Infrared-visible image fusion aims to integrate infrared and visible information into a single fused image. Existing 2D fusion methods focus on fusing images from fixed camera viewpoints, neglecting a comprehensive understanding of complex scenarios, which results in the loss of critical information about the scene. To address this limitation, we propose a novel Infrared-Visible Gaussian Fusion (IVGF) framework, which reconstructs scene geometry from multimodal 2D inputs and enables direct rendering of fused images. Specifically, we propose a cross-modal adjustment (CMA) module that modulates the opacity of Gaussians to solve the problem of cross-modal conflicts. Moreover, to preserve the distinctive features from both modalities, we introduce a fusion loss that guides the optimization of CMA, thus ensuring that the fused image retains the critical characteristics of each modality. Comprehensive qualitative and quantitative experiments demonstrate the effectiveness of the proposed method.

Unlike traditional target tracking tasks, UAV tracking scenarios in consumer electronics applications present complex challenges such as background interference, target occlusion, and aerial perspective change, which directly affect tracking performance. To tackle these challenges, we propose FCADtrack, a Spatial Context Awareness-Based Adaptive Real-Time UAV Tracking method for consumer electronics. FCADtrack focuses on fine-tuning to meet UAV-specific requirements and introduces a spatial information interaction mechanism to enhance tracking robustness. First, it incorporates an Adapter Fine-Tuning Module into the ViT backbone, trained on UAV-specific datasets, enabling the model to capture distinctive UAV tracking features and improve adaptability in complex scenarios. Additionally, the Feature Enhanced Contextual Attention (FECA) module is introduced, comprising the Local Spatial Feature Enhanced Module (FEM) and the Global Spatial-Channel Attention Module (SCAM). The FEM extracts rich target features and suppresses background interference through diverse convolutional operations, while the SCAM enhances global feature representation by integrating spatial and channel information using Global Average Pooling (GAP) and Global Max Pooling (GMP). These components significantly boost target feature representation and reduce background interference. The proposed FCADtrack has achieved competitive tracking performance compared with existing methods. Furthermore, it runs a real-time processing speed of 30.3 FPS on the Jetson Orin NX16 platform, demonstrating its potential for real-time consumer electronics applications. Code is released at: https://github.com/qin490/FCADtracker.

The current two-stage tracking method locates the target using the position with the highest confidence score, and updates the template using a carefully designed template update strategy. However, we identify two key issues with these trackers: (1) the update strategy lacks continuous, cost-free template adaptation, leading to suboptimal tracking under appearance changes and (2) the location with the highest confidence score does not always yield accurate bounding boxes, potentially resulting in incomplete target coverage. In this article, we propose a novel tracker that incorporates two key innovations. First, the tracker employs a dual update strategy that performs online template updates at both the image and feature levels. This strategy enables continuous adaptation to target appearance changes without introducing additional computational overhead. Second, we enhance the existing loss function by introducing a Classification-Regression Interaction (CRI) loss, which guides the training process to produce confidence scores that more accurately reflect the quality of the predicted bounding boxes. Extensive experiments are conducted to evaluate the performance of our tracker and the effectiveness of the proposed methods. The experimental results show that our method has achieved a comprehensive improvement over the baseline on five datasets, and achieves competitive performance compared to state-of-the-art trackers.

Existing visual trackers mainly operate in a non-interactive, fire-and-forget manner, making them impractical for real-world scenarios that require human-in-the-loop adaptation. To overcome this limitation, we introduce Interactive Tracking, a new paradigm that allows users to guide the tracker at any time using natural language commands. To support research in this direction, we make three main contributions. First, we present InteractTrack, the first large-scale benchmark for interactive tracking, containing 150 videos with dense bounding box annotations and timestamped language instructions. Second, we propose a comprehensive evaluation protocol and evaluate 25 representative trackers, showing that state-of-the-art methods fail in interactive scenarios—strong performance on conventional benchmarks does not transfer. Third, we introduce Interactive Memory-Augmented Tracking (IMAT), a new baseline that employs a dynamic memory mechanism to learn from user feedback and update tracking behavior accordingly. Our benchmark, protocol, and baseline establish a foundation for developing more intelligent, adaptive, and collaborative tracking systems, bridging the gap between automated perception and human guidance.

