Papers

Large vision-language models (LVLMs) can be adapted to specialized medical imaging tasks via parameter-efficient fine-tuning approaches such as low-rank adaptation (LoRA), leading to a growing ecosystem of expert models tailored to specific imaging modalities and clinical scenarios. However, deploying multiple expert LVLMs in practice incurs substantial computational and operational overhead. Model merging provides a promising solution by consolidating multiple experts into a single model without retraining, yet it remains largely unexplored in the medical domain. In this work, we present the first systematic study of model merging for medical LVLMs. We introduce MergeMedBench, a comprehensive benchmark spanning eight imaging modalities and diverse clinical task types, comprising 16 LoRA fine-tuned models built upon two mainstream architectures. We conduct an extensive evaluation of existing merging methods and further propose winner-take-all, a simple and hyperparameter-free approach that retains only the most dominant parameters across expert models. By preserving the critical parameters that govern model behavior and discarding weaker ones, our method avoids the information dilution inherent in averaging- or alignment-based strategies. Despite its simplicity, winner-take-all consistently outperforms existing approaches, offering both a new perspective on LoRA merging and a strong practical baseline for future research.

Multimodal AI agents increasingly rely on persistent long-term memory to ground generation in past visual and textual episodes. We show that unconditional trust in visual data creates a critical vulnerability. We propose Lucid, a black-box adversarial framework that compromises multimodal memory pipelines under a strictly image-bounded threat model, requiring no access to the target MLLM, target retrieval encoder, or the text channel. Lucid crafts imperceptible perturbations to enable two distinct failure modes based on the availability of historical context: (1) Memory poisoning, an in-context attack where the adversarial image replaces a benign one whose content is reinforced by prior textual context, reliably corrupting visual recall and steering the agent toward attacker-chosen narratives; (2) Memory injection, an out-of-context attack where the adversarial image replaces a benign one in a conversation turn devoid of prior textual grounding, causing the agent to generate attacker-influenced responses with no corrective signal from memory. We evaluate Lucid across various conversation domains and five black-box memory architectures, including graph-structured, LLM-summarized, and commercially deployed systems. Lucid achieves 61.6% ASR on poisoning and 58.4% ASR on injection, exposing a structural vulnerability in multimodal memory pipelines.

Bundle adjustment (BA) remains a critical refinement module for image-based 3D reconstruction and continues to improve geometric accuracy even in learning-based pipelines. However, in low-parallax and near-rotational regimes, classical Schur-based Levenberg–Marquardt (LM) often becomes ill-conditioned and yields unreliable pose and calibration estimates. We propose Gate-Guided CSS-BA, a solver-side modification of Schur-LM that preserves the classical BA objective and trust-region framework while constraining each update to a geometrically informed low-dimensional subspace. By integrating Column Space Search (CSS) with geometry-aware gating, the method stabilizes the Schur-LM update without altering the estimation problem. In contrast to keyframe or state-selection approaches, all camera and point parameters remain in the optimization problem; only the update direction is restricted. The method serves as a drop-in replacement for existing BA pipelines. Experiments on both generic and challenging weak-geometry scenarios show more stable optimization, improved relative pose accuracy, and competitive calibration behavior while maintaining reprojection quality.

Recent advances in AI-generated content have driven widespread adoption of Diffusion Transformers (DiTs) for high-resolution, long-duration content generation. While parallelization techniques accelerate diffusion inference, they face significant scalability challenges due to excessive communication overhead in multi-node environments. We observe that sequence partitions in Context Parallelism (CP) exhibit distinct heterogeneity: spatially proximate partitions contribute more significantly to attention computation results. By mapping this heterogeneous pattern to hierarchical communication topology, we can access high-contribution partitions with reduced communication cost. This insight motivates our novel selective attention state mechanism that strategically balances partial attention computation and historical result reuse across denoising steps. We present DiTango, an efficient parallel framework for DiT generation. DiTango features an anchor-guided state selection planner that optimizes computation-reuse decisions for each partition, complemented by a runtime that orchestrates efficient state-centric operations. This design achieves superior system efficiency while preserving generation quality. Experimental evaluation on popular diffusion models demonstrates that DiTango achieves up to 1.9x end-to-end and 3.2x attention speedup with near-linear scaling in multi-node settings, while maintaining generation quality comparable to state-of-the-art approaches.

