arXiv subjects

Di Wang

Publications and source records attributed to Di Wang.

At least 181 records · Page 10Linked to original sources

arXiv2025

Towards Analysing Invoices and Receipts with Amazon Textract

This paper presents an evaluation of the AWS Textract in the context of extracting data from receipts. We analyse Textract functionalities using a dataset that includes receipts of varied formats and conditions. Our analysis provided a qualitative view of Textract strengths and limitations. While the receipts totals were consistently detected, we also observed typical issues and irregularities that were often influenced by image quality and layout. Based on the analysis of the observations, we propose mitigation strategies.

cs.IR↗

arXiv2025

A Dual-Branch Local-Global Framework for Cross-Resolution Land Cover Mapping

Cross-resolution land cover mapping aims to produce high-resolution semantic predictions from coarse or low-resolution supervision, yet the severe resolution mismatch makes effective learning highly challenging. Existing weakly supervised approaches often struggle to align fine-grained spatial structures with coarse labels, leading to noisy supervision and degraded mapping accuracy. To tackle this problem, we propose DDTM, a dual-branch weakly supervised framework that explicitly decouples local semantic refinement from global contextual reasoning. Specifically, DDTM introduces a diffusion-based branch to progressively refine fine-scale local semantics under coarse supervision, while a transformer-based branch enforces long-range contextual consistency across large spatial extents. In addition, we design a pseudo-label confidence evaluation module to mitigate noise induced by cross-resolution inconsistencies and to selectively exploit reliable supervisory signals. Extensive experiments demonstrate that DDTM establishes a new state-of-the-art on the Chesapeake Bay benchmark, achieving 66.52\% mIoU and substantially outperforming prior weakly supervised methods. The code is available at https://github.com/gpgpgp123/DDTM.

cs.CV↗

arXiv2025

Scale-Invariant Robust Estimation of High-Dimensional Kronecker-Structured Matrices

High-dimensional Kronecker-structured estimation faces a conflict between non-convex scaling ambiguities and statistical robustness. The arbitrary factor scaling distorts gradient magnitudes, rendering standard fixed-threshold robust methods ineffective. We resolve this via Scaled Robust Gradient Descent (SRGD), which stabilizes optimization by de-scaling gradients before truncation. To further enforce interpretability, we introduce Scaled Hard Thresholding (SHT) for invariant variable selection. A two-step estimation procedure, built upon robust initialization and SRGD--SHT iterative updates, is proposed for canonical matrix problems, such as trace regression, matrix GLMs, and bilinear models. The convergence rates are established for heavy-tailed predictors and noise, identifying a phase transition where optimal convergence rates recover under finite noise variance and degrade optimally for heavier tails. Experiments on simulated data and two real-world applications confirm superior robustness and efficiency of the proposed procedure.

stat.ME↗

arXiv2025

Anatomical Region-Guided Contrastive Decoding: A Plug-and-Play Strategy for Mitigating Hallucinations in Medical VLMs

Medical Vision-Language Models (MedVLMs) show immense promise in clinical applicability. However, their reliability is hindered by hallucinations, where models often fail to derive answers from visual evidence, instead relying on learned textual priors. Existing mitigation strategies for MedVLMs have distinct limitations: training-based methods rely on costly expert annotations, limiting scalability, while training-free interventions like contrastive decoding, though data-efficient, apply a global, untargeted correction whose effects in complex real-world clinical settings can be unreliable. To address these challenges, we introduce Anatomical Region-Guided Contrastive Decoding (ARCD), a plug-and-play strategy that mitigates hallucinations by providing targeted, region-specific guidance. Our module leverages an anatomical mask to direct a three-tiered contrastive decoding process. By dynamically re-weighting at the token, attention, and logits levels, it verifiably steers the model's focus onto specified regions, reinforcing anatomical understanding and suppressing factually incorrect outputs. Extensive experiments across diverse datasets, including chest X-ray, CT, brain MRI, and ocular ultrasound, demonstrate our method's effectiveness in improving regional understanding, reducing hallucinations, and enhancing overall diagnostic accuracy.

cs.CV↗

arXiv2025

CheXPO-v2: Preference Optimization for Chest X-ray VLMs with Knowledge Graph Consistency