Robust multimodal sensing is essential for reliable object tracking in complex environments. Although RGB-thermal (RGBT) sensing leverages complementary spectral modalities to enhance tracking performance, existing methods often rely on static template matching, thereby neglecting the inherent temporal continuity of sensor streams. This limitation compromises tracking stability under challenging sensing conditions, such as abrupt illumination changes, severe occlusions, and thermal crossover. To address these challenges, this article proposes TFTrack, a new type of closed-loop video-stream RGBT tracking framework that has adopted a deep-collaborative two-component time design: a dynamic template library, which provides spatial priors for trajectory modeling by storing the prediction box results of historical templates; and autoregressive time label propagation, which compensates for the continuity deficiency of pure template methods. Unlike traditional sparse sampling paradigms, TFTrack exploits continuous temporal informatics by integrating historical appearance features and motion trajectories within a dynamic template library, ensuring consistent perception across time. Furthermore, to maximize multisensor synergy, we develop an input-feature-guided attention-fusion (IFGAFusion) module. By dynamically recalibrating spatial importance weights under the guidance of input features, this module effectively fuses heterogeneous data streams from RGB and thermal infrared (TIR) sensors. Extensive qualitative and quantitative experiments demonstrate that the proposed method achieves state-of-the-art performance. Notably, TFTrack operates at 30 FPS on the LasHeR benchmark with a precision rate of 74.7% and a success rate of 59.6%, satisfying the stringent real-time and reliability requirements for practical vision-based sensing systems.

Frontier commercial generative models face a growing threat from distillation, whereby a distiller harvests generated responses and trains a competing model of its own at drastically lower cost. Existing defenses either rely on modifying the models outputs, thereby sacrificing response quality for benign users, or on behavioral detection methods, which can be readily circumvented by distributing queries across multiple accounts. In this work, we propose Lossless Anti-Distillation Sampling (LADS), a novel sampling scheme specifically designed to counter multi-account distillation while maintaining a lossless experience for benign users. Concretely, LADS derives the randomness underlying each generation from a private seed determined by the semantic content of the query and the number of times the user has queried the model. By construction, every benign user receives a response independently sampled from the original model at each visit, and thus experiences no distortion. In contrast, for a distiller, different accounts share latent randomness whenever their queries fall in the same semantic bucket. As a result, the harvested data becomes correlated, potentially reducing sample diversity and degrading generalization. Using uniform convergence theory, we show that LADS provably degrades the convergence rate of the distillers generalization gap relative to standard i.i.d. sampling in both unconditional and conditional generation settings. Experiments on image generation, mathematical reasoning, and code generation confirm that LADS substantially degrades the performance of distilled students while preserving exact statistical fidelity for individual users.

Virtual machine (VM) consolidation strategies are widely deployed in Cloud Data Centers (CDCs) to optimize resource utilization and improve the Quality of Service (QoS). However, the host overload detection algorithms in current VM consolidation strategies are static. That means, once the overload threshold is calculated, it will not change until the next recalculation. The current algorithms are not suitable for the environment of highly dynamic workloads which results in additional energy consumption and potential Service Level Agreement Violations (SLAVs) which will affect the QoS of CDC. In PMPHD, a novel host dynamic threshold adjustment algorithm is proposed. In the proposed algorithm, the PMs are classified into mildly overloaded, normal, and severely overloaded based on the resource utilization. If the PM is predicted to be severely overloaded in the next moment, the threshold of this PM will be proactively reduced. The PM is determined to be overloaded, and some VMs in this PM will be migrated in advance. Thus, this PM will be in normal in the next moment, and the VM performance degradation resulting from SLAV and VM migration overlap in the next moment will be avoided. If the PM is predicted to be mildly overloaded, the threshold will be appropriately increased to transit it to be in normal state in the next moment, and the VM in the PM will not be migrated. Since the PMs’ workloads are dynamic, the PMPHD overload algorithm predicts the resource utilization rate of PM continuously, and adjusts the overload threshold of PM. Compared with other algorithms, PMPHD maintains high efficiency while having lower ESV (a combination metric for balancing energy consumption and SLAV).