Hyperspectral image (HSI) classification systems are increasingly deployed on platforms with strict computational budgets, such as UAVs and small spaceborne sensors. In these settings, accuracy alone is not enough; the model must also run within tight latency and memory constraints. Most recent HSI classifiers, however, focus on accuracy and pay relatively little attention to these constraints. We propose BCG-Former, a lightweight CNN-Transformer hybrid that targets this trade-off. The model introduces three innovations: (1) Band-Contextual Gating (BCG) for adaptive spectral recalibration using local inter-band context and learnable temperature sharpening, (2) a spectral summary token that bridges spectral and spatial features, and (3) single-pass Band-RoPE combined with linear attention for efficient joint representation learning. Evaluated on classical airborne (Pavia University, Salinas, Indian Pines, Houston 2013/2018) and UAV-borne benchmark datasets (WHU-Hi-LongKou, HongHu, and HanChuan), BCG-Former achieves over-all accuracy ranging from 91.51% on Houston 2018 to 99.49% on Houston 2013, while maintaining sub-millisecond inference latency (0.91-0.95ms) and using only 0.10-0.23M parameters. Across all eight benchmarks, BCG-Former consistently resides on or near the Pareto frontier of accuracy versus latency, outperforming or matching recent CNN-, Transformer-, and Mamba-based methods at a fraction of their computational cost. Ablation studies confirm that all three components are complementary, with BCG providing the largest individual contribution. These results establish BCG-Former as a strong accuracy-efficiency Pareto candidate for real-time and large-scale remote sensing applications.

The primate visual system is typically divided into two streams - the ventral stream, responsible for object recognition, and the dorsal stream, responsible for encoding spatial relations and motion. Recent studies have shown that convolutional neural networks (CNNs) pretrained on object recognition tasks are remarkably effective at predicting neuronal responses in the ventral stream, shedding light on the neural mechanisms underlying object recognition. However, similar models of the dorsal stream remain underdeveloped due to the lack of large scale datasets encompassing dorsal stream areas. To address this gap, we present STSBench, a dataset of large-scale, single neuron recordings from over 2,000 neurons in the superior temporal sulcus (STS), a nearly 50-fold increase over existing dorsal stream datasets, collected while Rhesus macaques viewed thousands of unique, natural videos. We show that our dataset can be used for benchmarking encoding models of dorsal stream neuronal responses and reconstructing visual input from neural activity.

Multi-reference image generation aims to synthesize images by integrating attributes from multiple reference images under textual instructions. As the number of references increases, the task necessitates complex semantic comprehension, such as correctly associating attributes with the intended subjects and planing out coherent spatial arrangement between subjects and their environments. Existing approaches, which rely solely on natural language instruction, often fail to capture these complex intentions precisely, leading to semantic misalignment and inconsistent generation. We identify two key factors behind these limitations: natural language instructions are often verbose and ambiguous, and high-quality multi-reference data is scarce. To address these issues, we propose StructGen, which employs a structured, dictionary-like format to encode multiple reference images, thereby enabling explicit and unambiguous specification of generation intentions. To support this design, we construct a structured dataset based on high-quality real images and develop a corresponding training framework, along with a dedicated benchmark for challenging multi-reference scenarios. Extensive experiments on both public benchmarks and our proposed benchmark demonstrate that StructGen consistently outperforms existing methods on both semantic alignment and detailed reference-generation consistency, especially under complex instructions with multiple references. The code is available at https://jianingpeng0382.github.io/StructGen/

Vision-language models trained with contrastive objectives have shown promise in medical image analysis. However, conventional global image-text alignment is ill-suited for mammography, where diagnostically relevant lesions are spatially localized and occupy only a small fraction of the image. Subtle morphological cues critical for malignancy assessment can be diluted when representations are learned at the whole-image level. In this work, we propose a novel region-grounded vision-language learning method for detection-guided mammographic lesion classification. The method mirrors radiologists' diagnostic paradigm. First, a region-text contrastive pretraining stage aligns lesion-specific features with structured clinical descriptors derived from radiology metadata. To mitigate semantic collapse and background bias in low-vocabulary settings, we introduce a multi-component objective incorporating positive alignment, fine-grained semantic hard negatives, and background suppression. Second, an auxiliary lesion detection head is jointly optimized with contrastive classification to preserve spatial sensitivity and enable localization-aware malignancy classification. Extensive experiments on two independent datasets, CBIS-DDSM and VinDr-Mammo, show superior performance of our method compared to related methods under in-domain, cross-dataset, and transfer learning settings.

