Recurrent pregnancy loss (RPL) is three or more times of consecutive spontaneous loss of pregnancy. The underlying cause is complicated and the etiology of over 50% of RPL patients is unclear. In the present study, bone marrow mesenchymal stem cells were isolated from CBA/J female mice and exosomes were isolated from cell culture medium by ultracentrifugation. CBA/J female mice were paired with male DBA/2 to generate abortion prone mouse model, and CBA/J females paired with male BALB/c mice were used as control. Exosomes were injected through uterine horns into pregnant CBA/J mice on day 4.5 of gestation in abortion‐prone matting. On day 13.5 of pregnancy, abortion rates were calculated and the level of transforming growth factor‐β (TGF‐β), interleukin 10 (IL‐10), interferon g (IFN‐γ), and tumor necrosis factor a (TNF‐α) in CD4+ T cells and macrophages in deciduas were evaluated by flow cytometry. Exosomes injection improved the pregnancy outcomes in abortion prone mice. The IL‐4 and IL‐10 levels on CD4+ T cells were upregulated in the maternal–fetal interface; meanwhile, the TNF‐α and IFN‐γ levels on CD4+ T cells were reduced. The IL‐10 level was increased and IL‐12 was reduced on the monocytes that separated from deciduas. miR‐101 level was increased in the CD4+ T cells in the deciduas. In conclusion, the treatment of ESCs‐derived exosomes modulates T cells' function and macrophages activities in the maternal–fetal interface that resulted in a decreased embryo resorption rate, and provides a therapeutic potential to treat RPL.

The rapid evolution of deep learning technologies has substantially influenced the field of urban-scene point cloud semantic segmentation (USPCSS), a cornerstone for various applications ranging from road-level applications such as autonomous driving, traffic management, and railway track inspection to urban-level applications such as disaster management, urban planning, and rail infrastructure monitoring. Despite remarkable progress, the complexity of urban environments presents unique challenges. This study conducts an in-depth review of both state-of-the-art and pioneering deep learning models and datasets specifically tailored for USPCSS. First, we categorize the existing mainstream USPCSS datasets into road-level and urban-level, summarize their metadata and unique characteristics, and discuss their associated challenges in semantic segmentation. Second, we classify the USPCSS deep learning models into four categories—image-based, voxel-based, point-based, and fusion-based—highlighting unique strengths and weaknesses, architectures, and innovations. Third, we introduce the mainstream USPCSS evaluation metrics, and conduct comparative analyses of these models on selected datasets. Findings reveal that the recent point-based methods such as PTv3 significantly dominate the accuracy metrics across road-level benchmarks, indicating the robustness of using transformer-based architectures for the direct processing of complex urban-scene point clouds. In the end, we address the pressing challenges and gaps in current research, offering insights into future directions for advancing the USPCSS field. Through this review study, we aim to provide thorough guidance to researchers, engineers, and decision-makers who are seeking comprehensive understanding and cutting-edge solutions in the current field of USPCSS.

Monocular camera localization in known environments is a critical task for applications like autonomous navigation, augmented reality, and robotic positioning, requiring precise spatial awareness. Unlike localization in unknown environments, which builds maps in real time, this leverages pre-existing data for higher accuracy. This review comprehensively analyzes monocular camera localization methods in known environments, categorizing them into 2D-2D feature matching, 2D-3D feature matching, and regression-based approaches. It consolidates foundational techniques and recent advancements, providing inter-class and intra-class performance comparisons on mainstream datasets. Key findings show that 2D-3D methods generally offer the highest accuracy, especially in structured outdoor environments, due to robust use of 3D spatial information. However, recent scene coordinate regression methods, such as ACE and ACE++, achieve comparable or superior performance in indoor scenes with more efficient pipelines. This review highlights challenges and proposes future directions: (1) synthetic data generation to meet deep learning demands, while addressing domain gaps; (2) improving generalization to unseen scenes and reducing retraining; (3) multi-sensor fusion for enhanced robustness; (4) exploring transformer-based and graph neural network architectures; (5) developing lightweight models for real-time performance on resource-constrained devices. This review aims to guide researchers and practitioners in method selection and identify key research directions.

This paper presents a schematic plan outlining a methodology for determining train positions using digital twin technology and commercially available positioning sensors. The plan involves an offline step and an online step. The task in the offline step includes: building a library of 3D models for objects on railway tracks, and constructing a real digital twin environment. Then in the online step, train positions can be accurately determined by automatically generating real-time semantic point clouds from images and matching them with the 3D objects in the digital twin. Along with these tasks, a City Geography Markup Language (CityGML) Application Domain Extension (ADE) for modeling railway tracks in 3D will be developed and suggested as Open Geospatial Consortium (OGC) standard for similar applications. All these procedures will be conducted in a controlled environment using remote-controlled cars before being applied to an actual railway track. It is hoped that this localization methodology can enhance traditional positioning methods, ultimately leading to improved network operation and maintenance.

Building footprint datasets are valuable for various applications, but they are challenging to prepare due to the labor-intensive and time-consuming manual delineations of building footprint labels by human image interpreters. Although deep learning-based building extraction methods have been introduced in recent years, most of them rely on a two-stage approach involving a polygon initialization network and a polygon regularization network. These methods often suffer from high computational costs and slow inference speeds, limiting their suitability for large-scale building extraction tasks. Furthermore, achieving consistent regularization across different building types remains a challenge. This paper introduces Poly R-CNN, a region-based convolutional neural network designed for efficient, large-scale, boundary-regularized building footprint extraction. Built on the Mask R-CNN model, Poly R-CNN incorporates a lightweight polygon prediction head that works alongside the existing bounding box and mask prediction heads. This new head comprises a Vertex Proposal Network (VPN) and a fast edge-tracing algorithm. The VPN predicts instance-aware building vertex proposals, while the edge-tracing algorithm selects, traces, and refines these proposals into enclosed building polygons. Poly R-CNN outperforms two state-of-the-art deep-learning-based methods, PolygonCNN and PolyMapper, with a higher F1-score (0.722 vs. 0.663 and 0.591, respectively) and faster inference speed (248.87 ms per image vs. 457.11 ms and 604.29 ms). Additionally, unlike previous methods limited to single-image processing, Poly R-CNN supports batched training and inference which further enhances its training and inference speed. Experimental results demonstrate that Poly R-CNN can effectively extract regularized and visually appealing building polygons from large-scale urban scenes.