The estimation of brain effective connectivity is crucial in the identification of biomarkers associated with neurological disorders. Nevertheless, modeling nonlinear interactions between brain regions remains a considerable challenge. To overcome this issue, we introduce an innovative multi-model framework termed the Multimodel Segmented Causal Network (MSCN), which aims to estimate effective connectivity networks distinguished by strong generalization capacities and physiological interpretability. The MSCN incorporates two causality-driven modules, leveraging VAE_attention and ResNet-low-rank architectures. Collectively, these modules collaboratively model fMRI time series data to extract average effective connectivity patterns. Additionally, a functional connectivity-based assisted learning strategy is employed to enhance the estimation of effective connectivity. The constructed networks are assessed for their efficacy in classifying ASD using SVM and ridge classifiers. Experimental results from 13 ABIDE sites demonstrate that our method not only effectively identifies significant physiological mechanisms but also exhibits superior classification performance compared to existing state-of-the-art methods. Beyond its excellent performance, utilizing the proposed framework for modeling interpretable brain networks offers a promising approach for elucidating the ASD pathogenesis.

The path to fully autonomous web agents is currently hindered by a critical bottleneck: their limited ability to handle CAPTCHA. Existing agent benchmarks largely ignore this practical challenge, failing to evaluate an agent’s real-world capacity to solve CAPTCHA. To bridge this gap, we conduct a comprehensive analysis of real-world CAPTCHA distributions and introduce MirrorCAPTCHA, a benchmark annotated with Weighted Pass Rate and a newly proposed metric Completion Degree. MirrorCAPTCHA is designed to serve as a “mirror” that faithfully reflects the automation capabilities of agents in real scenarios. We filter 2095 websites from Common Crawl, identify the CAPTCHA deployed on these sites, and cluster them into 18 distinct categories using K-means algorithm. To ensure practicality, we extract a web subgraph from Common Crawl covering these websites and use random walks to simulate real-world CAPTCHA encounter frequencies, yielding a realistic measure of agents’ ability. Additionally, we develop a lightweight synthetic data pipeline to train Ovis2-Agent-CAPTCHA-8B, which significantly outperforms current state-of-the-art closed-source models on MirrorCAPTCHA, achieving a 9.4% higher average Weighted Pass Rate and a 2.13% higher average Completion Degree than the runner-up, Gemini-2.5-Pro.

This paper presents V2DGS, a novel multi-sensor fusion reconstruction system designed to enhance 2D Gaussian Splatting (2DGS) for outdoor scene reconstruction. Our method integrates LiDAR, camera, and IMU within a SLAM framework to jointly estimate camera poses and construct a surfel-based visual voxel map. Unlike conventional image-only pipelines that rely on Structure-from-Motion (SfM), V2DGS leverages geometric and photometric priors derived from this voxel map, where each surfel encodes position, color, scale and orientation. These rich priors are used to initialize 2D Gaussian primitives, significantly improving convergence and reconstruction quality. In addition, the SLAM-estimated poses are further refined through global bundle adjustment to enhance overall consistency. Experiments on the FAST-LIVO dataset demonstrate that our approach outperforms other Gaussian-based methods, including 2DGS, 3DGS, and SuGaR, in terms of both geometric accuracy and rendering efficiency.

Feed-forward 3D reconstruction models have recently shown strong generalization across diverse scenes, yet most of them recover geometry only up to an unknown global scale. This scale ambiguity limits their use in applications that require metric understanding of the environment. Existing metric reconstruction methods commonly rely on large-scale metric annotations or accurate camera calibration, both of which are costly or unreliable in many real-world settings. We propose a satellite-guided framework for resolving scale ambiguity in feed-forward 3D reconstruction. The key idea is to use readily available satellite imagery as a global metric reference. Given a coarse camera pose, our method retrieves a local satellite patch and integrates it with a feed-forward reconstruction backbone through bidirectional cross-view interaction. By enforcing consistency between the reconstructed scene and the satellite reference, the model infers absolute scale, refines scene geometry, and estimates camera pose in a metric coordinate frame. Experiments on KITTI, nuScenes, and Oxford RobotCar show consistent improvements in metric depth estimation, multi-view point-cloud reconstruction, and cross-view camera localization, while preserving strong generalization across datasets and geographic regions.