Visible and infrared image fusion aims to integrate complementary information from both modalities to produce high-quality fused images that enhance downstream computer vision tasks. However, existing fusion methods tend to use equal weight to fuse images, ignoring the different importance of the two modalities in spatial distribution and the need for dynamic adjustment of fusion methods in different scenarios. To address these limitations, we propose a novel RGB-T fusion method via semantic-guided attention and a dynamic gate mechanism to improve the robustness towards different scenarios. Specifically, by incorporating semantic attention importance maps, we propose a dual-modal semantic-driven feature alignment module that comprises a cross-modal query compensation module and an intra-modal query enhancement module, which explore the varying significance of different spatial regions in the two input images. Subsequently, we introduce a dynamic multi-path gating mechanism that enables the network to adjust the weights of each module according to the input of different scenarios, which ultimately improves the robustness of the fusion algorithm across various scenarios. Comprehensive experiments conducted on four benchmark datasets demonstrate that our approach achieves state-of-the-art performance in both qualitative and quantitative evaluations.

The crux of Referring Video Object Segmentation (RVOS) lies in modeling dense text-video relations to associate abstract linguistic concepts with dynamic visual contents at pixel-level. Current RVOS methods typically use vision and language models pretrained independently as backbones. As images and texts are mapped to uncoupled feature spaces, they face the arduous task of learning Vision-Language (VL) relation modeling from scratch. Witnessing the success of Vision-Language Pretrained (VLP) models, we propose to learn relation modeling for RVOS based on their aligned VL feature space. Nevertheless, transferring VLP models to RVOS is a deceptively challenging task due to the substantial gap between the pretraining task (static image/region-level prediction) and the RVOS task (dynamic pixel-level prediction). To address this transfer challenge, we introduce a framework named VLP-RVOS which harnesses VLP models for RVOS through temporal-aware adaptation. We first propose a temporal-aware prompt-tuning method, which not only adapts pretrained representations for pixel-level prediction but also empowers the vision encoder to model temporal contexts. We further customize a cube-frame attention mechanism for robust spatial-temporal reasoning. Besides, we propose to perform multi-stage VL relation modeling while and after feature extraction for comprehensive VL understanding. Extensive experiments demonstrate that our method performs favorably against state-of-the-art algorithms and exhibits strong generalization abilities.

Existing hyperspectral object tracking methods are limited by data scarcity, spectral band gaps, and the high computational cost associated with the large number of bands, which severely hinders progress in this field. To address these challenges, we propose a hyperspectral object tracking method based on general expert adapter (EA-Track). Specifically, we first design a multi-stream joint feature extraction module to capture visual-spectral information across different bands. Then, we introduce a general hyperspectral adapter module to achieve adaptive interaction of spectral information across bands. Furthermore, to exploit the latent visual-spectral-spatial features among different bands, we incorporate a mixture-of-experts model, enabling accurate tracking in complex scenarios. Finally, experiments on the HOTC2022 and IMEC25 datasets demonstrate the superior performance of our proposed method.

Leveraging multiple templates across video frames has become a popular direction in visual object tracking to enhance robustness against target appearance variations. However, existing multi-template tracking methods suffer from a significant increase in computational complexity as the number of templates grows, which severely degrades inference speed. In this work, we propose a novel tracking framework based on the only-decoder paradigm, which redefines tracking as a template-guided feature extraction process. To this end, we propose a unified attention mechanism that integrates self-attention and cross-attention, enabling the model to efficiently extract discriminative features from the search region while incorporating guidance from multiple templates. This design ensures that the growth in complexity slows as the number of templates increases, thus dramatically improving inference efficiency. To further enhance runtime performance, we incorporate a Key–Value caching mechanism to eliminate redundant computations associated with template features during inference. In addition, we employ a Kalman filter to model target motion across video frames, refining the tracker’s predictions and further improving tracking accuracy. Extensive experiments show that the tracking framework we proposed has a significant advantage in speed compared to other multi-template tracking frameworks, and it also maintains very competitive accuracy on multiple datasets. The source code is available at: https://github.com/cute-koala/ODVTrack/tree/main.

