Remote sensing change detection (RSCD) aims to identify the spatio-temporal changes of land cover, providing critical support for multidisciplinary applications (e.g., environmental monitoring, disaster assessment, and climate change studies). Existing methods focus either on extracting features from localized patches or pursue processing entire images holistically, which leads to the cross-temporal feature matching deviation and exhibits sensitivity to radiometric and geometric noise. Following the above issues, we propose a dual-module collaboration-guided hierarchical adaptive aggregation framework (HA2F), namely HA2F, which consists of a dynamic hierarchical feature calibration module (DHFCM) and a noise-adaptive feature refinement module (NAFRM). The former dynamically fuses adjacent-level features through perceptual feature selection, suppressing irrelevant discrepancies to address multitemporal feature alignment deviations. The NAFRM utilizes the dual feature selection mechanism to highlight the change-sensitive regions and generate spatial masks, suppressing the interference of irrelevant regions or shadows. Extensive experiments verify the effectiveness of the proposed HA2F, which achieves state-of-the-art performance on LEVIR-CD, WHU-CD, and SYSU-CD datasets, surpassing existing comparative methods in terms of both precision metrics and computational efficiency. In addition, ablation experiments show that DHFCM and NAFRM are effective. Code can be found at https://huggingface.co/InPeerReview/RemoteSensingChangeDetection-RSCD.HA2F.

Trajectory prediction for traffic agents is critical for safe autonomous driving. However, achieving effective zero-shot generalization in previously unseen domains remains a significant challenge. Motivated by the consistent nature of kinematics across diverse domains, we aim to incorporate domain-invariant knowledge to enhance zero-shot trajectory prediction capabilities. The key challenges include: 1) effectively extracting domain-invariant scene representations, and 2) integrating invariant features with kinematic models to enable generalized predictions. To address these challenges, we propose a novel generalizable Physics-guided Causal Model (PCM), which comprises two core components: a Disentangled Scene Encoder, which adopts intervention-based disentanglement to extract domain-invariant features from scenes, and a CausalODE Decoder, which employs a causal attention mechanism to effectively integrate kinematic models with meaningful contextual information. Extensive experiments on real-world autonomous driving datasets demonstrate our method's superior zero-shot generalization performance in unseen cities, significantly outperforming competitive baselines. The source code is released at https://github.com/ZY-Zong/Physics-guided-Causal-Model.

Active Speaker Detection (ASD) aims to identify who is currently speaking in each frame of a video. Most state-of-the-art approaches rely on late fusion to combine visual and audio features, but late fusion often fails to capture fine-grained cross-modal interactions, which can be critical for robust performance in unconstrained scenarios. In this paper, we introduce GateFusion, a novel architecture that combines strong pretrained unimodal encoders with a Hierarchical Gated Fusion Decoder (HiGate). HiGate enables progressive, multi-depth fusion by adaptively injecting contextual features from one modality into the other at multiple layers of the Transformer backbone, guided by learnable, bimodally-conditioned gates. To further strengthen multimodal learning, we propose two auxiliary objectives: Masked Alignment Loss (MAL) to align unimodal outputs with multimodal predictions, and Over-Positive Penalty (OPP) to suppress spurious video-only activations. GateFusion establishes new state-of-the-art results on several challenging ASD benchmarks, achieving 77.8

Black-box adversarial attacks remain challenging due to limited access to model internals. Existing methods often depend on specific network architectures or require numerous queries, resulting in limited cross-architecture transferability and high query costs. To address these limitations, we propose JAD, a latent diffusion model framework for black-box adversarial attacks. JAD generates adversarial examples by leveraging a latent diffusion model guided by attention maps distilled from both a convolutional neural network (CNN) and a Vision Transformer (ViT) models. By focusing on image regions that are commonly sensitive across architectures, this approach crafts adversarial perturbations that transfer effectively between different model types. This joint attention distillation strategy enables JAD to be architecture-agnostic, achieving superior attack generalization across diverse models. Moreover, the generative nature of the diffusion framework yields high adversarial sample generation efficiency by reducing reliance on iterative queries. Experiments demonstrate that JAD attack offers improved attack generalization, generation efficiency, and cross-architecture transferability compared to existing methods, providing a promising and effective paradigm for black-box adversarial attacks.

