Accurate segmentation of aortic vascular structures is critical for diagnosing and treating cardiovascular diseases. Traditional Transformer-based models have shown promise in this domain by capturing long-range dependencies between vascular features. However, their reliance on fixed-size rectangular patches often influences the integrity of complex vascular structures, leading to suboptimal segmentation accuracy. To address this challenge, we propose the adaptive Morph-Patch Transformer (MPT), a novel architecture specifically designed for aortic vascular segmentation. Specifically, MPT introduces an adaptive patch partitioning strategy that dynamically generates morphology-aware patches aligned with complex vascular structures. This strategy can preserve semantic integrity of complex vascular structures within individual patches. Moreover, a Semantic Clustering Attention (SCA) method is proposed to dynamically aggregate features from various patches with similar semantic characteristics. This method enhances the model's capability to segment vessels of varying sizes, preserving the integrity of vascular structures. Extensive experiments on three open-source datasets (AVT, AortaSeg24 and TBAD) demonstrate that MPT achieves state-of-the-art performance, with improvements in segmenting intricate vascular structures.

Social event detection (SED) involves identifying and analyzing significant real-world events using data generated on social media platforms. With the rapid growth of platforms like Weibo and Twitter, users are sharing not just text but also images. However, most existing SED methods remain text-focused, limiting their ability to fully capture the complexity of real-world social dynamics. Moreover, the lack of multi-modal datasets specifically designed for SED has blocked the development of models that can effectively exploit these rich content types. To address these limitations, we introduced WEIBO2022, an extensive multi-modal SED dataset that includes both text and image data. The dataset is available in two versions: WEIBO2022-Medium, containing 25,435 entries and WEIBO2022-Large, containing 79,825 entries. In addition, we presented a novel network called the Key Instance-driven, Quality-aware Graph Neural Network (KQGNN), which features a key instance-driven library, a quality-aware learning process, and a multi-modal fusion module, enhancing its ability to detect events accurately in both offline and real-time settings. Extensive experiments showcase the exceptional performance and superiority of the proposed model, showing improvements in detection accuracy and effective prevention of catastrophic forgetting during continuous training.

3D Gaussian Splatting has demonstrated notable success in large-scale scene reconstruction, but challenges persist due to high training memory consumption and storage overhead. Hybrid representations that integrate implicit and explicit features offer a way to mitigate these limitations. However, when applied in parallelized block-wise training, two critical issues arise since reconstruction accuracy deteriorates due to reduced data diversity when training each block independently, and parallel training restricts the number of divided blocks to the available number of GPUs. To address these issues, we propose Momentum-GS, a novel approach that leverages momentum-based self-distillation to promote consistency and accuracy across the blocks while decoupling the number of blocks from the physical GPU count. Our method maintains a teacher Gaussian decoder updated with momentum, ensuring a stable reference during training. This teacher provides each block with global guidance in a self-distillation manner, promoting spatial consistency in reconstruction. To further ensure consistency across the blocks, we incorporate block weighting, dynamically adjusting each block's weight according to its reconstruction accuracy. Extensive experiments on large-scale scenes show that our method consistently outperforms existing techniques, achieving a 12.8% improvement in LPIPS over CityGaussian with much fewer divided blocks and establishing a new state of the art. Project page: https://jixuan-fan.github.io/Momentum-GS_Page/

3D Gaussian Splatting (3DGS) has emerged as a high-quality and efficient alternative to Neural Radiance Fields (NeRF), offering distinct advantages in scene representation and suitability for multi-view 3D object detection (MV-3DOD) tasks. However, conventional 3DGS-based detection methods typically employ a reconstruction-then-detection pipeline, which is time-consuming and unsuitable for real-time applications. This approach arises from 3DGS’s original design for reconstruction tasks, which lacks network-based training. In this paper, we propose GS-3Det, an online Gaussian detection framework for MV-3DOD, achieving real-time performance through a single forward pass for scene reconstruction and detection. Specifically, we introduce a detection-aware Gaussian grid that enables directly prediction of explicit 3D Gaussians from multi-view images, enabling efficient and robust 3D scene understanding. Additionally, we propose a Dual-Path Consistency module that leverages 3D constraints to improve the accuracy of Gaussian grid representation and detection. Experiments on the ScanNet V2 dataset demonstrate that GS-3Det surpasses state-of-the-art methods by 3.5% in [email protected] and 2.7% in [email protected], underscoring its generalization capability and real-time performance.