Adverse weather image restoration aims to recover clean background scenes from images degraded by various weather conditions, such as haze, rain, and snow. With the rapid development of deep learning, single-task restoration methods targeting specific weather types have achieved remarkable progress and attracted increasing attention in recent years. More recently, to address the limited generalization of task-specific models, All-in-One (AiO) methods have emerged to handle multiple degradations within a unified framework. However, existing surveys mostly focus on individual degradation types or specific restoration paradigms, and unified reviews of deep learning-based adverse weather restoration are still limited. In this paper, we present a comprehensive survey that jointly organizes single-task and AiO restoration models from the perspectives of network architectures and learning paradigms. We further review widely used datasets, loss functions, and evaluation metrics across different restoration tasks. In addition, we summarize benchmark results of representative methods on public datasets to analyze their performance and generalization ability. Finally, we discuss key challenges and promising research directions to support future developments in this rapidly evolving field.

In the field of recommendation systems, click-through rate (CTR) prediction is essential for measuring user engagement and product interest. However, input features are typically high-dimensional and sparse, requiring effective modeling of high-order feature interactions. Existing methods learn low-dimensional representations and identify useful feature combinations, but they often struggle to capture implicit interactions in non-Euclidean spaces, leading to noisy feature relationships, reduced interpretability, and suboptimal prediction performance. To address these challenges, we introduce a similarity-based adaptive graph attention network (SGAT) for CTR prediction. The SGAT employs a dual-attention mechanism that integrates similarity-based adaptive attention with softmax attention to selectively emphasize beneficial feature interactions while suppressing irrelevant ones. This strategy reduces noise, enhances interaction modelling, and mitigates overfitting. Experimental results on two public benchmark datasets show that SGAT achieves superior performance compared with several state-of-the-art baseline methods.

While hyperspectral imaging provides rich spatial-spectral information across hundreds of narrow wavelength bands for precise material identification, ground-based hyperspectral pre-trained backbones remain absent, constrained by varying spectral configurations across sensors, the scarcity and inconsistency of labels, and the limited scale and scene diversity of existing datasets. To address these challenges and enable universal perception, we propose HyperVision, the first ground-based hyperspectral pre-trained backbone. First, to handle varying spectral configurations, HyperVision adopts a channel-adaptive dynamic embedding mechanism to map heterogeneous inputs into a unified token space. Second, we develop an unsupervised representation learning framework. Specifically, to address label scarcity and inconsistency, a multi-source pseudo-labeling method is introduced to fuse spatial structures from SAM2 and fine-grained spectral material information from HyperFree. Furthermore, to enrich scene diversity and compensate for limited dataset scale, a cross-modal knowledge distillation mechanism is utilized to transfer rich semantic representations from a pre-trained RGB vision model to our backbone. Pre-trained on a collection of 15k images from 26 diverse ground-based datasets, HyperVision demonstrates exceptional generalization. Requiring only efficient head-only adaptation without adjusting backbone parameters, it achieves state-of-the-art performance compared to task-specific methods across three downstream tasks under varying sensor configurations, yielding up to a 16.3

Given the significant potential of 3D Gaussian Splatting for high-fidelity avatar reconstruction, we propose an end-to-end framework that reconstructs a high-quality 3D Gaussian head from a single face image without any 2D post-processing. Given a near-frontal face image, the network directly predicts a 2D UV-aligned Gaussian attribute map consistent with the FLAME template, enabling efficient sampling of Gaussian primitives and fast rendering from arbitrary viewpoints, including unseen rear views. To balance global identity information and fine details, we adopt a dual-branch design: a pretrained VQ-VAE–based face prior branch encodes identity and coarse geometry via a codebook. In contrast, a ViT-based branch extracts high-frequency texture details from the input image. These branches fuse in UV space to produce precise Gaussian attributes for position, scale, color, and opacity. This dual-branch setup maximally captures face-relevant information from the input. We train on a large synthetic multi-view dataset, using cross-view loss formulation and Gaussian attribute regularization to supervise rendered images. Experiments demonstrate that our method outperforms existing single-image reconstruction approaches on various metrics and excels in generalization, identity preservation, and multi-view consistency.

Cross-resolution change detection aims to identify changes between bitemporal images acquired from heterogeneous platforms (e.g., drones and satellites) with varying spatial resolutions. It plays a vital role in applications such as rapid disaster assessment. However, semantic inconsistencies caused by resolution differences and spatial misalignment resulting from imperfect registration pose significant challenges to robust change detection. To address these issues, this paper proposes a Semantic-Aware Alignment Network for cross-resolution change detection. For spatial alignment, a semantic-aware Transformer is introduced to establish global semantic correspondences and estimate geometric transformation parameters between the two images. To enforce semantic consistency, the high-resolution image is used as a reference and degraded to simulate low-resolution observations. By enforcing feature-level consistency between the high-resolution and low-resolution representations, the two inputs are mapped into a unified, resolution-invariant feature space, thereby substantially reducing semantic discrepancies. Benefiting from the joint semantic consistency learning and spatial alignment strategies, the proposed method outperforms existing state-of-the-art approaches, achieving F1-scores of 77.99%, 90.37%, and 51.40% on the HTCD, MRCDD, and DECD datasets, respectively.

