Although text-to-image diffusion models exhibit remarkable generative power, concept erasure techniques are essential for preventing harmful content. Existing adversarial probes evaluate these methods by testing whether erased concepts can still be recovered. However, existing erasure and probe methods remain largely text-centric, focusing on whether the text-to-image mapping is severed while overlooking whether the corresponding visual knowledge remains. To investigate this question from a visual perspective, we leverage diffusion inversion to probe whether a generative trajectory can reconstruct visual instances of an erased concept. Under a null-text condition, standard inversion avoids the textual pathway but amplifies approximation errors, hindering faithful trajectory recovery. To address this challenge, we introduce TINA+, a diffusion-consistent Text-free INversion Attack equipped with optimization-based inversion. We also find that unconstrained diffusion inversion may discover spurious trajectories, even allowing a randomly initialized diffusion model to reconstruct the target concept. Such trajectories may falsely indicate residual visual knowledge. TINA+ therefore introduces Diffusion-Consistent Trajectory Regularization to suppress this failure mode. By penalizing trajectories that fall far below the expected marginal energy evolution of diffusion, TINA+ suppresses spurious inversion paths while preserving its ability to recover erased concepts. Experiments across twelve erasure methods, four concept-erasure tasks, and different model architectures demonstrate that TINA+ reliably probes residual visual knowledge through diffusion-consistent visual trajectories. These results provide stronger evidence that current methods often obscure concepts by severing text-image links rather than eliminating the underlying visual knowledge.

In this report, we present our champion solution for the NTIRE 2026 Challenge on Video Saliency Prediction held in conjunction with CVPR 2026. To exploit complementary inductive biases for video saliency, we propose Video Saliency with Adaptive Gated Experts (ViSAGE), a multi-expert ensemble framework. Each specialized decoder performs adaptive gating and modulation to refine spatio-temporal features. The complementary predictions from different experts are then fused at inference. ViSAGE thereby aggregates diverse inductive biases to capture complex spatio-temporal saliency cues in videos. On the Private Test set, ViSAGE ranked first on two out of four evaluation metrics, and outperformed most competing solutions on the other two metrics, demonstrating its effectiveness and generalization ability. Our code has been released at https://github.com/iLearn-Lab/CVPRW26-ViSAGE.

Although text-to-image diffusion models exhibit remarkable generative power, concept erasure techniques are essential for their safe deployment to prevent the creation of harmful content.This has fostered a dynamic interplay between the development of erasure defenses and the adversarial probes designed to bypass them, and this co-evolution has progressively enhanced the efficacy of erasure methods.However, this adversarial co-evolution has converged on a narrow, text-centric paradigm that equates erasure with severing the text-to-image mapping, ignoring that the underlying visual knowledge related to undesired concepts still persist.To substantiate this claim, we investigate from a visual perspective, leveraging DDIM inversion to probe whether a generative pathway for the erased concept can still be found.However, identifying such a visual generative pathway is challenging because standard text-guided DDIM inversion is actively resisted by text-centric defenses within the erased model.To address this, we introduce TINA, a novel Text-free INversion Attack, which enforces this visual-only probe by operating under a null-text condition, thereby avoiding existing text-centric defenses.Moreover, TINA integrates an optimization procedure to overcome the accumulating approximation errors that arise when standard inversion operates without its usual textual guidance.Our experiments demonstrate that TINA successfully regenerates erased concepts from models treated with state-of-the-art unlearning.The success of TINA proves that current methods merely obscure concepts, highlighting an urgent need for paradigms that operate directly on internal visual knowledge.Code will be released upon acceptance.