Low-light imaging is inherently challenging because images captured under insufficient illumination often suffer from poor quality. In this paper, we introduce a new invertible flow framework, which maps low-light images to latent data distribution. Then the parameter of the conditional distribution can be estimated by the proposed attention-flow network. Specifically, the network combines multi-scale attention mechanisms with flow-based generative models to enhance low-light images. Owing to the property of invertibility in the network, it effectively captures and reconstructs crucial visual details and contextual information, producing images that are perceptually realistic and free from artifacts. Experimental results on several public low-light datasets demonstrate the superior performance of the proposed method compared to existing approaches, highlighting its robustness across a range of low-light conditions. The source code is available at https: //github.com/Rehgar3/AttnFlow.

Cybersecurity experts are actively exploring and implementing automated technologies to extract and present attack information from Cyber Threat Intelligence. However, there are multiple relations among security entities within Cyber Threat Intelligence, a feature that existing technologies often overlook. Additionally, integrating external security knowledge into cyber threat intelligence intuitively during analysis and presentation poses challenges. We propose the Hyper Attack Graph (HAG) framework, the first work to apply hypergraph data structures in the analysis of cyber threat intelligence. Our approach uses a joint extraction model that incorporates a multi-head selection mechanism, effectively addressing the extraction of multiple relations among security entities. We use hypergraph to display tactics and techniques in cyber threat intelligence. Our evaluation of the HAG framework on 685 real-world cyber threat intelligence reports shows an increase in the F1 score for security entity extraction by 11.12% and for relation extraction by 6.71% over existing efforts. Furthermore, HAG’s ability to visually represent external security knowledge on hypergraphs demonstrates its potential as a valuable tool in cybersecurity analysis.

We introduce Ego4D, a massive-scale egocentric video dataset and benchmark suite. It offers 3,670 hours of daily-life activity video spanning hundreds of scenarios (household, outdoor, workplace, leisure, etc.) captured by 931 unique camera wearers from 74 worldwide locations and 9 different countries. The approach to collection is designed to uphold rigorous privacy and ethics standards with consenting participants and robust de-identification procedures where relevant. Ego4D dramatically expands the volume of diverse egocentric video footage publicly available to the research community. Portions of the video are accompanied by audio, 3D meshes of the environment, eye gaze, stereo, and/or synchronized videos from multiple egocentric cameras at the same event. Furthermore, we present a host of new benchmark challenges centered around understanding the first-person visual experience in the past (querying an episodic memory), present (analyzing hand-object manipulation, audio-visual conversation, and social interactions), and future (forecasting activities). By publicly sharing this massive annotated dataset and benchmark suite, we aim to push the frontier of first-person perception. Project page: https://ego4d-data.org/

Advanced Persistent Threats (APTs) pose significant security risks to the community. Researchers thereby propose techniques to capture the complex and stealthy scenarios of APT attacks through the use of provenance graphs to model system entities and their dependencies. Particularly, to mitigate the dependency explosion problem in attack investigation using provenance graphs, tag-based and priority-based provenance graphs are frequently utilized for analyzing attacks. These methods use threat tag propagation and threat prioritization to reduce the size of the provenance graph for faster analysis. Unfortunately, these methods can allow more complex and potential attacks to evade detection. To overcome these difficulties, we propose an APT attack investigation system, ProTaging, for APT detection and forensic analysis. By using Tactics, Techniques, and Procedures (TTPs) rules to assign and update the node's threat tag, splittable tag propagation to control the scope of threat information, and threat weight aggregation and prioritized backward analysis during the forensic analysis phase, ProTaging effectively reconstructs attack paths in seconds without dependency explosion. Experimental results on both the simulation dataset, DARPA TC E3, E5 dataset, and DARPA OpTC dataset demonstrate that ProTaging generates smaller dependency graphs (2.5 times smaller) and has fewer false positives (6.7 times fewer) compared to state-of-the-art solutions. Additionally, ProTaging significantly reduces manual investigation effort by approximately 99.9%.