Focal cortical dysplasia (FCD) is one of the leading structural causes of drug-resistant focal epilepsy, yet its subtle and heterogeneous imaging characteristics make accurate identification and delineation challenging on conventional magnetic resonance imaging (MRI). Although T1-weighted (T1w) and fluid-attenuated inversion recovery (FLAIR) images are routinely acquired for presurgical evaluation, the contribution of different MRI representations to deep learning-based FCD segmentation remains poorly understood. In this study, we present a systematic benchmark of MRI representations for automated FCD segmentation using the nnU-Net framework. A publicly available presurgical MRI dataset comprising 85 FCD subjects and 25 healthy controls was used to evaluate eight input configurations, including conventional MRI contrasts (T1w and FLAIR), ratio-derived representations, and their multimodal combinations. To isolate the effect of MRI representation, all experiments employed identical preprocessing, network architecture, optimization strategy, and five-fold cross-validation. Among the evaluated single-modality representations, FLAIR achieved the strongest overall performance, whereas ratio-derived representations alone were insufficient for reliable identification of subtle FCD. Incorporating ratio-derived representations with conventional T1w and FLAIR images consistently improved lesion delineation, with the four-channel multimodal configuration achieving the highest overall Dice score (0.376), representing a 5.0% relative improvement over the conventional T1w+FLAIR representation. These findings demonstrate that MRI representation design is an important yet underexplored component of deep learning-based FCD segmentation and should be optimized alongside network architecture.

This paper proposes a neural network for low light image enhancement (LLIE) based on retinex theory to make LLIE robust for various dynamic range scenes. The retinex theory is an image formulation model inspired by a human color perception hypothesis, where a low light image is decomposed into intrinsic color context (i.e., reflectance map) and scene-dependent illumination (i.e., illumination map). Due to non-uniqueness of its decomposition, existing retinex-based LLIE methods often fail to achieve stable decomposition, which lead to over-enhancement. Typically, they are sensitive to the dynamic ranges that vary in different lighting conditions. To tackle this issue, we propose WREN: An LLIE neural network with double U-Net-like structures. WREN consists of two U-Net-like sub-networks. The first network has one encoder and two decoders that decompose an input image into the reflectance and illumination maps. The second network with a customized Transformer block between an encoder and a decoder only enhances the illumination map obtained from the first network: This completely follows the assumption of the retinex theory. Finally, the enhanced illumination map is recombined with the reflectance map. The network is trained end-to-end with a scale-invariant loss function, which gives robustness against the illumination scaling. Numerical results show that our method achieves the state-of-the-art performance across multiple datasets. Our code is available online.

Monocular depth foundation models have demonstrated remarkable generalization capabilities across diverse environments. However, they continue to struggle with metric depth estimation in diverse environments. This limitation stems from the inherent scale ambiguity of single-view inference, leading to misaligned scale predictions even when the relative geometry is accurate. Conversely, recent multi-view foundation models leverage cross-view cues to learn robust scene-level geometry and consistent scale. Yet, these benefits typically vanish during single-image inference, as the absence of explicit geometric constraints causes performance to degrade. To bridge this gap, we propose a novel framework that transfers the scale-aware geometric priors of multi-view models into monocular depth foundation models. Specifically, we introduce an Epipolar Distillation (EpiDistill), an approach utilizing Rectified Stereo Tokens, which enables the single-view prediction model to retain epipolar attention patterns and maintain geometric consistency without requiring multi-view inputs at inference. Experimental results demonstrate that our method significantly improves zero-shot metric depth estimation, particularly on challenging datasets like ETH3D and DIODE where scale alignment is critical. Furthermore, our approach is model-agnostic, consistently boosting the performance of state-of-the-art ViT-based models, including UniDepthV2 and DepthPro.

