Searcharxiv⌕ Search

arXiv subjects

Andreas Dengel

Publications and source records attributed to Andreas Dengel.

At least 19 recordsLinked to original sources
arXiv2026

Biological Congruence: Evaluating Microscopy Generation Across Representations

Generative models increasingly produce realistic microscopy images, yet conventional evaluation primarily characterizes visual fidelity and distributional coverage rather than the biological information represented by synthetic data. We introduce biological congruence as an evaluation principle describing whether biologically relevant information forms a coherent description across representations and remains consistent with its empirical biological context. To study this principle, we develop a text-conditioned diffusion framework that jointly generates microscopy images and instance masks, exposing the same generated biological state through conditioning, visual appearance, and instance structure. Across prompt--mask, prompt--image, image--mask, and population-level analyses, we find strongly feature-dependent behavior: density and confluency are consistently preserved, morphology exhibits recurrent disagreement, while nearest-neighbor distance reveals that strong image--mask agreement can coexist with weaker conditioning and population fidelity. Stress testing increasingly unsupported phenotype combinations further shows that target controllability can increase while untargeted phenotypes reorganize and the complete generated state moves toward configurations better supported by real data. These findings demonstrate that generative quality, controllability, cross-representation agreement, and population fidelity capture complementary properties, motivating biological congruence as a relational perspective for evaluating scientific generative models.

cs.CV↗
arXiv2026

DocXEval: A Systematic Evaluation of Explainability in Multimodal Document Understanding

Multimodal transformer models have driven unprecedented performance breakthroughs in modern document understanding. However, the inherent black-box nature of these models remains a major obstacle to their safe deployment in high-stakes sectors. Despite significant progress in explainability for vision and language tasks, explainable AI (XAI) research tailored to multimodal document understanding has received little attention. To address this gap, we present \ourframework{}, the first comprehensive explainability framework that adapts existing XAI methods and evaluation metrics to the multimodal (text, layout, and image) document understanding context, while introducing efficient multi-target adaptations of both methods and metrics for domain-specific tasks such as semantic entity recognition. Using \ourframework{}, we systematically evaluate 13 XAI methods across 3 document understanding tasks, 8 datasets, 4 models, and 17 evaluation metrics, and find that only a few methods perform consistently across all evaluation criteria, with Occlusion emerging as the most faithful and interpretable method overall. Through a dedicated meta-evaluation, we observe moderate to strong agreement among several faithfulness and complexity metrics while identifying certain ones that produce unreliable results. Finally, through multiple dataset-wide diagnostic analyses, we uncover key insights into model behavior, including the contribution of individual modalities across tasks and labels, spatial attribution patterns around target entities, and potential dataset biases. Overall, this work serves as both a benchmark and a practical guide for future XAI research in multimodal document understanding. To support reproducibility, we publicly release our code at https://github.com/saifullah3396/docxeval/.

cs.CL↗
arXiv2026

LUMA: Benchmarking Segmentation via a Lightweight Universal Mask Adapter

Comparing transformer backbones for image segmentation is confounded: each is paired with a different decoder, recipe, and pretraining, so reported differences rarely reflect the backbone itself. We introduce the Lightweight Universal Mask Adapter (LUMA), a lightweight, backbone-agnostic mask-transformer head that treats any backbone as a black-box feature extractor, letting a set of queries read from its features through cheap cross-attention. LUMA matches the accuracy of EoMT, the state-of-the-art efficient ViT-segmenter, at cost parity, while attaching unchanged to hierarchical, convolutional, and mixture-of-experts backbones alike. Holding this head fixed, we benchmark 20 backbones, 11 pretraining schemes and a range of resolutions on ADE20K, COCO, and Cityscapes under one modern recipe. We find that ``efficient'' token mixers fail to deliver efficiency even at the high resolutions that motivate them, with plain ViT holding the throughput Pareto front at every resolution. Additionally, the pretraining objective, not the architecture, the lever the field has tuned hardest, governs segmentation quality. Code: https://github.com/tobna/LUMA

cs.CV↗
arXiv2026

Rethinking Microscopy Generation: Co-Designed Diffusion for Biologically Interpretable Single-Cell Synthesis

