utilization of multiscale hybrid features. To address these issues, we propose a novel nonlinear multi-head cross-attention network with programmable gradient information (MCA-PGI), which synthesizes the advantages of CNNs and the Transformer. The programmable gradient information is used to achieve reliable gradient propagation. An auxiliary branch is incorporated to integrate the gradient information, thereby retaining more original information than CNNs. In addition, nonlinear multi-head cross-attention is employed to fuse the global visual and multiscale hybrid features for more accurate gaze estimation. Experimental results on three publicly available datasets demonstrate that the proposed MCA-PGI exhibits strong competitiveness and outperforms most state-of-the-art methods, achieving 2.5% and 10.2% performance improvements on the MPIIFaceGaze and Eyediap datasets, respectively. The implementation code can be found at https://github.com/Yuhang-Hong/MCA-PGI.