Recent advancements in pre-trained vision-language models like CLIP, have enabled the task of open-vocabulary segmentation. CLIP demonstrates impressive zero-shot capabilities in various downstream tasks that require holistic image understanding. However, due to the image-level contrastive learning and fully global feature interaction, ViT-based CLIP struggles to capture local details, resulting in poor performance in segmentation tasks. Our analysis of ViT-based CLIP reveals that anomaly tokens emerge during the forward process, attracting disproportionate attention from normal patch tokens and thereby diminishing spatial awareness. To address this issue, we propose Self-Calibrated CLIP (SC-CLIP), a training-free method that calibrates CLIP to generate finer representations while preserving its original generalization ability-without introducing new parameters or relying on additional backbones. Specifically, we mitigate the negative impact of anomaly tokens from two complementary perspectives. First, we explicitly identify the anomaly tokens and replace them based on local context. Second, we reduce their influence on normal tokens by enhancing feature discriminability and attention correlation, leveraging the inherent semantic consistency within CLIP's mid-level features. In addition, we introduce a two-pass strategy that effectively integrates multi-level features to enrich local details under the training-free setting. Together, these strategies enhance CLIP's feature representations with improved granularity and semantic coherence. Experimental results demonstrate the effectiveness of SC-CLIP, achieving state-of-the-art results across all datasets and surpassing previous methods by 9.5%. Notably, SC-CLIP boosts the performance of vanilla CLIP ViT-L/14 by 6.8 times. Furthermore, we discuss our method's applicability to other vision-language models and tasks for a comprehensive evaluation. Our source code is available at https://github.com/SuleBai/SC-CLIP.

The reconstruction of the Normalized Difference Vegetation Index (NDVI) is a crucial prerequisite for numerous spatiotemporal continuous studies. To address the limitations posed by satellite temporal resolution and challenging atmospheric conditions, the combination of Synthetic Aperture Radar (SAR) and optical images from diverse sources has proven to be effective and widely employed. In this study, we employ the Spatial-Temporal Savitzky-Golay algorithm to rectify MODIS NDVI maps and eliminate interruptions caused by noise. Random Forest and Gradient Boosting Decision Trees serve as a dual filter to select SAR indices with the highest impact on NDVI reconstruction, ensuring that the chosen indices encapsulate the most valuable information. Subsequently, we conduct a series of ablation experiments and develop a deep learning network named Residual-in-Residual Dense Block NDVI Reconstruction net (RDNRnet). This network effectively mitigates the impacts of MODIS coarse resolution and speckle noises in SAR data. We also evaluate the network performance in reconstructing NDVI across all seasons and land cover types. Our findings highlight that the modified dual-polarimetric SAR vegetation index and the standard deviation of the VV band are the most crucial SAR indices. The predictions for summer exhibit the highest performance, with a coefficient of determination (R2) reaching 0.9757. Optimal performances by land cover type are observed in forests, paddy fields, and dry farming fields, all with R2values exceeding 0.9580. Our adaptive NDVI reconstruction solution demonstrates robust performance across different data availability scenarios, effectively catering to all seasons and land cover types.

Wearable devices have garnered widespread attention as a mobile solution, and various intelligent modules based on wearable devices are increasingly being integrated. Additionally, image captioning is an important task in computer vision that maps images to text. Existing image captioning achievements are based on high-quality visible images. However, higher target complexity and insufficient light can lead to reduced captioning performance and mistakes. In this paper, we present an infrared image captioning framework designed to solve the problem of invalid visible image captioning in special conditions. Remarkably, we integrate the infrared image captioning model into the wearable device. Volunteers perform offline and real-time environmental analysis tasks in the real world to evaluate the framework's effectiveness in multiple scenarios. The results indicate that both the accuracy of infrared image captioning and the feedback from wearable device users are promising.