Generative models for microscopy can already produce realistic whole-image fields of view, but offer no principled way to control how each individual cell looks. We address this gap by recasting microscopy generation at the single-cell level and designing the input controls and the model's internal representation jointly, so that both are organized along the same biologically meaningful directions. On the input side, we aggregate a long list of standard cell measurements into a small set of semantic axes (size, elongation, brightness, texture, etc.) that are each named, measurable, and traversable. On the model side, we introduce BioVAE, an encoder-decoder whose latent space is explicitly aligned with the same set of measurements. A key finding is that the benefit of biologically structuring the latent emerges only when conditioning is also moved to the semantic axes, showing that the two ends must be designed jointly. Experiments on LIVECell show that semantic-axis conditioning substantially improves generative fidelity over high-dimensional conditioning, a blind study with biologists confirms the realism of generated cells, and per-axis traversals produce stable and interpretable phenotypic changes.

cs.CV↗
arXiv2026

PASTA: Prompt-Aware Annealed Diffusion Sampling

While Diffusion models produce high-quality images, real-world visual design requires intentionally preserving clean regions for headlines, logos, or UI elements. Existing layout-aware text-to-image methods struggle to strictly preserve reserved regions, often introducing unwanted content that complicates downstream editing. Existing approaches either manipulate attention to displace content or suppress conditioning via noise-level interventions. However, these methods sacrifice visual coherence, producing flat regions or noticeable style inconsistencies between reserved and non-reserved areas. To address this limitation, we propose a Prompt-Aware Shared-Latent Training-Free Annealed Diffusion Sampling (PASTA), a training-free framework that enforces reserved-region constraints by selectively applying semantic guidance to object-allowed regions during sampling. PASTA operates on a shared latent trajectory, where full-prompt and context-prompt guidance are fused through mask-guided sampling. An annealed schedule suppresses semantic leakage during early denoising stages while gradually restoring visual consistency in later stages. Extensive experiments demonstrate that PASTA improves reserved-region fidelity while preserving overall image realism and perceptual consistency.

cs.CV↗
arXiv2026

CARES: Coverage-Aware Replay Exemplar Selection for Continual Learning

Replay-based Continual Learning mitigates catastrophic forgetting by keeping a buffer of past samples, making buffer composition critical under a fixed memory budget. Recent coreset-based replay methods improve buffer quality using model-derived gradients, losses, or auxiliary optimization, but incur substantial computational overhead as memory is repeatedly updated. We propose CARES, a continual model-optimization-free approach for buffer construction that instead exploits the geometry of a frozen pretrained feature space using class-aware, density-weighted coverage. We evaluate CARES across balanced, imbalanced, and label-noisy streams under online class- and task-incremental settings. On Tiny-ImageNet, CARES is 2.3 to 8.8 times faster than existing coreset-based selectors. Although CARES achieves this speedup using no training signal from the continual learner, it ranks among the state of the art across balanced, imbalanced, and label-noisy streams under online class- and task-incremental settings, outperforming the strongest baseline by up to 7.58 percentage points under class imbalance while also attaining higher average accuracy on Tiny-ImageNet. Code will be released upon publication

cs.LG↗
arXiv2026

Where Did the Light Come From? Diagnosing Cast-Shadow Reasoning in VLMs and Borrowing the Missing Shadow Prior from Image Generation Models

