Searcharxiv⌕ Search

arXiv subjects

Deyu Meng

Publications and source records attributed to Deyu Meng.

At least 19 recordsLinked to original sources
arXiv2026

Rotation Equivariant Mamba for Vision Tasks

Rotation equivariance is an important structural prior for a broad range of vision tasks, particularly in domains such as remote sensing and medical imaging where objects have no canonical orientation, yet it remains notably absent from current Mamba-based vision architectures. Despite the success of Mamba in natural language processing and its growing adoption in computer vision, existing visual Mamba models fail to account for rotational symmetry in their design. This omission renders them sensitive to image rotations, thereby constraining their robustness and cross-task generalization. To address this limitation, we incorporate rotation symmetry, an important geometric prior in many visual domains, into Mamba-based architectures. Specifically, we introduce EQ-VMamba, the first rotation equivariant visual Mamba architecture for vision tasks. The core components of EQ-VMamba include a carefully designed rotation equivariant cross-scan strategy and group Mamba blocks. Moreover, we provide a rigorous theoretical analysis of the intrinsic equivariance error, demonstrating that the proposed architecture enforces end-to-end rotation equivariance under the p4 rotation group. Extensive experiments across multiple benchmarks---including high-level image classification, mid-level semantic segmentation, and low-level image super-resolution---demonstrate that EQ-VMamba consistently improves rotation robustness and achieves superior or competitive performance compared to non-equivariant baselines on most benchmarks, while requiring 40-70% fewer parameters. These results indicate that embedding rotation equivariance not only effectively bolsters the robustness of visual Mamba models against rotation transformations, but also improves accuracy on data exhibiting rotational symmetry, with substantially improved parameter efficiency. Code is available at https://github.com/zhongchenzhao/EQ-VMamba.

cs.CV↗
arXiv2026

HotPOT: Operator Transformer with Heterogeneous Physics Experts for Large-Scale PDE Pre-training

Large-scale PDE pre-training trains a single model on the data from multiple partial differential equations (PDEs). However, adding heterogeneous PDEs with different governing equations, physical variables, and variable interactions does not always improve predictive performance across tasks. Existing PDE mixture-of-experts models mitigate cross-task interference through learned routing, but typically use homogeneous expert architectures, making it difficult to directly reflect differences in state evolution across equations. We introduce HotPOT, an operator transformer with heterogeneous physics experts for large-scale PDE pre-training. HotPOT uses a shared encoder to capture information common across PDEs, while each PDE family uses a neural-numerical expert that pairs a learnable predictor with a fixed numerical update containing no trainable parameters. Each predictor outputs the quantities required by its paired numerical update, which then computes the next state. This design allows different PDEs to share a common encoder while retaining their own learnable predictors and numerical updates. Experiments across five PDE families and six settings show that HotPOT outperforms MoE-POT in all 24 matched-scale comparisons before and after fine-tuning, reducing rollout error by 50.7% on average. Notably, the 17M model also surpasses substantially larger MoE-POT models, and the performance gains extend to zero-shot prediction under unseen physical conditions. Code is available at https://github.com/cruiseresearchgroup/HotPOT-code.

cs.LG↗
arXiv2026

Diagnosing and Mitigating Evidence Misgrounding in Multimodal Knowledge Editing

Multimodal knowledge editing aims to update visually grounded facts in Multimodal Large Language Models (MLLMs) while preserving unrelated text and visual behaviors. However, standard answer-level metrics can hide evidence misgrounding: an edit may produce the target answer on the original image yet remain sensitive to incidental context rather than the intended visual evidence. To examine this issue in a controlled setting, we construct WaterBird-Edit, an analysis set that decouples bird habitat labels from background context and measures whether edited predictions remain stable when the target evidence is preserved but the background changes. Motivated by this diagnosis, we propose VICE (Visual In-Context Editing), a retrieval-based framework that stores edits in external memory and applies them through visual in-context demonstrations designed to expose task-relevant evidence. Experiments on three standard benchmarks show that VICE achieves strong lifelong editing performance, while results on WaterBird-Edit show that it substantially reduces sensitivity to the controlled background shifts. Code and dataset are available in https://github.com/Wang-Z-W/VICE-main.