RGB-optical flow dual-stream architectures constitute a key approach for zero-shot video object segmentation. However, these methods suffer from excessive dependence on optical flow, with performance severely degraded when optical flow estimation fails in challenging scenarios such as occlusion and static scenes. To address this issue, we propose an uncertainty-aware multimodal adaptive fusion framework with mixture-of-experts for zero-shot video object segmentation, dubbed as UEMF. It incorporates a Cross-Modal Uncertainty Correlation mechanism (CMUC), an Uncertainty-guided Feature Purification Module (UFPM), and a Mixture-of-Experts (MoE) adaptive fusion strategy to establish a tri-modal architecture integrating RGB, optical flow, and depth information, enhancing feature representation and achieving more robust segmentation. Comprehensive experiments on DAVIS2016, FBMS, and YTOBJ datasets demonstrate superior performance. Experimental results show that UEMF effectively leverages depth information to maintain segmentation accuracy when optical flow fails, successfully compensating for performance degradation caused by unreliable optical flow.

Chain-of-Thought (CoT) prompting has emerged as a powerful technique for enhancing language model's reasoning capabilities. However, generating long and correct CoT trajectories is challenging. Recent studies have demonstrated that Looped Transformers possess remarkable length generalization capabilities, but their limited generality and adaptability prevent them from serving as an alternative to auto-regressive solutions. To better leverage the strengths of Looped Transformers, we propose RELAY (REasoning through Loop Alignment iterativelY). Specifically, we align the steps of Chain-of-Thought (CoT) reasoning with loop iterations and apply intermediate supervision during the training of Looped Transformers. This additional iteration-wise supervision not only preserves the Looped Transformer's ability for length generalization but also enables it to predict CoT reasoning steps for unseen data. Therefore, we leverage this Looped Transformer to generate accurate reasoning chains for complex problems that exceed the training length, which will then be used to fine-tune an auto-regressive model. We conduct extensive experiments, and the results demonstrate the effectiveness of our approach, with significant improvements in the performance of the auto-regressive model. Code will be released at https://github.com/qifanyu/RELAY.

RGB-T detectors leverage the complementary strengths of visible and thermal infrared modalities, achieving robust performance under challenging conditions. Many of them resort to heavy dual backbones and exhaustive cross-modality fusion across the entire image, leading to impractically high computational costs. We observe that most image regions are smooth backgrounds (e.g., sky, ground) that can be easily handled by lightweight single-modality models. In light of this observation, we propose a sparse fusion mechanism for efficient RGB-T detection: first rapidly scanning the image to identify the proposals and then carefully examining the remaining sparse proposals via feature fusion. We propose a two-stage framework to instantiate this mechanism, which performs detection in two stages: 1) a lightweight and modality-specific detection stage that produces high-recall RoIs, and 2) a fusion-driven examination and refinement stage that filters out the false positives and refines the bounding boxes. This design enables the detector to adaptively allocate more computational resources to the potential foregrounds, improving the efficiency while ensuring detection accuracy. Extensive experiments show that our method achieves competitive performance with substantially fewer parameters and lower cost, while maintaining strong scalability to high-resolution images.

The task of visible-to-infrared (V2I) person re-identification (ReID) presents greater challenges than visible-to-visible (V2V) ReID, primarily due to limited availability of training samples and significant cross-modal discrepancy. To address these challenges, we propose Flow2Flow, a unified framework designed to simultaneously expand training samples and generate cross-modal images for V2I person ReID. Flow2Flow operates by learning bijective transformations from both the visible and infrared image domains to a shared isotropic Gaussian domain, utilizing invertible flow-based generators for each modality. This framework enables the generation of training samples by transforming latent Gaussian noise into visible or infrared images and the generation of cross-modal images by transforming existing modality images through the latent Gaussian space into the target modality. To ensure proper identity and modality alignment of the generated images, we devise two adversarial training strategies. Specifically, we design an image encoder and a modality discriminator for each modality. The image encoder enhances the similarity between generated images and real images of the same identity through identity adversarial training, while the modality discriminator ensures the generated images are indistinguishable from real images through modality adversarial training. Experimental results on the SYSU-MM01 and RegDB datasets demonstrate that both training sample generation and cross-modal image generation substantially improve V2I ReID accuracy.