Medical Vision-Language Models (VLMs) are prone to hallucinations, compromising clinical reliability. While reinforcement learning methods like Group Relative Policy Optimization (GRPO) offer a low-cost alignment solution, their reliance on sparse, outcome-based rewards inadvertently encourages models to "overthink" -- generating verbose, convoluted, and unverifiable Chain-of-Thought reasoning to justify answers. This focus on outcomes obscures factual errors and poses significant safety risks. To address this, we propose CheXPO-v2, a novel alignment framework that shifts from outcome to process supervision. Our core innovation is a Knowledge Graph Consistency Reward mechanism driven by Entity-Relation Matching. By explicitly parsing reasoning steps into structured "Disease, Relation, Anatomy" triplets, we provide fine-grained supervision that penalizes incoherent logic and hallucinations at the atomic level. Integrating this with a hard-example mining strategy, our approach significantly outperforms GRPO and state-of-the-art models on benchmarks like MIMIC-CXR-VQA. Crucially, CheXPO-v2 achieves new state-of-the-art accuracy using only 5k samples, demonstrating exceptional data efficiency while producing clinically sound and verifiable reasoning. The project source code is publicly available at: https://github.com/ecoxial2007/CheX-Phi4MM.

cs.CV↗

arXiv2025

Advancing Weakly-Supervised Change Detection in Satellite Images via Adversarial Class Prompting

Weakly-Supervised Change Detection (WSCD) aims to distinguish specific object changes (e.g., objects appearing or disappearing) from background variations (e.g., environmental changes due to light, weather, or seasonal shifts) in paired satellite images, relying only on paired image (i.e., image-level) classification labels. This technique significantly reduces the need for dense annotations required in fully-supervised change detection. However, as image-level supervision only indicates whether objects have changed in a scene, WSCD methods often misclassify background variations as object changes, especially in complex remote-sensing scenarios. In this work, we propose an Adversarial Class Prompting (AdvCP) method to address this co-occurring noise problem, including two phases: a) Adversarial Prompt Mining: After each training iteration, we introduce adversarial prompting perturbations, using incorrect one-hot image-level labels to activate erroneous feature mappings. This process reveals co-occurring adversarial samples under weak supervision, namely background variation features that are likely to be misclassified as object changes. b) Adversarial Sample Rectification: We integrate these adversarially prompt-activated pixel samples into training by constructing an online global prototype. This prototype is built from an exponentially weighted moving average of the current batch and all historical training data. Our AdvCP can be seamlessly integrated into current WSCD methods without adding additional inference cost. Experiments on ConvNet, Transformer, and Segment Anything Model (SAM)-based baselines demonstrate significant performance enhancements. Furthermore, we demonstrate the generalizability of AdvCP to other multi-class weakly-supervised dense prediction scenarios. Code is available at https://github.com/zhenghuizhao/AdvCP

cs.CV↗

arXiv2025

SAMCL: Empowering SAM to Continually Learn from Dynamic Domains with Extreme Storage Efficiency

Segment Anything Model (SAM) struggles in open-world scenarios with diverse domains. In such settings, naive fine-tuning with a well-designed learning module is inadequate and often causes catastrophic forgetting issue when learning incrementally. To address this issue, we propose a novel continual learning (CL) method for SAM, termed SAMCL. Rather than relying on a fixed learning module, our method decomposes incremental knowledge into separate modules and trains a selector to choose the appropriate one during inference. However, this intuitive design introduces two key challenges: ensuring effective module learning and selection, and managing storage as tasks accumulate. To tackle these, we introduce two components: AugModule and Module Selector. AugModule reduces the storage of the popular LoRA learning module by sharing parameters across layers while maintaining accuracy. It also employs heatmaps-generated from point prompts-to further enhance domain adaptation with minimal additional cost. Module Selector leverages the observation that SAM's embeddings can effectively distinguish domains, enabling high selection accuracy by training on low-consumed embeddings instead of raw images. Experiments show that SAMCL outperforms state-of-the-art methods, achieving only 0.19% forgetting and at least 2.5% gain on unseen domains. Each AugModule requires just 0.233 MB, reducing storage by at least 24.3% over other fine-tuning approaches. The buffer storage for Module Selector is further reduced by up to 256$\times$.