cs.CV↗
arXiv2026

Registration-Grounded Spectral Fusion for Unregistered WLI/NBI Endoscopic Lesion Segmentation

White-light imaging (WLI) and narrow-band imaging (NBI) provide complementary cues for endoscopic lesion segmentation, yet their sequential acquisition often results in spatially inconsistent observations. The central challenge is not merely to establish cross-modal correspondence, but to determine how complementary information should be incorporated when such correspondence is uncertain. To address this issue, we propose a correspondence-conditioned, modality-asymmetric complex spectral fusion framework for paired-but-unregistered WLI/NBI segmentation. The framework first establishes bidirectional feature correspondence through topology-regularized registration and derives local reliability from post-warping feature discrepancies. Rather than simply reweighting fused features, the estimated reliability directly controls the construction of modality-asymmetric polar representations, where WLI-derived cues primarily parameterize appearance-related moduli and NBI-derived cues parameterize structure-sensitive angular components. The resulting complex representations are further refined through Fourier-domain recalibration with reliability-biased residual reconstruction. Experiments on FAHSYSU and PICCOLO demonstrate competitive improvements over existing segmentation methods. Module ablations, modality-role reversal, and reliability analyses further support the effectiveness of correspondence-conditioned polar representation learning.

cs.CV↗
arXiv2026

FisherEdit: Fisher Information Guided Knowledge Editing for Large Language Models

Sequential knowledge editing in Large Language Models (LLMs) often leads to a degradation of general capabilities due to cumulative parameter interference. In this paper, we revisit prominent projection-based methods like AlphaEdit and reveal a surprising mismatch: their robustness is primarily driven by update regularization rather than the intended null-space projection. Based on this insight, we propose FisherEdit, a theoretically grounded method built on Fisher geometry, which replaces coarse activation-space projections with importance-aware regularization in parameter space. It penalizes changes to sensitive parameters more strongly while retaining flexibility for new facts, and admits an efficient row-wise closed-form solution. Experiments on Llama-3, Qwen2.5, and Olmo-3 across multiple benchmarks demonstrate that FisherEdit achieves state-of-the-art performance while better preserving model stability and downstream task accuracy.

cs.CL↗
arXiv2026

Exact Endpoint Asymptotics and Almost-Sure Limits for Uniform Empirical Measures in Two Dimensions

For the empirical measure of the first $n$ points of one infinite sequence of independent uniform samples in the unit square, we identify the leading constant in the bounded-Lipschitz distance to area measure, equivalently in $W_1$, both in expectation and almost surely along all sample sizes. The constant is $c_*=κ_1/(2\sqrtπ)\in(0,1/4]$, where $κ_1$ is the linear-growth Gaussian variational constant of Armegioiu--Goldman--Grotto--Trevisan. This answers the two-dimensional bounded-Lipschitz convergence question posed by Yukich in 1989. We give the complete endpoint argument, square--torus comparison, empirical reduction, and full-sequence probability argument.

math.PR↗
arXiv2026

Where to Adapt Matters: Layer-Selective Fine-Tuning for Capability Retention

Parameter-efficient fine-tuning (PEFT) enables large language models (LLMs) to adapt to specialized tasks, but often at the cost of degrading general capabilities acquired during pretraining. Existing approaches primarily mitigate this trade-off through data replay or regularization, relying on additional data or explicit optimization constraints. We instead focus on a different question: where should adaptation be applied? We find that fine-tuning different Transformer layers produces different target-task gains and degrees of capability degradation, suggesting that not all layers are equally suitable for adaptation. To characterize this difference, we use layer-wise empirical Fisher information to measure target-task sensitivity. However, computing Fisher scores requires backward computation and becomes increasingly expensive for large models. We therefore introduce input--output cosine similarity as a lightweight, forward-only proxy for ranking layer sensitivity. Across models and tasks, layers with lower input--output similarity consistently exhibit higher empirical Fisher scores. Building on this observation, we propose Layer-Selective LoRA (LS-LoRA), which places trainable LoRA adapters only in layers with low input--output similarity. Experiments on mathematical reasoning and code generation show that LS-LoRA improves average target-task performance while retaining substantially more commonsense reasoning capability than standard all-layer LoRA, demonstrating that carefully choosing where to adapt can provide a simple and effective way to balance target-task adaptation and general capability retention.