Generating multiview-consistent $360^\circ$ ground-level scenes from satellite imagery is a challenging task with broad applications in simulation, autonomous navigation, and digital twin cities. Existing approaches primarily focus on synthesizing individual ground-view panoramas, often relying on auxiliary inputs like height maps or handcrafted projections, and struggle to produce multiview consistent sequences. In this paper, we propose SatDreamer360, a framework that generates geometrically consistent multi-view ground-level panoramas from a single satellite image, given a predefined pose trajectory. To address the large viewpoint discrepancy between ground and satellite images, we adopt a triplane representation to encode scene features and design a ray-based pixel attention mechanism that retrieves view-specific features from the triplane. To maintain multi-frame consistency, we introduce a panoramic epipolar-constrained attention module that aligns features across frames based on known relative poses. To support the evaluation, we introduce VIGOR++, a large-scale dataset for generating multi-view ground panoramas from a satellite image, by augmenting the original VIGOR dataset with more ground-view images and their pose annotations. Experiments show that SatDreamer360 outperforms existing methods in both satellite-to-ground alignment and multiview consistency.

Hyperspectral object tracking offers superior performance over conventional color-based tracking by leveraging rich spectral information to enhance material discrimination ability. Due to the limited availability of hyperspectral video datasets, many hyperspectral trackers rely on spectral correlation modeling to bridge hyperspectral images (HSIs) and color images. They are often combined with pre-trained deep feature extractors for robust representation. However, these methods face two key limitations: 1) they ignore the intrinsic relationship between the object template in the initial frame and the search image in the current frame during spectral correlation modeling. This limits the ability to distinguish spectral differences between objects and backgrounds; and 2) they insufficiently utilize temporal information, which prevents the construction of a robust spatial-spectral-temporal representation and thereby limits the improvement of tracking performance. To overcome these two issues, we propose CSSTrack, a novel unified network for end-to-end hyperspectral object tracking. First and foremost, we propose a spectral-aware representation enhancement (SaRE) module that employs physical models of spectral self-expression to perform cross-frame spectral correlations between the template and search images. Different from previous works, our method enhances the discrimination of foreground-background spectral differences, thereby facilitating the extraction of discriminative spatial-spectral features. Moreover, we design a spatial-spectral-temporal modeling (S2TM) module, which utilizes a sequence of autoregressive temporal embeddings to capture motion dynamics across spectral bands and integrates static and dynamic features through a fusion network. Extensive experiments on the HOT2020 and IMEC25 datasets demonstrate the effectiveness of our proposed CSSTrack, which achieves state-of-the-art tracking performance. The source code is available at https://github.com/hscv/CSSTrack.

Multimodal learning integrates data from diverse sensors to effectively harness information from different modalities. However, recent studies reveal that joint learning often overfits certain modalities while neglecting others, leading to performance inferior to that of unimodal learning. Although previous efforts have sought to balance modal contributions or combine joint and unimodal learning—thereby mitigating the degradation of weaker modalities with promising outcomes—few have examined the relationship between joint and unimodal learning from an information-theoretic perspective. In this paper, we theoretically analyze modality competition and propose a method for multimodal classification by maximizing the total correlation between multimodal features and labels. By maximizing this objective, our approach alleviates modality competition while capturing inter-modal interactions via feature alignment. Building on Mutual Information Neural Estimation (MINE), we introduce **T**otal **C**orrelation **N**eural **E**stimation (**TCNE**) to derive a lower bound for total correlation. Subsequently, we present TCMax, a hyperparameter-free loss function that maximizes total correlation through variational bound optimization. Extensive experiments demonstrate that TCMax outperforms state-of-the-art joint and unimodal learning approaches. Our code is available at https://anonymous.4open.science/r/TCMax_Experiments.