Shadows are observable consequences of latent physical causes, arising from the interaction between illumination, object geometry, and scene structure. A model that understands shadows should be able not only to read a shadow's direction, but also to infer where a shadow should fall when it is absent, and to judge whether a shadow is physically consistent with the scene. We introduce ShadowBench, built from real photographs, which poses these three questions to vision-language models (VLMs). From each scene, we build a correct-shadow, a shadow-removed, and a human-annotated wrong-direction state. Evaluating 19 open VLMs (2B to 78B) and frontier proprietary models, we find that VLMs read a visible shadow's direction well above chance and improve with scale, yet infer the absent direction only weakly and fall to chance at judging physical consistency even under a bias-free two-alternative test that removes the yes-bias of single-image naturalness questions. Only Gemini-3.1-pro exceeds chance on validity, and even then only modestly and far below humans. Chain-of-Thought (CoT) does not help. The missing ingredient is not reasoning but knowledge that image generation models (IGMs) already hold. Routing a VLM through an IGM lifts its inference well above its direct attempt and helps it flag the wrong-direction shadows it otherwise accepts. Cast-shadow direction is a single facet of illumination understanding, but the sharp gap it reveals between VLMs and IGMs suggests image-generation or reconstruction objectives as a route toward illumination-aware VLMs.

cs.CV↗
arXiv2026

Nesting or resolution? Facility proxies for variance estimation versus confounding adjustment in claims data without facility identifiers

In claims databases, facility identifiers are often removed during de-identification, leaving only coarse attributes such as bed size. Yet facility information is needed for variance estimation, confounding adjustment and assessment of treatment availability. We examined whether accreditation-gated billing items can build strata standing for hospitals, and how to construct a facility proxy for each use. In 1,730,736 adults undergoing surgery (Medical Data Vision database, Japan, 2019-2023), we built marker-pattern strata from 66 accreditation-gated receipt codes (markers) and evaluated each grouping against a true facility identifier in an insurer-based database (DeSC Healthcare) and in simulation. The variance explained by the marker-pattern strata decreased with the extent to which a variable was patient- rather than hospital-determined ($R^2$ 0.921 for an accreditation item, 0.036 for sex). For variance estimation, the supplied-attribute cells, which can merge but never split hospitals, kept the type I error rate at 4.8% with a wild cluster bootstrap, against 5.4% for the true identifier and 39% when clustering was ignored; every finer marker-based grouping did worse. For adjustment, evaluated with real facility-level confounders and simulated null exposures, the marker pattern removed 51% of the confounding related to accreditation-linked facility characteristics, against 8% for the cells. The proportion of patients in groups mixing hospitals with and without the relevant accreditation was 49% for the marker pattern against 98% for the cells. No single facility proxy serves every purpose: nesting governs variance estimation, and resolution governs adjustment and treatment availability. Without a facility identifier, the supplied-attribute cells can carry the standard errors, and marker-pattern strata are worth considering for the other two uses.

stat.ME↗
arXiv2026

The Self-Improvement Problem

With the rise of large language models, agentic systems are becoming increasingly autonomous, as they write code, call tools, revise prompts, and propose modifications to their own components. This autonomy raises a fundamental question for safe self-improvement: can an automated gate like a reward model or an automated reviewer reliably decide whether a proposed update should be accepted? The difficulty is structural. The gate must be a terminating, computable procedure, while the property we ultimately care about, namely whether an update is a genuine semantic improvement, depends on potentially unbounded program behavior. We formalize this mismatch by modeling agents as programs and defining improvement in terms of a semantic score that need not be computable. We study two evaluation abstractions. The first is a certifier that approves or rejects an update. The second is a search procedure that, given access to a proposal generator, attempts to output some improving candidate if one exists. We show that, in the fully general setting, no terminating procedure can simultaneously guarantee reliability (never accepting a non-improvement) and non-conservatism (never missing an available improvement): self-referential constructions can force any such gate or searcher to fail. Randomized gates with bounded error face the same limit. We then identify a standard decidable regime used in practice (bounded evaluation and a finite proposal budget) under which exhaustive simulation yields a terminating search procedure with the desired guarantees. We also show the limit of this recovery: a chain of accepted updates can raise the bounded score in every round while the true score falls. Our results draw a boundary between unbounded, universal settings where correctness guarantees collapse and bounded settings where they can be recovered by design.