cs.AI↗
arXiv2026

HyperSAM: A Promptable Foundation Model for Hyperspectral Remote Sensing

Hyperspectral remote sensing provides dense spectral measurements that are indispensable for material-level Earth observation, yet the construction of a general-purpose hyperspectral foundation model remains difficult. Two bottlenecks are especially limiting. First, large hyperspectral corpora rarely provide high spatial resolution together with reliable dense annotations. Second, many hyperspectral models are still trained almost from scratch, so the geometric and interactive priors learned by modern vision foundation models are not fully reused. To alleviate these issues, we \highlight{present} \textbf{HyperSAM}, a promptable hyperspectral foundation model that couples a data-centric hyperspectral synthesis pipeline with a spectral adaptation architecture based on Segment Anything Model 3 (SAM3). On the data side, HyperSAM synthesizes full-spectrum hyperspectral cubes from high-resolution SpaceNet multispectral imagery through a physics-informed abundance-transfer generator, while SAM3-derived pseudo-masks provide object-centric supervision. On the model side, the latest implementation uses a frozen SAM3 RGB image branch, a trainable hyperspectral side encoder initialized from the RGB vision transformer (ViT), ControlNet-style zero-initialized feature injection, and a lightweight mixture-of-experts mask refiner. To enhance training robustness against noisy pseudo-labels, Cross-modal Sample Selection (CromSS)-style confidence selection is incorporated for noisy-label weighting. Extensive experiments show that HyperSAM obtains strong generalization on diverse hyperspectral tasks (e.g., classification, anomaly detection, change detection, target detection, and airborne oil-spill mapping) and that high-quality synthetic hyperspectral data can be more effective than simply scaling noisy hyperspectral supervision.

cs.CV↗
arXiv2026

Flash EQ-Linear: Accelerating Equivariant Linear Layers via Group-wise Discrete Fourier Transform

Equivariant networks embed geometric symmetries as structural priors through weight sharing, achieving remarkable parameter efficiency across vision tasks. However, this parameter efficiency does not translate into compute efficiency: most existing implementations unroll the structured weights into dense matrices and dispatch them to generic dense kernels, so an equivariant layer costs no fewer MACs than its non-equivariant counterpart. In this paper, we observe that the equivariant linear (EQ-Linear) layer---the most fundamental and frequently used module in modern equivariant architectures---is essentially a circular convolution along the group dimension composed with a linear transform along the channel dimension. Building on this observation, we propose Flash EQ-Linear, an exact acceleration algorithm that reduces the cost to $2(T-1)/T^2$ of the original dense formulation ($T$ is the equivariant group size) by combining the Fourier convolution theorem along the group dimension with the conjugate symmetry of the real DFT. To translate these computational savings into wall-clock speedups, we further develop dedicated CUDA kernels for the $\mathrm{p}4$ group. At the operator level, Flash EQ-Linear achieves up to $2.1\times$ forward speedup over PyTorch's highly optimized F.linear; at the network level, Flash EQ-ViT achieves up to ${1.7\times}$ end-to-end speedup over both equivariant and non-equivariant baselines. As an operator-level acceleration algorithm, Flash EQ-Linear provides plug-and-play acceleration for diverse pretrained equivariant models, including EQ-ViT, EQ-Swin, EQ-VMamba, and EQ-INR, without retraining or architectural changes. Code is available at https://github.com/zhongchenzhao/FlashEQLinear.

cs.CV↗
arXiv2026

Multi-Stage NeRF for Efficient 3D Coronary Artery Reconstruction from Two Narrow-Angle Angiographic Projections