The increasing popularity of egocentric cameras has generated growing interest in studying multi-camera interactions in shared environments. Although large-scale datasets such as Ego4D and Ego-Exo4D have propelled egocentric vision research, interactions between multiple camera wearers remain underexplored-a key gap for applications like immersive learning and collaborative robotics. To bridge this, we present TF2025, an expanded dataset with synchronized first- and third-person views. In addition, we introduce a sequence-based method to identify first-person wearers in third-person footage, combining motion cues and person re-identification.

As the complexity and frequency of cyberattacks, such as Advanced Persistent Threats (APTs) and ransomware, continue to escalate, traditional anomaly detection methods have proven inadequate in addressing these sophisticated, multi-faceted threats. Recently, Host Provenance Graphs (HPGs) have played a crucial role in analyzing system-level interactions, detecting anomalous behaviors, and tracing attack chains. However, existing provenance-based detection methods primarily rely on single-dimensional feature analysis, which fails to capture the dynamic and multi-dimensional patterns of modern APT attacks, resulting in insufficient detection performance. To overcome this limitation, we introduce STGAN, a model that integrates spatial-temporal graphs into host provenance graph modeling. STGAN applies temporal and spatial encoding to dynamic provenance graphs to extract temporal, spatial, and semantic features, constructing a comprehensive feature representation. This representation is further fused and enhanced using a multi-head self-attention mechanism, followed by anomaly detection. Through extensive evaluations on three widely-used provenance graph datasets, we demonstrate that our approach consistently outperforms current state-of-the-art techniques in terms of detection performance. Additionally, we contribute to the research community by releasing our datasets and code, facilitating further exploration and validation.

Understanding individuals' trip purposes is essential for transportation planning, land use evaluation, and sustainable urban development. Existing methods face challenges in generalizability, scalability, and reliance on labeled data. This study proposes a novel tensor-based method for inferring trip purposes using large-scale mobile signaling (MS) data. First, user trajectories are encoded into a structured third-order tensor representation that captures interactions among cell sectors, temporal context, and semantic features. Second, using the prior distribution of trip purposes, we develop a Bayesian log-normal decomposition model that factorizes the tensor into latent components, infers trip purposes, and reconstructs activity chains. Finally, a subdistrict-based expansion method is introduced to scale individual-level inferences to the population level by adjusting the market shares of mobile network operators. The proposed method is evaluated on a large-scale MS dataset collected in Nanjing, China. The effectiveness of the method is validated through a multi-perspective evaluation with travel survey data and high-confidence labels generated by strict rules. The results demonstrate that the model can accurately infer trip purposes without relying on labeled data. This study provides a scalable, interpretable solution for trip purpose inference and contributes to developing data-driven methods for urban mobility management and planning.

DNN(Deep Neural Networks)-based image/video watermarking models have shown outstanding robustness against various distortions. However, because the size of DNN-based watermarking models is typically large, it is not practical to deploy them in resource-constrained devices like laptops and smartphones. In this paper, we propose a two-stage approach based on a synergy of model pruning and knowledge distillation to compress DNN-based watermarking models. First, we propose to evaluate the importance of each module in the model through model pruning techniques and identify the redundant and less essential modules that can be safely removed without significantly degrading the model’s accuracy. Then, based on the importance of each module revealed by the pruning process, we design a compact student model via knowledge distillation during the model distillation phase. The student model retains the essential modules while simplifying or removing the less crucial ones. Through knowledge distillation, the student model can learn the expressive power of the original complex model while achieving a much lower computational cost. We can obtain a tailored student model that maintains the performance level of the original model while being significantly more efficient and suitable for deployment on resource-constrained devices. Extensive experiments demonstrate that this framework provides an effective solution for model compression and efficient model deployment.