Point cloud sampling plays a crucial role in reducing computational cost while maintaining geometric and semantic fidelity for downstream 3D tasks. Traditional task-agnostic methods, such as Farthest Point Sampling (FPS), often suffer from the loss of discriminative structures, while task-specific approaches require additional training and lack generality. Even the emerging task-agnostic learnable samplers largely optimize for global representativeness, yet still neglect the critical preservation of fine-grained edge structures and localized geometric details. To address these issues, we propose SE-PCS, a task-agnostic sampling framework that combines edge-aware geometric priors with semantic constraints from self-supervised vision models. Our approach explicitly emphasizes high-curvature regions to retain structural details, while a DINOv3-driven semantic consistency loss encourages the sampled subset to capture discriminative features across tasks. Experiments on ModelNet40 and ShapeNet Core55 show that SE-PCS consistently surpasses classical and learning-based baselines in classification, retrieval, and completion. Remarkably, it achieves near-Oracle accuracy with only 256 points, demonstrating both efficiency and adaptability. These results highlight the effectiveness of integrating geometric cues with semantic guidance to obtain compact yet informative point cloud representations.

Contribution evaluation is essential for incentivizing high-quality data sharing in federated learning (FL), yet existing Shapley-value-based methods are prohibitively expensive and overlook temporal influence propagation. In this paper, we propose Ripple Shapley, a novel attribution framework that enables accurate, real-time data valuation within a single federated training run. Our method decomposes each sample’s impact into an instantaneous drop term and a recursive ripple term, the latter capturing downstream influence via a Jacobian chain over global updates. To scale computation, we introduce a low-rank approximation of the Jacobian product and construct a shared subspace for efficient ripple accumulation. Extensive experiments on CIFAR-10 and MNIST show that Ripple Shapley achieves up to 62× speedup over existing Shapley-based FL methods while maintaining high attribution fidelity, significantly improving efficiency, robustness, and fairness in federated environments. We further demonstrate its effectiveness in dynamic federated learning scenarios and its potential for real-time data pricing.

Vision Transformers (ViTs) have emerged as a dominant backbone for various visual tasks, yet their high computational cost limits deployment in real-time applications. While structured pruning offers a viable solution by removing entire neurons to enable hardware-friendly acceleration, it often leads to significant performance degradation due to coarse-grained pruning. This paper proposes a novel recovery algorithm for structured pruning in ViTs, leveraging multivariate linear regression to approximate pruned neurons using remaining ones, thereby mitigating accuracy loss. Unlike traditional methods requiring labeled data and retraining, our approach utilizes unlabeled data and statistical relationships between neurons for restoration. Specifically, we introduce a regression-based recovery mechanism where the regression matrix is absorbed into the model parameters, ensuring no additional inference overhead. Furthermore, we design a neuron selection strategy based on estimated recovery error to optimize pruning decisions, balancing computational efficiency and restoration quality. Experiments demonstrate that our method effectively preserves model performance while maintaining hardware compatibility, offering a practical solution for deploying lightweight ViTs in resource-constrained scenarios.

Hyperspectral object tracking significantly outperforms traditional color video-based tracking due to its material identification ability enabled by rich spectral bands. However, existing hyperspectral object tracking methods suffer from two critical limitations: the difficulty of learning discriminative features due to spectral band redundancy, and the challenge of fully modeling spatial-spectral-temporal (SST) dependencies. To address these limitations, we propose a novel video-based hyperspectral tracking framework, termed DRSST-Net, which integrates a dimensionality reduction module with a context-aware SST modeling module. The dimensionality reduction module employs a nonlinear, data-driven weighting strategy that integrates an intraband discrimination block and a cross-band weighting block. This design suppresses spectral redundancy while enhancing heterogeneity, thereby improving material discrimination ability. The context-aware SST modeling module integrates spatial-spectral features with temporal contextual information by combining the Mamba architecture with a Mixture-of-Experts (MoE) mechanism. The Mamba is employed to capture temporal dependencies, while the MoE dynamically activates task-relevant experts to enhance adaptability across diverse scenes. Extensive experiments on four benchmark datasets, e.g., HOT2020, HOT2022, IMEC25, and MUST, demonstrate that our DRSST-Net achieves state-of-the-art tracking performance, improving tracking accuracy and robustness. The source code will be available at https://github.com/hscv/DRSST-Net.

Recently, Graph Neural Networks (GNNs) have shown significant improvements in the recognition of neurological disorders by incorporating brain networks/graphs. However, most existing approaches have three main limitations. First, these methodologies rely on precomputed brain networks as input, typically derived from statistical metrics (e.g., Pearson correlation), which are inherently not learnable. Second, methods often assume that the magnitude of the brain interactions remains constant across the whole scan duration. Third, representations produced by models often lack interpretability and robustness when applied across brain disorders. To address these limitations, we propose a novel model called the Effective Brain Inference Graph Neural Network (EBIGNN), which infers dynamic Effective Connectivity (dEC) to characterize brain networks trained with direct feedback from downstream tasks within a unified end-to-end framework. EBIGNN is highly flexible in learning the most relevant graph structures customized to the specific underlying brain condition. The proposed model offers strong interpretability, providing valuable insights into the temporal evolution and altered connectivity patterns essential for understanding brain disorders. The model is validated on three publicly available datasets, demonstrating superior performance compared to other state-of-the-art methods. Moreover, the findings are consistent with previous neuroimaging-derived evidence of biomarkers, underscoring the model's robustness in clinical settings.