cs.CV↗

arXiv2025

S5: Scalable Semi-Supervised Semantic Segmentation in Remote Sensing

Semi-supervised semantic segmentation (S4) has advanced remote sensing (RS) analysis by leveraging unlabeled data through pseudo-labeling and consistency learning. However, existing S4 studies often rely on small-scale datasets and models, limiting their practical applicability. To address this, we propose S5, the first scalable framework for semi-supervised semantic segmentation in RS, which unlocks the potential of vast unlabeled Earth observation data typically underutilized due to costly pixel-level annotations. Built upon existing large-scale RS datasets, S5 introduces a data selection strategy that integrates entropy-based filtering and diversity expansion, resulting in the RS4P-1M dataset. Using this dataset, we systematically scale up S4 into a new pretraining paradigm, S4 pre-training (S4P), to pretrain RS foundation models (RSFMs) of varying sizes on this extensive corpus, significantly boosting their performance on land cover segmentation and object detection tasks. Furthermore, during fine-tuning, we incorporate a Mixture-of-Experts (MoE)-based multi-dataset fine-tuning approach, which enables efficient adaptation to multiple RS benchmarks with fewer parameters. This approach improves the generalization and versatility of RSFMs across diverse RS benchmarks. The resulting RSFMs achieve state-of-the-art performance across all benchmarks, underscoring the viability of scaling semi-supervised learning for RS applications. All datasets, code, and models will be released at https://github.com/MiliLab/S5

cs.CV↗

arXiv2025

Hubble constant measurement from QPEs as electromagnetic counterparts to extreme mass ratio inspirals

Gravitational waves (GWs) accompanied by electromagnetic (EM) counterparts provide a novel methodology to measure the Hubble constant ($H_0$), known as bright sirens. However, the rarity of such multi-messenger events limits the precision of the $H_0$ constraint. Recently, the newly-discovered nuclear transient, quasi-periodic eruptions (QPEs) show intriguing evidence of a stellar-mass companion captured by a supermassive black hole (SMBH) in an extreme/intermediate mass-ratio inspiral (EMRI/IMRI), which is the most promising sources of the space-based GW detectors, such as LISA. Here, we model the secular orbital evolution of known QPE systems using two frameworks: a stripping scenario in which periodic mass transfer at periapsis drives the evolution; and an orbiter-disk collision scenario in which the companion interacts with a misaligned accretion disk, modulated by coupled orbiter-disk precession. For each framework, we assess detectability by LISA, together with the resulting constraints on $H_0$. Our principal findings are: (i) in the stripping scenario, no currently known QPE reaches detectability within a four-year LISA mission. (ii) in the orbiter-disk scenario, two sources-eRO-QPE2 and eRO-QPE4-are detectable with signal-to-noise ratios $\simeq 8.5-28.8$ and constrain $H_0$ with fractional uncertainty of 6.7-14.9\%. QPE systems remain uncertain on the decade-long secular evolution. Therefore, they motivate continued time-domain monitoring of QPE candidates.

astro-ph.HE↗

arXiv2025

Towards Reasoning-Preserving Unlearning in Multimodal Large Language Models

Machine unlearning aims to erase requested data from trained models without full retraining. For Reasoning Multimodal Large Language Models (RMLLMs), this is uniquely challenging: intermediate chain-of-thought steps can still leak sensitive information even when final answers are forgotten, and overly aggressive interventions easily damage general reasoning ability. Yet no benchmark jointly evaluates how well unlearning methods suppress reasoning-level leakage while preserving reasoning competence. We address this gap with RMLLMU-Bench, the first benchmark for RMLLM unlearning that extends standard forgetting metrics with dedicated measures of reasoning leakage and reasoning retention. A systematic evaluation on RMLLMU-Bench reveals that existing unlearning methods for MLLMs and Large (Language) Reasoning Models (LRMs) either leave substantial leakage in the reasoning process or severely degrade reasoning performance. To address these gaps, we propose R-MUSE (Reasoning-preserving MLLM Unlearning via Subspace guidance and Adaptive Steering), a training-free and inference-time intervention framework that steers internal representations to forget both answers and reasoning traces while explicitly preserving general reasoning. Experiments on RMLLMU-Bench demonstrate that R-MUSE achieves a substantially better balance between effective forgetting and reasoning retention.

cs.CL↗

arXiv2025

Performance Calibration of the Wavefront Sensor's EMCCD Detector for the Cool Planets Imaging Coronagraph Aboard CSST