X-ray coronary angiography is the clinical gold standard for coronary artery disease during real-time cardiac interventions, but provides only 2D projections of inherently 3D vessels. Existing learning-based 2D-to-3D reconstruction methods typically require wide angular coverage or multiple views, assumptions that are rarely satisfied in routine practice where only two projections with narrow angular separation are available. To address these challenges, we propose NeCA++, a multi-stage self-supervised neural radiance field (NeRF) framework tailored to clinically realistic acquisition constraints. The framework decomposes reconstruction into two stages that progressively refine spatial support and representation capacity. In the first stage, a coarse 3D representation of the vasculature is reconstructed, restricting the subsequent optimisation to regions with a higher likelihood of vessel presence, termed an active region. Afterward reconstruction is restricted to this region while higher-resolution representations are progressively activated to recover fine vascular details. This multi-stage strategy focuses learning on anatomically plausible regions, mitigates gradient dilution under extreme sparsity, and stabilises global topology before recovering fine vascular branches. Furthermore, two vessel-specific regularisations are introduced: a ray-aligned constraint to reduce projection-induced ambiguity, and a bimodal density penalty to enable early vessel-background separation. Extensive experiments across three datasets (ImageCAS, ASOCA, and Synthetic RCA) and four angular configurations demonstrate consistent superiority over state-of-the-art baselines, particularly under clinically realistic narrow-angle settings, while achieving reconstruction within 58 seconds per case.

cs.CG↗
arXiv2026

Tensor Decomposition Structure Search Framework from an Interaction Perspective

Recently, tensor decompositions have attracted increasing attention. Fundamentally, different interactions among factors induce distinct tensor decomposition structures (i.e., tensor decomposition). Identifying an appropriate interaction-induced tensor decomposition structure for given data is a fundamental yet challenging problem in tensor modeling, and remains largely under-explored. Existing tensor decomposition structure search methods are typically restricted to a predefined interaction family, such as tensor contraction. To address this problem, we suggest a tensor decomposition structure search framework (I-TSS) from an interaction perspective that can identify either a single structure beyond a predefined interaction family or a mixture of structures involving heterogeneous interaction families. Specifically, we first systematically review existing tensor decomposition structures from an interaction perspective and construct a heterogeneous interaction-induced candidate structure set. Based on it, we propose a unified energy-based rank estimation scheme for heterogeneous interaction-induced candidate structure set. We then introduce the top-$k$ gating mechanism with learnable gating scores to dynamically select a suitable single-candidate structure or combine suitable multi-candidate structures, thereby identifying the tensor decomposition structure in a data-adaptive manner. Theoretically, we derive an approximation error bound for I-TSS, thereby establishing its approximation capability. Extensive experiments on synthetic and real-world datasets demonstrate that I-TSS consistently outperforms state-of-the-art tensor decomposition methods.

cs.CV↗
arXiv2026

Pre-Trained Low-Rank Tensor Decomposition for Multi-Dimensional Image Recovery

Recently, tensor decompositions are prevalent for multi-dimensional image representation, which learn the instance-specific structure of each image from scratch. However, tensor decompositions neglect the common structure across different images, leading to limited semantic modeling capability, high computational cost, and a large number of learnable parameters. To address this challenge, we suggest the first pre-trained low-rank tensor decomposition (PLTD) framework, which organically integrates the pre-trained large vision model into the classical tensor decomposition framework. Beyond the shallow and untrained deep tensor decomposition, the suggested PLTD achieves an unprecedented balance among higher recovery fidelity, fewer learnable parameters, and smaller carbon footprint. Specifically, PLTD factorizes the target tensor into a latent tensor and a learnable transform that maps the latent tensor back to the original data domain. The latent tensor consists of two indispensable and complementary terms, i.e., a fixed pre-trained latent tensor and a learnable low-rank latent tensor. The fixed pre-trained latent tensor is distilled from a pre-trained large vision model (i.e., DINOv3) to capture the common structure of the target tensor, while the learnable low-rank latent tensor characterizes the instance-specific structure of the target tensor. To examine the potential of PLTD, we develop the corresponding multi-dimensional image recovery model and theoretically justify the advantages of this framework. Additionally, we discuss the connections between PLTD and classical tensor decomposition frameworks. Extensive experiments on multi-dimensional image recovery demonstrate that PLTD consistently achieves superior performance compared with state-of-the-art methods.