cs.AI↗
arXiv2026

iBitter-Stack: A Multi-Representation Ensemble Learning Model for Accurate Bitter Peptide Identification

The identification of bitter peptides is crucial in various domains, including food science, drug discovery, and biochemical research. These peptides not only contribute to the undesirable taste of hydrolyzed proteins but also play key roles in physiological and pharmacological processes. However, experimental methods for identifying bitter peptides are time-consuming and expensive. With the rapid expansion of peptide sequence databases in the post-genomic era, the demand for efficient computational approaches to distinguish bitter from non-bitter peptides has become increasingly significant. In this study, we propose a novel stacking-based ensemble learning framework aimed at enhancing the accuracy and reliability of bitter peptide classification. Our method integrates diverse sequence-based feature representations and leverages a broad set of machine learning classifiers. The first stacking layer comprises multiple base classifiers, each trained on distinct feature encoding schemes, while the second layer employs logistic regression to refine predictions using an eight-dimensional probability vector. Extensive evaluations on a carefully curated dataset demonstrate that our model significantly outperforms existing predictive methods, providing a robust and reliable computational tool for bitter peptide identification. Our approach achieves an accuracy of 96.09\% and a Matthews Correlation Coefficient (MCC) of 0.9220 on the independent test set, underscoring its effectiveness and generalizability. To facilitate real-time usage and broader accessibility, we have also developed a user-friendly web server based on the proposed method, which is freely accessible at https://ibitter-stack-webserver.streamlit.app/. This tool enables researchers and practitioners to conveniently screen peptide sequences for bitterness in real-time applications.

q-bio.QM↗
arXiv2026

Walking the Score Manifold: Continuous-time Generative Dynamics on Learned Data Manifolds

Generative modeling of time-dependent data is typically formulated on a discrete temporal grid, restricting supervision to the observed timestamps in the training data. We instead frame generation as continuous-time evolution on a learned data manifold. To this end, we leverage pretrained score-based models as geometric priors and learn a vector field that evolves data along score-induced interpolation paths. Because these dynamics follow transitions that respect the geometry learned by the score model, they support generation at arbitrary timestamps and temporal super-resolution beyond the discretization of the training data. Moreover, this geometric formulation allows us to train the vector field simulation-free through a regression objective. To improve long-horizon rollout robustness, we introduce an objective that promotes path-relative transverse exponential stability. While motivated by stability theory, it admits a practical interpretation as denoising score matching transverse to the interpolation path. Further, we extend the framework to a probabilistic setting that models a distribution over plausible future trajectories. We demonstrate the method on natural video and scientific dynamical data, including temporal super-resolution, PDE-based spatiotemporal fields, and molecular dynamics. Our results show that score-based priors provide a strong foundation for learning stochastic continuous-time generative dynamics.

cs.LG↗
arXiv2026

SLUM-i: Semi-supervised Learning for Urban Mapping of Informal Settlements and Data Quality Benchmarking

Very-high-resolution remote-sensing imagery provides a scalable basis for delineating informal settlements, but sparse annotations, severe imbalance between informal-settlement and background pixels, and cross-city heterogeneity in urban morphology and image--mask correspondence complicate model development. We present SLUM-i, a semi-supervised semantic segmentation framework together with a geographically diverse seven-city Earth observation benchmark spanning three continents. The benchmark combines a newly annotated Lahore dataset and companion Karachi and Mumbai datasets with four publicly released city datasets, totaling 14,458 RGB image--mask tiles. We quantify cross-city heterogeneity using class composition, boundary morphology, grayscale separability, correspondence between mask boundaries and image edges, and divergence between measured feature distributions. For label-efficient mapping, SLUM-i combines representation-guided unlabeled-pool curation, using embeddings from a vision foundation model (DINOv2-Small) to remove the least-similar tiles, and Class-Aware Adaptive Thresholding, which adapts pseudo-label acceptance by class through global mean-confidence and per-class mean-softmax exponential moving averages. Experiments at 10%, 20%, and 30% label budgets, using convolutional and vision-transformer backbones and five random seeds, demonstrate improvements over the corresponding UniMatch baselines in multiple city--budget settings, reaching +5.9 percentage points in mean intersection-over-union. At the 30% budget, the ResNet-101 configuration matches or exceeds its corresponding fully labeled supervised baseline in four of seven cities. Both components operate only during training and add no inference overhead.