With the increasing necessity for hyperspectral images (HSIs) with fine spatial-spectral details, the hyperspectral (HS) and multispectral (MS) image fusion scheme becomes a consequential mechanism to attain such images due to physical and expense constraints for the standard imaging instruments. With the rapid advancement of deep learning (DL) in the field of image processing, the ability to model long-range dependencies in both HS and MS images has become crucial for achieving superior performance in HS-MS image fusion. Transformer-based models address this challenge effectively through attention mechanisms, whereas state-space models offer a more efficient alternative in terms of both computation and memory usage. Therefore, we devise a degradation-aware transformer-Mamba MoE (TM-MOE) for HS-MS image fusion, which leverages the benefits of the cross-attention mechanism and Mamba to enhance the fusion consequences. The core part of the devised framework is the suggested degradation-aware cross transformer-Mamba (DACTM) model, which leverages the capacity of Mamba in processing long-range relationships with improved training efficiency and the performance of the transformer's feature fusion for the observed pair. Moreover, the suggested Mamba module utilizes a 3-D scan to explore the correlation between pixels in an HSI, and it incorporates a mixture-of-experts (MoE) network to enrich model capability while maintaining effortless active parameter use. While the fusion process is accomplished on various scales, the novel feature accumulation and reconstruction block (FARB) devised reconstructs the final output from the fused features at different scales, gradually. Extensive experiments demonstrate the effectiveness of the proposed TM-MOE. The code of this study will be available on https://github.com/Abdolraheemkhader/TM-MOE

Mainstream Test-Time Adaptation (TTA) methods for adapting vision-language models, e.g., CLIP, typically rely on Shannon Entropy (SE) at test time to measure prediction uncertainty and inconsistency. However, since CLIP has a built-in bias from pretraining on highly imbalanced web-crawled data, SE inevitably results in producing biased estimates of uncertainty entropy. To address this issue, we notably find and demonstrate that Tsallis Entropy (TE), a generalized form of SE, is naturally suited for characterizing biased distributions by introducing a non-extensive parameter q, with the performance of SE serving as a lower bound for TE. Building upon this, we generalize TE into Adaptive Debiasing Tsallis Entropy (ADTE) for TTA, customizing a class-specific parameter q^l derived by normalizing the estimated label bias from continuously incoming test instances, for each category. This adaptive approach allows ADTE, even without hyperparameter tuning required by TE, to accurately select high-confidence views and seamlessly integrate with label adjustment strategy to enhance adaptation. Besides, our investigation reveals that both TE and ADTE can serve as direct, advanced alternatives to SE in TTA, without any other modifications. Experimental results show that ADTE outperforms state-of-the-art methods on ImageNet and its five variants, and achieves the highest average performance on 10 cross-domain benchmarks, regardless of the model architecture or text prompts used. Our code is available at https://anonymous.4open.science/r/TTA-Entropy.

Image processing networks are known to be vulnerable to adversarial examples, where adding carefully crafted adversarial perturbations to the inputs can mislead the model. This paper addresses the problem of robust blind face restoration (BFR) against adversarial attacks. BFR refers to recovering the HQ images from the LQ images, which suffer from diverse unknown degradation, such as noise, blur, artifact removal, low resolution, etc. Although existing BFR methods exhibit good performance, they experience significant degradation when subtle distortions and perturbations are introduced into the input images. This paper is the first to investigate, improve comprehensively, and evaluate BFR methods towards adversarial attacks. Project Gradient Descent (PGD) is employed to generate adversarial examples, and multiple types of attacks were used to thoroughly assess the robustness of various BFR methods across different objectives, regions, and levels. We evaluate the robustness of multiple BFR methods and analyze the advantages of their structures and modules towards adversarial attacks. Experimental results demonstrate that the method utilizing latent feature encoding and pre-trained discrete HQ codebook achieves better robustness than other methods, with the latter outperforming the former. Similarly, multi-scale semantic guidance information also exhibits superior performance in enhancing robustness. Therefore, we propose a powerful BFR method to mitigate this issue while maintaining better performance. Extensive experiments on three real-world datasets demonstrate our method’s state-of-the-art robustness in different scenarios.