cs.CV↗
arXiv2026

ReUnit: Multi-Granularity Visual Unitization for Long Video Understanding

Long-video understanding is constrained by the limited visual input capacity of video multimodal large language models (Video-MLLMs). Existing methods mainly optimize which content to retain, while the presentation of retained content often remains fixed. As a result, the same balance between spatial detail and content coverage is imposed across the entire visual input. We propose ReUnit, a training-free and query-aware framework that jointly determines which content to retain and how it should be presented. Guided by frame-level query relevance, ReUnit constructs and allocates visual units at multiple presentation granularities. More relevant content receives finer presentation, while broader context is represented more compactly. It realizes these granularities using visual units that carry one, four, or nine source frames and are each rendered as a standard image. Across four benchmarks and visual input budgets from 4 to 32 units, ReUnit achieves the highest average score at every tested budget. It improves over Uniform Sampling by 8.3--10.1 points on average, and the gains transfer across three additional Video-MLLM families. Project resources are available at https://github.com/charon525/ReUnit.

cs.CV↗
arXiv2026

DOD-SA: Infrared-Visible Decoupled Object Detection with Single-Modality Annotations

Infrared-visible object detection has shown great potential in real-world applications, enabling robust all-day perception by leveraging the complementary information of infrared and visible images. However, existing methods typically require dual-modality annotations to output decoupled detection results, leading to high annotation costs and limiting scalability in large-scale remote sensing applications. To address this challenge, we propose a novel infrared-visible \textbf{D}ecoupled \textbf{O}bject \textbf{D}etection framework with \textbf{S}ingle-modality \textbf{A}nnotations, called DOD-SA. It is built upon a Single- and Dual-Modality Collaborative Teacher-Student Network (CoSD-TSNet), which consists of a single-modality branch (SM-Branch) and a dual-modality decoupled branch (DMD-Branch). This design enables cross-modality knowledge transfer from the labeled modality to the unlabeled modality, and facilitates effective cross-branch supervision. To further improve the quality of pseudo-labels, we introduce a Progressive and Self-Tuning Training Strategy (PaST) that trains the model in three stages: 1) SM-Branch self-training, 2) SM-Branch guiding the learning of DMD-Branch, and 3) DMD-Branch refinement. In addition, we design a Pseudo Label Assigner (PLA) to match labels across modalities, explicitly addressing modality misalignment during training.

cs.CV↗
arXiv2026

Making Every Step Count: Spatio-Temporal Information Allocation for Imaging Inverse Problems

Flow-based generative models have emerged as powerful image priors for training-free inverse problem solving, capturing coherent semantics and fine-grained structure. Despite these strengths, existing flow-based inverse solvers primarily focus on the design of individual updates, largely overlooking spatio-temporal information allocation under a fixed number of function evaluations (NFEs). Temporally, insufficient early exploration can trap the flow trajectory in an incorrect semantic basin, whereas excessive allocation of NFEs to early stages leaves little budget for late-stage refinement. Spatially, data consistency provides direct constraints only within observed regions, whereas the recovery of missing regions relies mainly on the generative prior. To address these two issues, we introduce two complementary and training-free components, i.e., Spectrum-Adaptive Scheduling (SAS) and Measurement-Prioritized Attention (MPA). For temporal allocation, SAS distributes the available NFEs over flow time according to the degradation spectrum and logSNR geometry, thus better balancing semantic exploration and detail refinement. For spatial propagation, MPA exploits data-prior conflicts to guide information toward weakly constrained regions, thereby enhancing semantic and structural fidelity. Extensive experiments on standard image inverse problems, e.g., super-resolution, motion deblurring, and inpainting, demonstrate that the proposed components can be integrated into existing flow-based inverse solvers in a plug-and-play manner without retraining or additional flow-model evaluations, and can also significantly improve the restoration quality of existing solvers.