Video Moment Retrieval (VMR) aims to retrieve temporal moments in videos that align with natural language queries, a task requiring cross-modal reasoning between video and text. Among various supervision paradigms, point-supervised VMR has emerged as a practical solution, leveraging single-frame annotations to significantly reduce annotation costs while maintaining competitive retrieval performance. However, this sparse supervision approach induces cross-modal representation shift. This shift complicates the model's ability to accurately capture action sequences and associate text with visual content. To tackle this, we propose a novel framework called pseuDo fRame-based tempOral and semaNtic rEfinement (DRONE) with two key modules: (1) Pseudo-Frame Temporal Alignment (PTA), which embeds textual queries as pseudo-frames to enhance temporal coherence, and (2) Curriculum-Guided Semantic Refinement (CSR), which uses a progressive contrastive learning strategy to refine semantic representations from easy to hard cases. Extensive experiments show that DRONE achieves effective retrieval performance while keeping annotation costs low.

Image-text retrieval (ITR) is a pivotal task in cross-modal research. However, existing methods often suffer from a fundamental yet overlooked challenge: redundancy. This issue manifests as both semantic redundancy within unimodal representations and relationship redundancy in cross-modal alignments. This not only inflates computational costs but also degrades retrieval accuracy by masking salient features and reinforcing spurious correlations. In this work, we are the first to explicitly analyze and address the ITR problem from a redundancy perspective by proposing the iMage-text rEtrieval rEdundancy miTigation (MEET) framework. MEET employs a cascaded, two-stage process to systematically mitigate both forms of redundancy. First, for Semantic Redundancy Mitigation, it repurposes deep hashing and quantization as synergistic tools, producing compact yet highly discriminative representations. Second, for Relationship Redundancy Mitigation, it progressively refines the cross-modal alignment space by filtering misleading negative samples and adaptively reweighting informative pairs. The structural integration of these modules under a unified optimization objective provides a clear and interpretable pathway to retrieval. Extensive experiments on multiple benchmarks demonstrate that MEET consistently surpasses state-of-the-art methods, validating its effectiveness and generalizability.

Spatial intelligence is essential for low-altitude unmanned aerial vehicle (UAV) perception, collaboration, and navigation. However, existing UAV benchmarks often emphasize image-level recognition, single-view understanding, or narrow answer formats, leaving 3D spatial inference, multi-view collaboration, scene dynamics, and diverse task formulations insufficiently evaluated. To address these gaps, we introduce SpatialUAV, a real low-altitude UAV benchmark comprising 4,331 curated instances across 14 fine-grained task types, covering semantic discrimination, spatial relation, aerial–aerial collaboration, aerial–ground collaboration, and motion understanding. SpatialUAV organizes all samples into a unified visual-input–question–answer schema, while supporting seven input configurations and nine answer formats, including option labels, region identifiers, geometric values, cross-view correspondences, and free-form motion descriptions. To ensure reliable and grounded evaluation, our data construction pipeline integrates detector-assisted regions, depth supervision, metadata-derived rules, extensive manual annotation, blind filtering, and multi-turn human validation, together with task-specific metrics for heterogeneous outputs. Evaluating representative vision-language models across three categories, we show that current models remain far from human-level performance, with pronounced bottlenecks in cross-view association, structured grounding, geometric reasoning, and temporal viewpoint understanding. These results offer empirical guidance for advancing low-altitude UAV spatial intelligence. Code and data are available at https://github.com/Hyu-Zhang/SpatialUAV.

With the explosive growth of images, efficiently retrieving such data has become a critical research challenge. Online hashing has received increasing attention since its advantages in high computational efficiency and low storage burden in the realm of streaming data retrieval. Despite the evolution of online hashing techniques, core issues are still at large. Most existing methods fail to account for the difficulty level of the samples and categories involved in the learning process, which leaves the hash learning vulnerable to outliers and noise. To tackle this issue, we introduce a novel Self-PAced weighting Online Hashing (SPAOH) method. Specifically, drawing inspiration from cognitive learning theories, we introduce an online self-paced learning method that mimics the human learning process by progressively learning samples and categories from streaming data, starting with the easier ones and gradually moving to the more complex ones. Besides, we introduce an information-preserving strategy into the hash function learning to obtain potent and efficient hash projection. In addition, we put forward a new online optimization algorithm for solving the objective functions precisely and efficiently. Extensive experiments on four benchmark datasets demonstrate the superiority of SPAOH. The source code is available at https://github.com/ZCyueternal/SPAOH.git.