Notwithstanding the tremendous success of deep neural networks in a range of realms, previous studies have shown that these learning models are exposed to an inherent hazard called adversarial example — images to which an elaborate perturbation is maliciously added could deceive a network, which entails the study of countermeasures urgently. However, existing solutions suffer from some weaknesses, e.g. parameters are usually determined empirically in some processing-based detection methods might result in a sub-optimal effect, and the directly performed processing on images might affect the classification of benign samples, leading to increment of false positive. In this paper, we propose a novel imAge-DepenDent noIse reducTION (ADDITION) model based on deep learning for adversarial detection. The ADDITION model can adaptively convert the adversarial perturbation in each image to approximate Gaussian noise by injecting image-dependent additional noise, then perform noise reduction to eliminate the adversarial perturbation, and finally detect adversarial examples by examining the classification inconsistency between the input image and its denoised version. The ADDITION model is trained end-to-end on benign samples without any prior knowledge of adversarial attacks, and thus avoid time-consuming task of generating adversarial examples in practical use. We generate more than 220,000 adversarial examples based on six attack algorithms for evaluation and present state-of-the-art comparisons on three real-word datasets. Extensive experiments demonstrate that our proposed method achieves improved performance in both detection accuracy rate and false positive rate.

Visible and infrared image fusion (VIF), combining thermal information with textural details for more informative output, has attracted widespread research interest. However, current studies in VIF primarily focus on enhancing the image quality of fusion results, while the robustness against adversarial attacks remains largely unexplored. The VIF system might be deceived to produce inferior output that affects the decision of subsequent high-level tasks, leading to potential security accidents. This work aims to bridge this gap. We focus on degrading the fusion quality by introducing an adversarial attack, which injects imperceptible perturbation into input images to deteriorate the fusion result. To this end, we design an adaptive target generation method to establish a benchmark that indicates the ideal attack effect, which is used to guide the generation of perturbations. Extensive experiments demonstrate that our attack effectively impairs the performance of VIF models, while concurrently exerting a notable impact on subsequent tasks.

As wearable cameras become more popular, an important question emerges: how to identify camera wearers within the perspective of conventional static cameras. The drastic difference between first-person (egocentric) and third-person (exocentric) camera views makes this a challenging task. We present PersonEnvironmentNet (PEN), a framework designed to integrate information from both the individuals in the two views and geometric cues inferred from the background environment. To facilitate research in this direction, we also present TF2023, a novel dataset comprising synchronized first-person and third-person views, along with masks of camera wearers and labels associating these masks with the respective first-person views. In addition, we propose a novel quantitative metric designed to measure a model's ability to comprehend the relationship between the two views. Our experiments reveal that PEN outperforms existing methods. The code and dataset are available at https://github.com/ ziweizhao1993/PEN.

DNN-based watermarking methods are rapidly developing and delivering impressive performances. Recent advances achieve resolution-agnostic image watermarking by reducing the variant resolution watermarking problem to a fixed resolution watermarking problem. However, such a reduction process can potentially introduce artifacts and low robustness. To address this issue, we propose the first, to the best of our knowledge, Resolution-Agnostic Image WaterMarking (RAIMark) framework by watermarking the implicit neural representation (INR) of image. Unlike previous methods, our method does not rely on the previous reduction process by directly watermarking the continuous signal instead of image pixels, thus achieving resolution-agnostic watermarking. Precisely, given an arbitrary-resolution image, we fit an INR for the target image. As a continuous signal, such an INR can be sampled to obtain images with variant resolutions. Then, we quickly fine-tune the fitted INR to get a watermarked INR conditioned on a binary secret message. A pre-trained watermark decoder extracts the hidden message from any sampled images with arbitrary resolutions. By directly watermarking INR, we achieve resolution-agnostic watermarking with increased robustness. Extensive experiments show that our method outperforms previous methods with significant improvements: averagely improved bit accuracy by 7%similar to 29%. Notably, we observe that previous methods are vulnerable to at least one watermarking attack (e.g. JPEG, Crop, or Resize), while ours are robust against all watermarking attacks.