Where should the question go in a vision-language model (VLM) prompt: before the image or after it? Intuition says before: knowing what is asked should tell the model where to look. Yet across visual question answering benchmarks, question-first prompting consistently underperforms the image-first ordering recommended for frontier VLMs, a phenomenon we term the question-first paradox. We trace this paradox to a conflict between two stages of VLM computation. Logit-lens and attention probes show that question-first prompting steers perception, shifting image patch representations toward question-relevant concepts. But downstream, stranded behind hundreds of image tokens, the question is barely attended by the answer token, which instead commits to image-driven, often wrong answers. Causal attention knockout confirms that the answer reads the question only when it follows the image. This diagnosis yields a training-free fix: question echoing, restating the question on both sides of the image so one copy steers perception while the other is available at answer time. A similar division of labor appears in a fifty-year-old finding on human 'adjunct questions', where repeating a question before and after a passage improves comprehension. Echoing the image as well brings further gains by restoring the whole-image view otherwise lost by a causal decoder. The paradox holds across five open VLMs, costing up to 17.5 group-accuracy points. Echoed prompts recover most of the gap and, on NaturalBench and Winoground, surpass the best single-pass ordering by up to 19 group-accuracy points on Winoground, with no training, fine-tuning, or architecture change. The paradox reveals a tension between steering what a model sees and preserving access to what it was asked; echoing resolves this through prompt design. Project Page: https://rakshanda-cmu.github.io/ask-twice-look-twice/

Recent visual place recognition (VPR) methods based on vision transformers, particularly foundation models, have achieved remarkable recognition performance. However, these models process all visual tokens throughout the entire network, resulting in substantial computational overhead, which hinders their deployment in real-time and resource-constrained scenarios. A natural question thus arises: are all visual tokens necessary for VPR? To answer this question, we present the first systematic benchmark of token reduction for efficient visual place recognition. Our benchmark comprehensively evaluates representative token pruning, token merging, and hybrid pruning-merging methods across multiple state-of-the-art VPR models and diverse benchmark datasets covering urban, suburban, and natural environments. We further investigate token reduction from multiple perspectives, including recognition performance under different reduction configurations, computational complexity, inference speed, qualitative visualization, and deployment efficiency on edge devices. Through extensive experiments and in-depth analysis, our benchmark reveals multiple important characteristics of token reduction in VPR and provides several practical insights into the trade-offs between accuracy and inference efficiency. For example, token reduction can reduce computational cost by up to 29% and improve throughput by up to 44%, while incurring less than 1% degradation in recognition accuracy. Overall, this work establishes a comprehensive foundation for future research on token-efficient VPR and efficient visual retrieval systems. Our codes and models will be available at https://github.com/Tong-Jin01/TokenReduction4VPR

3D CT vision-language models (VLMs) classify abnormalities from text prompts in a zero-shot manner, enabling cross-institution deployment where labels are scarce and clinical tasks shift faster than supervised models can be retrained. A real CT scan, however, typically contains several co-occurring abnormalities, and the reliability of zero-shot multi-label prediction under distribution shift remains poorly understood. Test-time adaptation (TTA) updates a model on unlabeled target scans without source data or target annotations, yet existing TTA methods target multi-class softmax prediction on natural images or 2D medical segmentation, and none addresses unsupervised multi-label adaptation for zero-shot 3D CT VLMs. We study when TTA helps zero-shot 3D CT VLMs. A controlled diagnostic analysis shows that TTA is conditional: the volumetric input must preserve the encoder's depth structure, and the base representation must transfer to the target cohort, with depth reduction alone lowering internal AUROC by more than 0.12. We then focus on the regime where the base model already separates present from absent abnormalities. We introduce CARVE (Cardinality-Aware Retained-View Entropy), the first TTA method for this setting. CARVE estimates a sample-specific positive-label cardinality $\hat{k}$, optimizes a top-$\hat{k}$ objective to preserve co-occurring abnormalities, and performs memory-efficient multi-view adaptation by scoring weak 3D views without gradients before updating on a retained subset. Across contrastive CT-CLIP and anatomy-aware fVLM, CARVE provides the most consistent improvements across multi-label, three-class, and binary CT tasks when the base model is already discriminative. These results establish multi-label TTA for zero-shot 3D CT VLMs as a distinct problem and CARVE as a cardinality-aware solution.

On-the-fly reconstruction is a key requirement for many applications in robotics and autonomous navigation. Variational Bayes Gaussian Splatting (VBGS) enables continual learning without replay buffers using Coordinate Ascent Variational Inference (CAVI), but its per-frame iterations over all observed points make it too slow for real-time use with strict memory and latency requirements. We present ImprovedVBGS, an accelerated framework for on-the-fly continual reconstruction. This is achieved primarily through (i) spatially truncated variational inference, and (ii) improved reassignment that uses forwarding, truncation and eliminates wasteful dynamic recompilation. On the NeRF synthetic dataset, we reduce mean per-frame latency from ~84.0 s to ~0.050 s on an RTX 3070 Ti, a 1680x speed-up while maintaining reconstruction quality. We also improve novel-view synthesis quality using an exact renderer with no added training costs.