The wavefront sensor (WFS), equipped with an electron-multiplying charge-coupled device (EMCCD) detector, is a critical component of the Cool Planets Imaging Coronagraph (CPI-C) on the Chinese Space Station Telescope (CSST). Precise calibration of the WFS's EMCCD detector is essential to meet the stringent requirements for high-contrast exoplanet imaging. This study comprehensively characterizes key performance parameters of the detector to ensure its suitability for astronomical observations. Through a multi-stage screening protocol, we identified an EMCCD chip exhibiting high resolution and low noise. The electron-multiplying gain (EM Gain) of the EMCCD was analyzed to determine its impact on signal amplification and noise characteristics, identifying the optimal operational range. Additionally, noise properties such as readout noise were investigated. Experimental results demonstrate that the optimized detector meets CPI-C's initial application requirements, achieving high resolution and low noise. This study provides theoretical and experimental foundations for the use of EMCCD-based WFS in adaptive optics and astronomical observations, ensuring their reliability for advanced space-based imaging applications

astro-ph.IM↗

arXiv2025

Understanding Private Learning From Feature Perspective

Differentially private Stochastic Gradient Descent (DP-SGD) has become integral to privacy-preserving machine learning, ensuring robust privacy guarantees in sensitive domains. Despite notable empirical advances leveraging features from non-private, pre-trained models to enhance DP-SGD training, a theoretical understanding of feature dynamics in private learning remains underexplored. This paper presents the first theoretical framework to analyze private training through a feature learning perspective. Building on the multi-patch data structure from prior work, our analysis distinguishes between label-dependent feature signals and label-independent noise, a critical aspect overlooked by existing analyses in the DP community. Employing a two-layer CNN with polynomial ReLU activation, we theoretically characterize both feature signal learning and data noise memorization in private training via noisy gradient descent. Our findings reveal that (1) Effective private signal learning requires a higher signal-to-noise ratio (SNR) compared to non-private training, and (2) When data noise memorization occurs in non-private learning, it will also occur in private learning, leading to poor generalization despite small training loss. Our findings highlight the challenges of private learning and prove the benefit of feature enhancement to improve SNR. Experiments on synthetic and real-world datasets also validate our theoretical findings.

cs.LG↗

arXiv2025

Topological diagrams of doubly charmed baryon decays in the $SU(3)_F$ limit

The doubly charmed baryon was first observed by LHCb via the non-leptonic decay $Ξ_{cc}^{++}\to Λ^+_cK^-π^+π^+$ in 2017. Subsequently, ongoing efforts have been made to identify other doubly charmed baryons. However, there is no systematic analysis of the topological decomposition for non-leptonic decays of doubly charmed baryons. In this work, we study the topological amplitudes of doubly charmed baryon decays in the $SU(3)_F$ limit. Tree- and penguin-induced topological diagrams for the $\mathcal{B}_{cc}\to \mathcal{B}_c M$ and $\mathcal{B}_{cc}\to \mathcal{B} D$ decays are presented. The linear relations between the topological amplitudes and the $SU(3)$ irreducible amplitudes are derived through tensor contraction and $SU(3)$ decomposition. The magnitude pattern of the topological diagrams is analyzed in the rescattering dynamics and the large $N_c$ expansion. In addition, some amplitude relations are derived to test the Körner-Pati-Woo theorem in the isospin limit.

hep-ph↗

arXiv2025

PIXEL: Adaptive Steering Via Position-wise Injection with eXact Estimated Levels under Subspace Calibration

Reliable behavior control is central to deploying large language models (LLMs) on the web. Activation steering offers a tuning-free route to align attributes (e.g., truthfulness) that ensure trustworthy generation. Prevailing approaches rely on coarse heuristics and lack a principled account of where to steer and how strongly to intervene. To this end, we propose Position-wise Injection with eXact Estimated Levels (PIXEL), a position-wise activation steering framework that, in contrast to prior work, learns a property-aligned subspace from dual views (tail-averaged and end-token) and selects intervention strength via a constrained geometric objective with a closed-form solution, thereby adapting to token-level sensitivity without global hyperparameter tuning. PIXEL further performs sample-level orthogonal residual calibration to refine the global attribute direction and employs a lightweight position-scanning routine to identify receptive injection sites. We additionally provide representation-level guarantees for the minimal-intervention rule, supporting reliable alignment. Across diverse models and evaluation paradigms, PIXEL consistently improves attribute alignment while preserving model general capabilities, offering a practical and principled method for LLMs' controllable generation. Our code is available at https://github.com/V1centNevwake/PIXEL-Adaptive-Steering