cs.CV↗
arXiv2026

Layers Matter: Why Continual Learning Regularization Should Be Layer-Adaptive

Continual learning regularizers like EWC fight forgetting by penalizing changes from previous-task parameters with per-parameter importance, typically diagonal Fisher values. Per-parameter looks more flexible than per-layer, but each layer's diagonal Fisher is a weak summary of its actual curvature, missing the top-eigenvalue information that controls forgetting. Adversarial bit-flip attacks and Hessian-spectrum studies show that this missing per-layer sensitivity spans orders of magnitude in neural networks. Under a block-diagonal Hessian assumption, the layer-level analogue of EWC's existing diagonal assumption, we prove three things. Forgetting decomposes as a sum of per-layer terms weighted by each layer's top Hessian eigenvalue. Diagonal-Fisher weights cannot recover this eigenvalue. For instance, two layers with identical Fisher averages can have top eigenvalues differing by a factor as large as the layer width. For the same level of forgetting, uniform regularization loses new-task performance by an amount scaling with the layer condition number. Our theoretical analysis leads to a simple recipe: protect early layers strongly, let deeper layers move. We apply this recipe to EWC and SLCA and show clear improvements in average performance and forgetting metrics.

cs.LG↗
arXiv2026

When Pretty Isn't Useful: Investigating Why Modern Text-to-Image Models Fail as Reliable Training Data Generators

Recent text-to-image (T2I) diffusion models produce visually stunning images and demonstrate excellent prompt following. But do they perform well as synthetic vision data generators? In this work, we revisit the promise of synthetic data as a scalable substitute for real training sets and uncover a surprising performance regression. We generate large-scale synthetic datasets using state-of-the-art T2I models released between 2022 and 2025, train standard classifiers solely on this synthetic data, and evaluate them on real test data. Despite observable advances in visual fidelity and prompt adherence, classification accuracy on real test data consistently declines with newer T2I models as training data generators. Our analysis reveals a hidden trend: These models collapse to a narrow, aesthetic-centric distribution that undermines diversity and real data distribution coverage. Overall, our findings challenge a growing assumption in vision research, namely that progress in generative realism implies progress in data realism. We thus highlight an urgent need to rethink the capabilities of modern T2I models as reliable training data generators.

cs.CV↗
arXiv2026

CogRad: A Cognitively-Inspired Multi-Agent Framework for Radiology Report Generation

Automated radiology report generation (RRG) can ease radiologist workload, yet most existing systems produce a report in a single forward pass, with no mechanism to check a claim against the image or revisit a finding once stated. We present CogRad, a cognitively inspired multi-agent framework that structures generation around four stages of a radiologist's reading process. A Scout agent discovers anatomical regions directly from image patches via slot attention and assigns region and disease-level triage scores; an Investigator agent concentrates representational capacity on the regions Scout flags as suspicious; a Writer agent compiles these signals into a disease gated visual prefix for a large language model; and a Verifier agent supervises training with a visual entailment loss and, at inference, re-examines its own draft sentence by sentence, regenerating any report it judges insufficiently grounded. On CheXpert Plus, CogRad attains a BLEU-4 of 0.316 and a CIDEr of 0.322, the best scores among the methods we compare against. On IU X-Ray, it attains a BLEU-4 of 0.201 and a CIDEr of 0.724, leading every baseline on every standard NLG metric. We further evaluate CogRad with RadGraph F1, CheXbert F1, and a hallucination analysis to assess clinical accuracy beyond standard text-overlap metrics, complemented by ablation studies and Grad-CAM-based visualizations that characterize each agent's contribution and the model's visual grounding.