In this report, we present our champion solution for the DataMFM Challenge Track 2: Chart Understanding. This track requires models to recover structured chart data and generate faithful natural-language summaries from chart images. To address the complementary requirements of accurate data extraction and factual narration, we propose ChartLens, a dual-branch framework for chart data correction and summary refinement. ChartLens consists of two key modules: Structure-Aware CSV Verification and Correction (SAVC) and Text-Retention-Guided Summary Refinement (TRSR). SAVC improves the reliability of structured data extraction through verification and correction, while TRSR enhances summary generation by preserving critical textual and numerical evidence from charts. By combining model adaptation, correction-based generation, and OCR-assisted evidence grounding, ChartLens improves both structured data recovery and summary factuality. On the test set, our final system achieves an overall score of 69.10 and ranks first in Track 2, demonstrating its effectiveness for accurate chart understanding. Our code will be released at: https://github.com/iLearn-Lab/CVPRW26-ChartLens.

Glance-supervised video moment localization, which requires only single-frame annotations within target moments, offers a promising balance between annotation efficiency and localization precision. However, existing methods suffer from a fundamental limitation: the inherent uncertainty regarding precise moment boundaries when given only glance supervision. Current approaches either assume deterministic relationships between glance annotations and boundaries or employ predefined soft-label distributions, both failing to capture the complexity of real-world boundary uncertainty. We introduce UDist, an uncertainty-aware distillation network that explicitly models boundary prediction uncertainty as learnable probability distributions. Our framework consists of two key components: 1) an Uncertainty-aware Offset Distribution (UOD) mechanism, transforming implicit boundary uncertainty into explicit probability distributions over discretized offset bins; and 2) a dual-branch distillation framework that transfers uncertainty knowledge from a teacher to a student branch through a competitive distillation process. This “certainty to uncertainty to refined certainty” paradigm converts glance annotation certainty into uncertainty representations, then distills them into precise boundary predictions. Experiments on Charades-STA, ActivityNet Captions and TACoS show UDist improves localization accuracy while maintaining computational efficiency.

In this report, we present our third-place solution for the DataMFM Challenge Track 1: Document Parsing. This track requires models to recover structured Markdown documents from document page images while preserving textual content and document structure. To address the complementary requirements of accurate content recovery and faithful structure reconstruction, we propose ParseFixer, an agentic framework for backbone parsing and selective correction. ParseFixer consists of two key modules: Full-Page Backbone Parsing (FBP) and Agentic Selective Correction (ASC). FBP produces stable initial Markdown outputs with MinerU2.5 Pro, while ASC detects high-value parsing failures and repairs them through a verify-and-rollback correction process. By placing selective multimodal correction after open-source backbone parsing, ParseFixer improves the recovery of key document elements without rewriting reliable backbone predictions. On the test set, our final system achieves an overall score of 61.78 and ranks third in Track 1, demonstrating its effectiveness for accurate document parsing. Our code will be released at: https://github.com/iLearn-Lab/CVPRW26-ParseFixer.

Masked Image Modeling (MIM), following "mask-andreconstruct" scheme, is a promising self-supervised method to learn scalable visual representation. Studies indicate that selecting an effective mask strategy is vital for MIM. However, existing approaches often rely on static pre-defined priors, which limit their ability to adapt mask strategies dynamically for network optimization. In this paper, we focus on the learning process of the network and introduce human-like curriculum into MIM for dynamic representation refinement, and propose an end-to-end framework Curriculum Masked Image Modeling (CurMIM). CurMIM consists of two components: Mask Priority Measurer, which acts as a curriculum learner to determine mask priority values using the network's intrinsic state information, and Dual Adaptive Selector, which serves as a curriculum scheduler to create effective masks based on these values. With negligible extra parameters, our curriculum-based method consistently establishes noticeable improvements across varying model sizes and benchmarks, showing effectiveness and generalization.