Recent deep learning approaches seek to automate CAD creation by representing a model as a sequence of discrete commands and parameters, and then generating them using autoregressive models or continuous diffusion operating in Euclidean embedding space. However, continuous diffusion perturbs representations in a continuous Euclidean domain that does not reflect the inherently discrete and heterogeneous nature of CAD tokens, often producing perturbed representations that map to semantically invalid symbols. To overcome this limitation, we propose a cascaded discrete diffusion framework for CAD generation, which consists of a command diffusion for generating CAD commands and a parameter diffusion conditioned on CAD commands. Unlike isotropic Gaussian perturbation, the forward process of our approach operates directly over categorical token distributions using delicate transition matrices. For commands, we adopt an absorbing-state transition matrix that progressively corrupts tokens to a designated symbol; for parameters, we introduce specific transition matrices tailored to heterogeneous attributes: a Gaussian kernel for coordinate continuity, a scale-invariant kernel for dimensional values, and a prior-preserving kernel for boolean attributes. The reverse process is achieved by two denoising networks: a Transformer-based encoder for command recovery, and a parameter network with extra local self-attention for command-level interaction and cross-attention for conditional injection. Experiments on the DeepCAD dataset show that the proposed approach surpasses existing autoregressive and continuous diffusion models on unconditional generation metrics, while qualitative results validate effective controllability in conditional generation tasks. Source codes will be released.

The advantage of RGB-Thermal (RGB-T) detection lies in its ability to perform modality fusion and integrate cross-modality complementary information, enabling robust detection under diverse illumination and weather conditions. However, under extreme conditions where one modality exhibits poor quality and disturbs detection, modality separation is necessary to mitigate the impact of noise. To address this problem, we propose a Modality-Decoupled RGB-T detection framework with Query Fusion (MDQF) to balance modality complementation and separation. In this framework, DETR-like detectors are employed as separate branches for the RGB and TIR images, with query fusion interspersed between the two branches in each refinement stage. Herein, query fusion is performed by feeding the high-quality queries from one branch to the other one after query selection and adaptation. This design effectively excludes the degraded modality and corrects the predictions using high-quality queries. Moreover, the decoupled framework allows us to optimize each individual branch with unpaired RGB or TIR images, eliminating the need for paired RGB-T data. Extensive experiments demonstrate that our approach delivers superior performance to existing RGB-T detectors and achieves better modality independence.

Temporal information is crucial for visual tracking, but existing multi-frame trackers are vulnerable to model drift caused by naively aggregating noisy historical predictions. In this paper, we introduce DTPTrack, a lightweight and generalizable module designed to be seamlessly integrated into existing trackers to suppress drift. Our framework consists of two core components: (1) a Temporal Reliability Calibrator (TRC) mechanism that learns to assign a per-frame reliability score to historical states, filtering out noise while anchoring on the ground-truth template; and (2) a Temporal Guidance Synthesizer (TGS) module that synthesizes this calibrated history into a compact set of dynamic temporal priors to provide predictive guidance. To demonstrate its versatility, we integrate DTPTrack into three diverse tracking architectures—OSTrack, ODTrack, and LoRAT—and show consistent, significant performance gains across all baselines. Our best-performing model, built upon an extended LoRATv2 backbone, sets a new state-of-the-art on several benchmarks, achieving a 77.5\% Success rate on LaSOT and an 80.3\% AO on GOT-10k.

The vision-language tracking task aims to perform object tracking based on various modality references. Existing Transformer-based vision-language tracking methods have made remarkable progress by leveraging the global modeling ability of self-attention. However, current approaches still face challenges in effectively exploiting the temporal information and dynamically updating reference features during tracking. Recently, the State Space Model (SSM), known as Mamba, has shown astonishing ability in efficient long-sequence modeling. Particularly, its state space evolving process demonstrates promising capabilities in memorizing multimodal temporal information with linear complexity. Witnessing its success, we propose a Mamba-based vision-language tracking model to exploit its state space evolving ability in temporal space for robust multimodal tracking, dubbed MambaVLT. In particular, our approach mainly integrates a time-evolving hybrid state space block and a selective locality enhancement block, to capture contextual information for multimodal modeling and adaptive reference feature update. Besides, we introduce a modality-selection module that dynamically adjusts the weighting between visual and language references, mitigating potential ambiguities from either reference type. Extensive experimental results show that our method performs favorably against state-of-the-art trackers across diverse benchmarks.