3D Gaussian Splatting (3DGS) captures scenes by coupling explicit geometry (position, covariance) with view-dependent photometry (Spherical Harmonics). However, building $\mathrm{SE}(3)$-equivariant architectures on these primitives presents a fundamental representation bottleneck. Color has been treated as a signal rather than a geometric entity, making it nontrivial to unify symmetry across geometry and appearance as the camera frame changes. While translations are handled by relative coordinates, rotations act heterogeneously across attributes: $\mu\mapsto R\mu$, $\Sigma\mapsto R\Sigma R^\top$, and $f_\ell\mapsto D^\ell(R)f_\ell$. This mismatch complicates strict equivariance, leading existing methods to either discard or flatten SH coefficients, thereby breaking symmetry. We propose a unified solution rooted in representation theory: for SH degrees $\ell\le2$, photometry is algebraically isomorphic to a rank-2 geometric tensor. We prove that the Wigner-$D$ action on these SH coefficients can be exactly reformulated as the conjugation action on $3\times3$ matrices. Leveraging this, we introduce the Unified Matrix Embedding, a lifting that maps all Gaussian attributes into a unified carrier space, $\mathfrak{gl}(3)$. Building on the "Color-as-Geometry" formulation, we present E3DGS, a rigid-body ($\mathrm{SE}(3)$) equivariant architecture that processes 3D Gaussians without Clebsch-Gordan tensor products. Evaluations on object vision and action-conditioned Gaussian world modeling demonstrate that our unified approach yields strong robustness under camera-frame changes and improved data efficiency.

Physical simulations that predict the behavior of urban disasters, such as climate-related flooding, play a crucial role in disaster prevention and the development of anomaly detection models. However, the severe shortage of flood data in real-world environments, combined with the inherent distortions of fisheye lens images, which are used for urban surveillance, has made high-precision simulations challenging. To address this, we propose a new physical simulation system PhysFlood that leverages Diffusion Models to synthesize realistic floods from just a single image captured by a fisheye lens. Our system not only enables simulation from a single image, but also features the ability to freely control and generate diverse flood scenarios by manipulating physically meaningful variables, such as water levels. In our evaluation experiments, we conducted a qualitative human study and demonstrated that the simulation images generated by PhysFlood exhibit both acceptable realism and robustness.

Four-finger SLAP fingerprints are flat live-scan impressions of the index, middle, ring, and little fingers of one hand, used for identity verification in border control and law enforcement. No benchmark has evaluated whether multimodal large language models (MLLMs) can verify identity from SLAP images. We introduce SLAPBench, the first benchmark for MLLM-based four-finger SLAP fingerprint verification, built from NIST SD302b with 7,832 pairs (176 mated, 7,656 non-mated). We evaluate four open-source MLLMs (InternVL3-8B, Qwen2.5-VL-7B, Qwen3-VL-8B, Gemma-3-12B) and the proprietary Claude Opus 4.8 under zero-shot, task-description, and similarity-scoring prompts. Prompting governs verification behavior. Task-description prompting collapses all four open-source models to near-100% False Accept Rate (FAR), and Gemma-3-12B collapses under zero-shot as well; Claude Opus 4.8 alone resists collapse under both binary prompts, giving the best binary result (FAR = 20.2%). Similarity scoring removes collapse across the open-source models and exposes wide capability gaps: Claude reaches AUC = 0.953 and Gemma-3-12B 0.837, while InternVL3-8B is inverted (AUC = 0.590) and Qwen2.5-VL-7B near random (0.567). Qwen3-VL-8B attains perfect separation (AUC = 1.000), which we treat as a diagnostic rather than as capability: SD302b holds one SLAP capture per finger position, so mated pairs are cross-resolution. A matched-resolution control leaves the perfect score intact, ruling out the resolution shortcut; what cannot be excluded within SD302b is near-duplicate detection, since a mated pair is one capture rendered twice. A fairness probe over gender, race, and age suggests disparity grows as discrimination weakens. SLAPBench establishes the first SLAP-specific MLLM baseline and shows that prompting governs collapse while model capability governs discrimination.