Video moment retrieval aims to determine the temporal boundaries of moments within a video that are most relevant to textual queries. Unlike fully-supervised and weakly-supervised methods, frame-supervised methods use a single annotated frame to model the similarities between the target moment and queries. This task is still in its infancy due to the following challenges: 1) indiscernible intra-modal information and 2) inflexible inter-modal information interaction. In light of these challenges, we introduce the Gaming fOr elAstic Localization (GOAL) method for frame-supervised video moment retrieval. It enables target moment boundary localization from a novel strategic game perspective. GOAL encompasses two core components: a game-based paradigm to find the most reliable moment and a Dynamic Updating Technique (DUT) to continuously optimize moment retrieval through dynamic gradients, thereby refining boundary predictions with different feedback. Extensive experiments on Charades-STA, ActivityNet Captions, and TACoS have validated the effectiveness of GOAL.

Localizing a desired moment within an untrimmed video via a given natural language query, i.e., cross-modal moment localization, has attracted widespread research attention recently. However, it is a challenging task because it requires not only accurately understanding intra-modal semantic information, but also explicitly capturing inter-modal semantic correlations (consistency and complementarity). Existing efforts mainly focus on intra-modal semantic understanding and inter-modal semantic alignment, while ignoring necessary semantic supplement. Consequently, we present a cross-modal semantic perception network for more effective intra-modal semantic understanding and inter-modal semantic collaboration. Concretely, we design a dual-path representation network for intra-modal semantic modeling. Meanwhile, we develop a semantic collaborative network to achieve multi-granularity semantic alignment and hierarchical semantic supplement. Thereby, effective moment localization can be achieved based on sufficient semantic collaborative learning. Extensive comparison experiments demonstrate the promising performance of our model compared with existing state-of-the-art competitors.

Video moment localization (VML) aims to identify the temporal boundary semantically matching the given query. Point-supervised VML balances localization accuracy and annotation cost but is still immature due to granularity alignment and scale perception issues. To this end, we propose a Semantic Granularity and Scale Correspondence Integration (SG-SCI) framework aimed at leveraging limited single-frame annotation for correspondence learning. It explicitly models semantic relations of different feature granularities and adaptively mines the implicit semantic scale, thereby enhancing feature representations of varying granularities and scales. SG-SCI uses granularity correspondence alignment to align semantics via latent prior knowledge and a scale correspondence learning to identify and address semantic scale differences. Extensive experiments on benchmark datasets have demonstrated the promising performance of our model over several state-of-the-art competitors.

Informative sensing facilitates the effectiveness and efficiency for environmental monitoring. This paper investigates a deep learning approach that can estimate the optimal results of informative sensing in a random field. First, a Gaussian process (GP) is designated to characterize an environmental field. Then, mutual information over the covariance function of GP is utilized to define the near-optimal locations as the most informative sampling places in the field. At last, a deep neural network is developed to learn the intercorrelation between the covariance function and the the most informative sampling places, which can provide an near online planning of the MI optimization process. In this paper, the experimental results on a real-world dataset validate the proposed learning framework.

Based on the human–robot interaction behavior of mobile robots in social navigation, this paper proposes a social navigation quality evaluation model based on combined weights for the problems of single indicators, rough quantification and non-convergence of information in social navigation quality evaluation. Firstly, three evaluation modules of comfort, naturalness and sociality are designed, and each module is refined into primary and secondary indicators. The robot path navigation data are calculated by the indicator quantification formula. Secondly, the subjective and objective weights of hierarchical analysis method and the entropy weight method are combined to determine the index weights at each level. The weighted sum is used to achieve the fusion of index information and obtain the optimal solution of the evaluation navigation algorithm. Finally, we simulate the social scene through visualization simulation experiments to obtain the trajectory data of the robot in the social scene. The experimental results verify the feasibility of the theoretical model and give the final scores and optimization opinions of the tested algorithms. Through the evaluation of the social navigation quality evaluation model, the path planning algorithm that best suits the comfort perception of pedestrians in the current scenario can be found in the tested algorithms.