cs.CV↗
arXiv2026

Sequential Monte Carlo with Gaussian Mixture Approximation for Infinite-Dimensional Statistical Inverse Problems

By formulating the inverse problem of partial differential equations (PDEs) as a statistical inference problem, the Bayesian approach provides a general framework for quantifying uncertainties. In the inverse problem of PDEs, parameters are defined on an infinite-dimensional function space, and the PDEs induce a computationally intensive likelihood function. Additionally, sparse data tends to lead to a multi-modal posterior. These features make it difficult to apply existing sequential Monte Carlo (SMC) algorithms. To overcome these difficulties, we propose new conditions for the likelihood functions, construct a Gaussian mixture based preconditioned Crank-Nicolson transition kernel, and demonstrate the universal approximation property of the infinite-dimensional Gaussian mixture probability measure. By combining these three novel tools, we propose a new SMC algorithm with Gaussian mixture approximation, together with an easy-to-use reduced version. For this new algorithm, we obtain a convergence theorem that allows Gaussian priors, illustrating that the sequential particle filter actually reproduces the true posterior distribution. Furthermore, the proposed new algorithm is rigorously defined on the infinite-dimensional function space, naturally exhibiting the discretization-invariant property. Numerical experiments demonstrate that the reduced version has a strong ability to probe the multi-modality of the posterior, significantly reduces the computational burden, and numerically exhibits the discretization-invariant property (important for large-scale problems).

math.NA↗
arXiv2026

SC-Diff: Semantically Calibrated Diffusion for Visible-to-Infrared Image Translation

Visible-to-infrared image translation provides a practical way to expand infrared training data using abundant visible images. Diffusion models are promising for this task because of their strong generative performance. However, existing diffusion-based methods typically use semantic priors only as external conditions, without explicitly regulating token interactions within the denoising network. Consequently, they struggle to preserve object locations, shapes, and semantic layouts required for reliable annotation reuse. We propose SC-Diff, a semantically calibrated latent diffusion framework that uses semantic priors for both conditional guidance and internal self-attention calibration. A pretrained SAM3 model with predefined text prompts first extracts category-specific semantic masks from visible images. These masks are merged into a semantic map and fused with the visible image as the input condition. The same map is converted into token-level semantic labels to calibrate self-attention in the denoising network. Based on these labels, we introduce Semantic-Guided Self-Attention Calibration (SGSC), which adaptively applies positive biases to query-key pairs of the same category. The query-wise calibration strength depends on the dispersion of attention across semantic categories and the attention assigned to the query's own category. The original attention scores further modulate the bias, giving greater calibration to same-category keys with stronger responses. This soft calibration reduces cross-category interference while retaining global contextual interactions, thereby improving semantic consistency in generated infrared images. Extensive experiments show that SC-Diff improves perceptual quality and produces more effective synthetic training data for downstream infrared object detection.

cs.CV↗
arXiv2026

Harnessing the Potential of Optimizing Data Mixtures via Bayesian Domain Reweighting

The performance of Large Language Models (LLMs) is fundamentally influenced by the distributional composition of multi-domain pre-training data. While manual heuristics were prevalent in early models, they increasingly fail to capture the intricate synergies between domains as data complexity grows. To overcome the issue, a dominant approach seeks to fit a proxy function mapping between domain weights and their corresponding validation losses, and then find the optimal domain weights to minimize validation losses. These methods rely on strong structural assumptions, such as rank invariance or scaling laws, which are often violated, resulting in non-negligible estimation bias. A promising approach is to directly optimize the weighting scheme from data. However, it suffers from unstable optimization trajectory and prohibitive computational overhead, limiting its potential to search better domain weights configurations. This paper presents a Bayesian domain weighting method to infer the weights from a Dirichlet distribution via introducing Gamma prior information learned from observations. Experimental results demonstrate that proposed method could achieve stable and efficient domain weights learning, and identifies optimal mixtures while consuming substantially less data than search-based function-fitting methods, revitalizing optimization-based domain weighting for large-scale applications.

cs.LG↗