Surgical phase recognition maps each video frame to a clinically meaningful workflow phase, supporting context-aware assistance, documentation, and postoperative analysis. Most methods treat phase annotations only as class IDs, whereas recent surgical vision-language models often require additional video–text data, captions, or instruction tuning. We propose LaSeD, a label-semantic self-distillation framework that uses phase names as privileged training-time context while retaining visual-only deployment without a ground-truth phase-name hint. LaSeD initializes a frozen teacher and a student from the same pretrained VLM checkpoint. The teacher receives the frame, a fixed task prompt, and the ground-truth phase-name hint; the student receives the same frame and prompt without the hint, and only its visual encoder is optimized. Training combines hard phase-token supervision with feature-level distillation from cached teacher representations. At inference, the teacher and hint are removed, and the student predicts one of the seven Cholec80 phases through constrained digit-token logits without an additional classifier head. On the Cholec80 evaluation split, LaSeD achieves 86.20% accuracy, 77.75% macro recall, 78.13% macro precision, and 64.15% macro Jaccard. Under the identical protocol, it improves a visual-only Qwen3-VL-4B baseline by 9.45, 9.42, 11.93, and 12.22 percentage points, respectively. Visual-only means that the image is the only sample-specific inference input, while all frames share the same fixed task prompt. These results suggest that phase names provide a useful low-cost signal for adapting VLMs to surgical workflow analysis. (The code will be published soon.)

Video diffusion models (VDMs) synthesize photorealistic content, yet they often fail to follow the course that a physical phenomenon should take within a given scene. Recent training-free methods let a vision-language model (VLM) plan the phenomenon and guide a frozen VDM toward the plan; however, such plans are derived from the prompt and consumed as whole keyframes or trajectories, which leaves unspecified where the consequences land in the observed scene and turns incidental visual details into optimization targets. We observe that a phenomenon specified in words unfolds as sparse, local changes to the physical state of the observed scene. Building on this observation, we present PhysPlan, a training-free image-to-video framework that represents a phenomenon as a grounded state graph and uses this graph to decide what, where, and when the guidance constrains. Grounded Physical State Reasoning decomposes the phenomenon into physical deltas, each stating which objects change, to what state, and by which physical rule, and translates each delta into graph edits, verified by deterministic checks, that leave all other objects unchanged. Graph-Guided Test-Time Optimization renders a keyframe for each state, measures the denoised estimates only along the properties selected by the edits, and concentrates the update on the edited objects. On PhyGenBench and Physics-IQ, PhysPlan raises its base model from 0.52 to 0.77 and from 27.1 to 38.2, surpassing the strongest prior I2V method (0.60 and 34.6), and lowers FVD by over 20%. Project page: https://physplan.github.io

With the increased adoption of robotic agents operating in human environments by scanning and sharing 3D representations (e.g., for fleet learning, cloud-based planning, or collaborative mapping), collected point clouds reveal not just the objects in a scene but also sensitive spatial context, such as room function or information that occupants never consented to disclose. Traditional point cloud encoders offer no principled control over this: either all is preserved, or none. Hence, we introduce RoboShape, an information theory guided compression head following the frozen Sonata encoder. We project voxel-level embeddings using the Donsker-Varadhan formulation of mutual information (MI). Specifically, we maximize the MI between embeddings and object-level understanding while minimizing it for private attributes. RoboShape leads to 87.5% smaller embeddings that retain 98.7% of object classification utility while collapsing sensitive attribute predictions by 39.3% across the three real-world indoor LiDAR datasets. Its privacy-preserving embeddings are cheaper to transmit over the network or to train a model for any downstream tasks. We release the RoboShape codebase to give the robotics community a practical, encoder-agnostic tool for building perception pipelines that are compact, privacy-aware, and deployment-ready.

Foundation vision models trained on natural images transfer to medical tasks without domain pre-training, but volumetric classification requires aggregating tens of thousands of patch tokens per study, and the aggregator constrains how the resulting model can be interpreted. We compare three aggregators on identical frozen DINOv3 ViT-H/16+ features for renal tumour/cyst detection on KiTS23 (966 kidneys; n=97 test): a CLS-token linear probe, gated attention multiple instance learning (MIL) over 55,296 patch tokens, and a prototype head following ProtoViT. Attention MIL achieves the highest AUROC for tumour (0.74, 95% CI 0.64-0.83) and cyst (0.80, 0.70-0.88), with attention enriched 7.5-9.8x over chance within annotated lesions. The prototype head does not transfer to cyst detection (AUROC 0.51), exposing an interpretability-performance trade-off at this token scale.