cs.CV↗
arXiv2026

ForAug: Mitigating Biases in Image Classification via Controlled Image Compositions

Large-scale image classification datasets exhibit strong compositional biases: objects tend to be centered, appear at characteristic scales, and co-occur with class-specific context. By exploiting such biases, models attain high in-distribution accuracy but remain fragile under distribution shifts. To address this issue, we introduce ForAug, a controlled composition augmentation scheme that factorizes each training image into a foreground object and a background and recombines them to explicitly manipulate object position, object scale, and background identity. ForAug uses off-the-shelf segmentation and inpainting models to (i) extract the foreground and synthesize a neutral background, and (ii) paste the foreground onto diverse neutral backgrounds before applying standard strong augmentation policies. Compared to conventional augmentations and content-mixing methods, our factorization provides direct control knobs that break foreground-background correlations. Across 10 architectures, ForAug improves ImageNet top-1 accuracy by up to 6 percentage points (p.p.) and yields gains of up to 7.3 p.p. on fine-grained downstream datasets. Moreover, the same control knobs enable targeted diagnostic tests: we quantify background reliance, foreground focus, center bias, and size bias via controlled background swaps and position/scale sweeps, and show that training with ForAug substantially reduces these shortcut behaviors and significantly increases accuracy on standard distribution-shift benchmarks by up to $19$ p.p. Our code and dataset are publicly available at https://github.com/tobna/ForAug.

cs.CV↗
arXiv2026

The Gentle Collapse: Distributional Metrics for Continual Learning

Accuracy degradation is the standard metric for Catastrophic Forgetting (CF), however, it records only whether forgetting occurred or not. It saturates at the extremes and collapses discretely at task boundaries, hiding the internal structure of what is being forgotten. We introduce six softmax-derived metrics spanning true-label rank (TLR), predictive confidence, and distributional divergence that characterize forgetting continuously, each normalized to [0, 1] with no modification to training. On CIFAR-100, these metrics carry information where accuracy does not: at 0% accuracy, the Confusion Margin spans an IQR of [0.32, 0.50] across classes that accuracy treats identically. We demonstrate that this richer signal is actionable in mitigating catastrophic forgetting. Per-sample metric scores used as loss weights reduce forgetting by 1.3 percentage points over uniform experience replay (ER) on CIFAR-100. Furthermore, the slope of a metric over a small window provides a stable sampling criterion: at a small-window size (e.g. 3 epochs), accuracy-trend degrades to 34.79% (std. = 2.32) while log-TLR achieves 41.07% (std. = 0.57). This gap is structural since reliable small-window trend estimation requires a continuous signal. On TinyImageNet, log-TLR trend sampling reduces forgetting by 7.7 percentage points over the ER baseline.

cs.LG↗
arXiv2026

GTA-Net: Cooperative Game Theory for Vision-Language Alignment in Chest X-Ray Report Generation

Automated chest X-ray report generation requires precise cross-modal grounding to ensure clinically reliable descriptions. However, existing vision-language models rely on implicit attention mechanisms that fail to enforce explicit region-word correspondence and disease-level consistency. We propose Game-Theoretic Alignment Network (GTA-Net), a vision-language framework that formulates report generation as a cooperative game-theoretic alignment problem. The model introduces a BinaryGameAligner that models interactions between image regions and text tokens using similarity-based payoff matrices with Shapley-inspired importance weighting. To enforce clinical semantics, we further develop a Disease-Aware Ternary Aligner, which captures joint interactions among images, reports, and structured disease concepts. GTA-Net combines a Swin-based visual encoder with a LoRA-adapted large language model and is trained with a unified objective for generation and alignment. Experiments on CheXpertPlus and IU-XRay demonstrate state-of-the-art performance across standard generation metrics and improved clinical consistency, highlighting the effectiveness of explicit game-theoretic alignment for medical vision-language generation.

cs.CV↗