Reconstructing the three-dimensional structure of a scene is a classic and fundamental problem in computer vision, but it has been revolutionized by recent progress in deep machine learning. In this paper, we survey this rich and growing area. We divide the work into four main threads: 3d reconstruction from two calibrated images from a binocular camera; 3d reconstruction from more than two images taken by the same camera or more than two calibrated cameras; object-focused 3D reconstruction with relaxed camera calibration; and SLAM-based techniques. We summarize each approach along five salient dimensions: algorithmic and deep network characteristics, output representation, datasets, and quantitative comparisons among different methods. We also discuss key challenges and future directions.

To accurately segment rock particles in the blasted rock pile image, this study proposes a new segmentation method based on fully residual rock segmentation network plus (FRRSnet + ), where FRRSnet + is a new deep learning network proposed by fusing U-net, ResNet, and ASPP. The method consists of two models, FRRSnet +-1 and FRRSnet +-2, where FRRSnet +-1 is responsible for segmenting rocks and background, and FRRSnet +-2 is responsible for complementing the rock edge of FRRSnet +-1 result. Compared with U-net and Segnet showed that better results were obtained by directly using FRRSnet +-1, which had MRE, RMSE, and R2 of 2.5%, 1.55, and 99.58%, respectively. More, after using the proposed segmentation method, the segmentation results can be reduced by 1.54% and 0.77% for MRE and RMSE, respectively, improved by 0.31% for R2 compared with the results of directly using FRRSnet +-1. The proposed network and segmentation method can provide a more accurate guidance for blast parameter optimization.

Video anomaly detection (VAD) has been extensively studied for static cameras but is much more challenging in egocentric driving videos where the scenes are extremely dynamic. This paper proposes an unsupervised method for traffic VAD based on future object localization. The idea is to predict future locations of traffic participants over a short horizon, and then monitor the accuracy and consistency of these predictions as evidence of an anomaly. Inconsistent predictions tend to indicate an anomaly has occurred or is about to occur. To evaluate our method, we introduce a new large-scale benchmark dataset called Detection of Traffic Anomaly (DoTA)containing 4,677 videos with temporal, spatial, and categorical annotations. We also propose a new VAD evaluation metric, called spatial-temporal area under curve (STAUC), and show that it captures how well a model detects both temporal and spatial locations of anomalies unlike existing metrics that focus only on temporal localization. Experimental results show our method outperforms state-of-the-art methods on DoTA in terms of both metrics. We offer rich categorical annotations in DoTA to benchmark video action detection and online action detection methods. The DoTA dataset has been made available at: https://github.com/MoonBlvd/Detection-of-Traffic-Anomaly

Edge intelligence (EI) migrates data and artificial intelligence (AI) to the “edge” of a network, enhancing the high-bandwidth and low-latency of wireless data transmission with the multiplier effect of 5G and AI, greatly improving the edges’ processing speed. Through integrating EI and computer vision technology, video surveillance systems in ITS can improve the processing capability of traffic information, which improves traffic efficiency and ensures traffic safety. Accordingly, first, we propose an edge intelligence-based improved-YOLOv4 vehicle detection algorithm, introducing an efficient channel attention (ECA) mechanism and a high-resolution network (HRNet) to enhance vehicle detection ability. Second, an edge intelligence-based improved DeepLabv3+ image segmentation algorithm is proposed, replacing the original backbone network with MobileNetv2 and using the softpool method, thus reducing the network size while improving the segmentation accuracy. Experimental results show that our proposed model has a higher average precision (AP) and can improve vehicle detection accuracy from 82.03% to 86.22%. The mean intersection over union (mIOU) of the image segmentation model improves from 73.32% to 75.63%.