This article presents DMFNet, a dual-backbone multiscale feature fusion framework with residual feature propagation and spatial attention for remote sensing scene classification. Existing approaches often face challenges in effectively capturing multiscale feature interactions and learning robust feature representations from complex aerial scenes with high intra-class variability and inter-class similarity. To address these limitations, the proposed framework employs two pretrained backbone networks to extract diverse hierarchical feature representations. A multiscale feature fusion mechanism with residual feature propagation is introduced to enhance feature interaction across multiple resolution levels. In addition, a spatial attention module is introduced to emphasize informative spatial regions in multi-object scenes. Further, a two-stage training strategy consisting of backbone freezing followed by selective fine-tuning is adopted to ensure stable optimization and improved generalization. Experiments conducted on the benchmark AID dataset demonstrate that the DMFNet achieves an average accuracy of 97.46% $\pm$ 0.14%. Ablative analysis further show the importance of various components in unison.

Attacks on general computer vision algorithms are often relegated to the digital domain, with the optimization performed purely in the digital world and then translated to physical mediums for implementation. In the field of biometrics, including facial recognition, physical presentation attacks targeting biometric sensors are dominant and present significant opportunity and risk. This paper highlights a critical vulnerability in the physical-to-digital pipeline of biometric sensors and provides a standardized approach for testing facial recognition system robustness against hardware attacks, going beyond and potentially complementing presentation attacks (as defined in ISO/IEC 30107 standard series). Specifically, in this work we (a) demonstrate that intentional electromagnetic interference is possible to be conducted with commonly accessible radio frequency (RF) equipment, (b) assess the robustness of state-of-the-art face recognition methods against RF-based attacks, and (c) provide a dataset composed of face images captured with and without electromagnetic interference to serve as a new benchmark for testing modern face matchers against RF-sourced interference.

We present a fully automated closed-loop AutoML framework that uses GPT-5, GPT-4o, and Claude Sonnet 4 as autonomous neural architecture designers for cross-lingual handwritten optical character recognition. Each large language model independently generates, trains, evaluates, and iteratively refines neural network architectures using performance feedback from previous trials. The framework is evaluated on Arabic, Persian, and English handwriting datasets through 270 independent experiments. It consistently discovers accurate and computationally efficient models without manual architecture design, domain-specific preprocessing, or hyperparameter tuning. The generated models achieve mean test accuracies above 93 percent, a best accuracy of 98.1 percent, and inference latency between 41 and 44 milliseconds. The results demonstrate that large language models can function as effective AutoML agents for neural architecture search, enabling scalable, script-adaptive, and reproducible handwriting recognition across languages.

Cross-view geo-localization matches ground-level observations against geo-tagged satellite imagery. Recent methods show that sequential queries such as video clips yield richer spatiotemporal cues than single images, yet they overlook a complementary sequential modality: route descriptions – which capture the same trajectory at a higher level of abstraction and are often the only input available (e.g., a user directing an autonomous vehicle to a pickup point). To bridge this gap, we introduce SeqGeo-VL, a dataset of $\sim$39K video-text-satellite triplets, and TrajLoc, a unified framework capable of processing both video clips and route descriptions. By leveraging both dense visual and abstract linguistic semantics, TrajLoc enables these modalities to mutually reinforce cross-view matching. We further propose TrajMod, a lightweight module that conditions query embeddings on trajectory geometry, yielding spatially-aware representations. Experiments show that TrajLoc achieves substantial gains over state-of-the-art methods on both video and text geo-localization. The project page is available at https://humblegamer.github.io/trajloc/.