cs.AI↗

arXiv2025

Backdooring CLIP through Concept Confusion

Backdoor attacks pose a serious threat to deep learning models by allowing adversaries to implant hidden behaviors that remain dormant on clean inputs but are maliciously triggered at inference. Existing backdoor attack methods typically rely on explicit triggers such as image patches or pixel perturbations, which makes them easier to detect and limits their applicability in complex settings. To address this limitation, we take a different perspective by analyzing backdoor attacks through the lens of concept-level reasoning, drawing on insights from interpretable AI. We show that traditional attacks can be viewed as implicitly manipulating the concepts activated within a model's latent space. This motivates a natural question: can backdoors be built by directly manipulating concepts? To answer this, we propose the Concept Confusion Attack (CCA), a novel framework that designates human-understandable concepts as internal triggers, eliminating the need for explicit input modifications. By relabeling images that strongly exhibit a chosen concept and fine-tuning on this mixed dataset, CCA teaches the model to associate the concept itself with the attacker's target label. Consequently, the presence of the concept alone is sufficient to activate the backdoor, making the attack stealthier and more resistant to existing defenses. Using CLIP as a case study, we show that CCA achieves high attack success rates while preserving clean-task accuracy and evading state-of-the-art defenses.

cs.CR↗

arXiv2025

Robust Drone-View Geo-Localization via Content-Viewpoint Disentanglement

Drone-view geo-localization (DVGL) aims to match images of the same geographic location captured from drone and satellite perspectives. Despite recent advances, DVGL remains challenging due to significant appearance changes and spatial distortions caused by viewpoint variations. Existing methods typically assume that drone and satellite images can be directly aligned in a shared feature space via contrastive learning. Nonetheless, this assumption overlooks the inherent conflicts induced by viewpoint discrepancies, resulting in extracted features containing inconsistent information that hinders precise localization. In this study, we take a manifold learning perspective and model $\textit{the feature space of cross-view images as a composite manifold jointly governed by content and viewpoint}$. Building upon this insight, we propose $\textbf{CVD}$, a new DVGL framework that explicitly disentangles $\textit{content}$ and $\textit{viewpoint}$ factors. To promote effective disentanglement, we introduce two constraints: $\textit{(i)}$ an intra-view independence constraint that encourages statistical independence between the two factors by minimizing their mutual information; and $\textit{(ii)}$ an inter-view reconstruction constraint that reconstructs each view by cross-combining $\textit{content}$ and $\textit{viewpoint}$ from paired images, ensuring factor-specific semantics are preserved. As a plug-and-play module, CVD integrates seamlessly into existing DVGL pipelines and reduces inference latency. Extensive experiments on University-1652 and SUES-200 show that CVD exhibits strong robustness and generalization across various scenarios, viewpoints and altitudes, with further evaluations on CVUSA and CVACT confirming consistent improvements.

cs.CV↗

arXiv2025

Tracing and Mitigating Hallucinations in Multimodal LLMs via Dynamic Attention Localization

Multimodal Large Language Models (MLLMs) achieve strong performance on tasks like image captioning and visual question answering, but remain prone to hallucinations, where generated text conflicts with the visual input. Prior work links this partly to insufficient visual attention, but existing attention-based detectors and mitigation typically apply uniform adjustments across layers and heads, obscuring where errors originate. In this paper, we first show these methods fail to accurately localize problematic layers. Then, we introduce two diagnostics: Layer Image Attention Entropy (LIAE) which flags anomalous layers, and Image Attention Focus (IAF) which scores attention heads within those layers. Analysis shows that LIAE pinpoints faulty layers and IAF reliably ranks heads that warrant correction. Guided by these signals, we propose Dynamic Layer-wise Entropy and Attention Fusion (D-LEAF), a task-agnostic, attention-guided method that dynamically localizes and corrects errors during inference with negligible overhead. Furthermore, by establishing a connection between D-LEAF and DPO, we provide theoretical justification for the effectiveness of D-LEAF. Results show our D-LEAF delivers a 53\% relative improvement on standard captioning benchmarks, and on VQA both accuracy and F1-score improve by approximately 4\%, substantially suppressing hallucinations while preserving efficiency.

cs.CV↗