Aiming at the problem of difficult quantization of robot comfort value in dynamic environments, we proposed a robot behavior evaluation network based on transformer and graphical attention mechanism. Firstly, we defined the characteristics of environment nodes, robot nodes and pedestrian nodes. Secondly, pedestrians and robots are taken as nodes of the relationship graph. Information of each node is aggregated using graphical neural network and self-attention mechanism, so as to obtain the characteristics of the whole scene, which gives a comfort score for pedestrians. Finally, the parameters of the evaluation model were trained using the manually labeled public dataset SocNav2, and the results of the test dataset show that our designed evaluation network is closer to the real human feelings. Meanwhile, we built different scenarios in gazebo environment to verify the value of this robot behavior evaluation network in applications. The experimental results showed that our designed evaluation network can objectively represent the impact of robot behavior on the comfort of surrounding pedestrians, which is instructive for designing robot behavior.

Regarding the problem of environmental field reconstruction, a method is proposed to establish the optimal observation position of the environmental field using a random process and complete the online planning of the optimal observation position through deep learning technology. First, this chapter uses a random process method to construct an environmental field model and describes the environmental object as a Gaussian Markov process. For the variable spatial field random process model, the information gain evaluation mechanism is used to define the sampling point with the most information in the field as the approximate optimal observation position, and the global optimal observation point is generated. Subsequently, this chapter proposes a high-precision fitting representation and real-time calculation framework based on a deep attention network, constructs a deep neural network with complex nonlinear representation learning ability, learns the relationship between the covariance function corresponding to the Gaussian random field and the optimal information observation point, simulates the global optimal observation position generated by the random process environment model, solves the real-time calculation of the global optimal position, and realizes the online representation and planning of the spatial field observation position.

Recently, temporal action localization (TAL) has garnered significant interest in information retrieval community. However, existing supervised/weakly supervised methods are heavily dependent on extensive labeled temporal boundaries and action categories, which is labor-intensive and time-consuming. Although some unsupervised methods have utilized the “iteratively clustering and localization” paradigm for TAL, they still suffer from two pivotal impediments: 1) unsatisfactory video clustering confidence, and 2) unreliable video pseudolabels for model training. To address these limitations, we present a novel self-paced iterative learning model to enhance clustering and localization training simultaneously, thereby facilitating more effective unsupervised TAL. Concretely, we improve the clustering confidence through exploring the contextual feature-robust visual information. Thereafter, we design two (constant- and variable- speed) incremental instance learning strategies for easy-to-hard model training, thus ensuring the reliability of these video pseudolabels and further improving overall localization performance. Extensive experiments on two public datasets have substantiated the superiority of our model over several state-of-the-art competitors.

With the proliferation of IoMT (Internet of Medical Things), billions of connected medical devices are constantly producing oceans of time series sensor data, dubbed as time series for short. Considering these time series reflect various functional states of the human body, how to effectively detect the corresponding abnormalities is of great significance for smart healthcare. Accordingly, we develop a horizontal visibility graph-based temporal classification model for disease diagnosis. We conduct extensive comparison experiments on the benchmark datasets to justify the superiority of our method in term of accuracy and efficiency. Besides, we have released the codes and parameters to facilitate the community research. We propose an identifiable temporal feature selection via horizontal visibility graph for time series classification (TSC) based disease diagnosis. We conduct comparison experiments on the benchmark datasets to justify the superiority of our method in term of accuracy and efficiency. As a side contribution, we have released the codes and parameters to facilitate the community research ( https://github.com/sdujicun/SSVG ).