Compact medical-image classifiers need efficiency and interpretable evidence, yet these goals are often addressed separately. We introduce qZACH-ViT, a quantization-aware extension of the zero-token (CLS-token-free), position-free ZACH-ViT backbone with recursive intrinsic patch-level class evidence. We also introduce Recursive Attribution-Stabilized Optimization (RASO), which norm-matches classification and attribution gradients and removes attribution components that conflict with classification. We evaluate four controlled conditions on seven MedMNIST datasets using 50 training images per class and ten fixed seeds, completing 280 runs. All 210 qZACH-ViT checkpoints are converted to executable mixed-precision ONNX INT8 graphs containing 16 signed INT8 MatMulInteger projections with INT32 accumulation. Deployed mixed-precision INT8 qZACH-ViT with Adam improves the FP32 ZACH-ViT baseline mean on all seven datasets, with a mean paired gain of 0.0313 in the dataset-specific primary metric; qZACH-ViT with RASO yields a mean gain of 0.0368. Across 964,920 source-to-INT8 test comparisons, prediction agreement is 99.9751%, with a mean absolute primary-metric change of 0.000133 and a maximum of 0.004386. Across 3,600 matched intrinsic maps, mean cosine similarity is 0.999955, mean rank correlation is 0.9944, and mean top-10% overlap is 0.9692. ONNX artifacts are 70.0% smaller than source checkpoints and provide $1.41\times$ and $2.39\times$ end-to-end CPU speedups with one and four threads. RASO significantly reduces sufficiency error and improves input-noise stability over Adam with the same attribution loss, but does not dominate every predictive or explainable artificial intelligence (XAI) metric. These results establish qZACH-ViT as a deployable compact intrinsically explainable model and RASO as a targeted stability-oriented optimization procedure.

We introduce DrawingVQA, the first benchmark designed to evaluate multimodal large language models (MLLMs) on real-world construction drawings – a core media in architecture, civil, and many other engineering practices. Unlike natural images or schematic floor plans, construction drawings fuse abstract geometry, symbolic notation, tabular data, annotations, and domain-specific text, forming a uniquely complex visual-textual domain core to engineering workflows. DrawingVQA bridges this gap with 33 "Issued for Construction" drawings and 92 expertly curated question-answer pairs, spanning three reasoning depths: perceptual understanding, contextual interpretation, and domain-expert reasoning. To evaluate model capabilities, we present a dual categorization framework to jointly analyze performance across seven construction-engineering and four MLLM capability dimensions – the first to explicitly map engineering workflows to AI reasoning competencies. Evaluations of state-of-the-art MLLMs reveal a substantial gap between model and expert performance, particularly at higher reasoning depths. This benchmark lays a foundation for domain-specialized multimodal reasoning to allow for advancement on integration of AI-driven understanding and real-world engineering workflows.

Reliable AI for screening mammography requires training data representative of the low cancer prevalence and subtle abnormalities found in screening populations. We examined whether supplementing such data with biopsy-confirmed cases from abnormal-enriched external datasets improves performance. Using the Newfoundland and Labrador Breast Screening Dataset (NLBSD) alongside CBIS-DDSM and CMMD, we evaluated an EfficientNet-B5 encoder initialized with Mammo-CLIP weights as a frozen linear probe under consistent preprocessing and patient-level splits. The NLBSD-only model achieved an AUC-ROC of 0.737 (95% CI [0.686, 0.785]). Adding external positive cases reduced performance in every configuration (AUC-ROC = 0.620–0.644; DeLong test, Holm-corrected $p < 0.05$), with degradation increasing as additional sources were introduced. Domain-matched evaluation produced modest gains only when the training and test domains coincided, and no configuration surpassed the NLBSD-only model. As a diagnostic, we reframed the task as predicting each examination's dataset of origin. The datasets were separated almost perfectly despite identical preprocessing, indicating that dataset-specific characteristics strongly influence the learned representation. These findings show that naïvely pooling abnormal-enriched mammography datasets can introduce domain shift that outweighs the benefit of additional positive cases. Differences in acquisition, intensity mapping, and dataset construction persist after normalization, motivating domain-aware strategies for combining heterogeneous mammography datasets.

Our study provides evidence that CNNs struggle to extract orientation features effectively. We show that using the Complex Structure Tensor, which contains compact orientation features with certainties, as input to CNNs consistently improves identification accuracy compared to grayscale inputs alone. Experiments also demonstrated that our inputs, provided by mini-complex convnets, combined with reduced CNN sizes, outperformed full-fledged, prevailing CNN architectures. This suggests that the upfront use of orientation features in CNNs, a strategy seen in mammalian vision, not only mitigates their limitations but also enhances their explainability and relevance to thin-clients. Experiments were conducted on publicly available datasets comprising periocular images (Cross-Eyed and PolyU) for biometric identification and verification in both Close-World and Open-World Scenarios using six CNN architectures. Our experiments on the Cross-Eyed and PolyU datasets yield a 5-26% reduction in EER, providing strong empirical evidence that explicit orientation priors mitigate CNN representational limits in Open-World and Close-World scenarios.