Vision Foundation Models

Interpreting the decisions of deep image classifiers remains challenging, particularly in black-box settings where model internals are inaccessible. We introduce OCCAM, a framework for open-set causal concept explanation and ontology induction in vision models. OCCAM discovers visual concepts in an open-set manner, localizes them via text-guided segmentation, and performs object-level interventions by removing concepts to measure changes in class confidence, estimating each concept's causal contribution. Beyond local explanations, OCCAM aggregates interventional evidence across a dataset to induce a structured concept ontology that captures how classifiers globally organize visual concepts. Reasoning over this ontology reveals consistent dependencies between concepts, exposes latent causal relations, and uncovers systematic model biases. Experiments on Broden and ImageNet-S across multiple classifiers show that OCCAM improves explanation quality in open-set black-box settings while providing richer global insight than per-image attribution methods.

Automated structural health monitoring is essential to prevent catastrophic infrastructure failures. Precise, pixel-level defect segmentation is needed to accurately assess structural integrity, but progress in defect segmentation for civil infrastructures has been held back by an extreme scarcity of data, which requires costly expert annotation. The need for data is accentuated by algorithmic hurdles intrinsic to the problem, including center-bias and the need to rely more on shape when inspecting nearly textureless building materials. To remove the bottleneck, we introduce Cracks in the Foundation (CiF), the largest and most detailed civil infrastructure (instance) segmentation dataset to date, comprising $\approx$150,000 high-resolution images meticulously curated over five years in collaboration with civil engineering experts. With the help of this unprecedented data source, we expose a blind spot of current visual AI: despite the advent of promptable Foundation Models (FMs) and Vision Language Models (VLMs), and despite the impressive abilities of today's specialised segmentation models, it turns out that dense image understanding in the built environment is nowhere near solved. Our evaluations indicate that even the most recent zero-shot FMs face significant challenges when deployed on real-world infrastructure and even the performance of specialised models with domain-specific supervision plateaus at $\approx$25% mAP. CiF establishes inspection of civil infrastructure, an elementary and seemingly easy perceptual task, as an open challenge that reveals fundamental weaknesses of present-day models trained predominantly on internet images, literally and figuratively highlighting cracks in the current foundation model paradigm.

In this work, we explore the largely unexplored direction of building a generalist image tokenizer directly on top of a frozen vision foundation model (VFM). To build this tokenizer, we utilize a frozen VFM as the encoder and introduce two key innovations: (1) a region-adaptive quantization framework to eliminate spatial redundancy in standard 2D grid features, and (2) a semantic reconstruction objective that aligns the decoded outputs with the VFM's representations to preserve semantic fidelity. Grounded in these designs, we propose VFMTok, a generalist visual tokenizer capable of operating seamlessly in both discrete and continuous latent spaces. VFMTok achieves substantial improvements in synthesis quality while drastically enhancing token efficiency. For discrete autoregressive (AR) generation, it accelerates model convergence by \textbf{3 times} and achieves a state-of-the-art gFID of \textbf{1.36} on ImageNet class-conditional synthesis. Similarly, for continuous-space generation, integrating VFMTok with a denoising model yields an exceptional gFID of \textbf{1.25}. Furthermore, because the latent space inherently captures rich spatial semantics, VFMTok enables high-fidelity class-conditional synthesis without classifier-free guidance (\textbf{w/o CFG}) across both generative paradigms, significantly accelerating inference speed. Beyond these remarkable empirical results, we systematically investigate the underlying mechanisms of our approach. We discover that the specific self-supervised learning objectives utilized during VFM pre-training dictate its effectiveness as a tokenizer. Specifically, a VFM jointly optimized with global contrastive learning and latent masked image modeling provides the optimal representations for image tokenization. These insights establish a strong foundation and offer valuable guidance for the design of future image tokenizers.

Physical adversarial attacks often overfit single surrogate models and optimization objectives. While ensemble attacks can mitigate this, existing methods struggle with severe gradient conflicts within restricted physical texture spaces, significantly degrading cross-model transferability. To bridge this gap, this paper proposes a Joint Multi-Objective and Multi-Model Optimization Framework (JMOF) that leverages quantitative similarity analysis to select the optimal surrogate model ensemble. Within JMOF, a dual-level mechanism jointly suppresses prediction outputs and flattens intermediate feature distributions, balancing attack efficiency with deep generalization. Additionally, an Orthogonal Gradient Alignment (OGA) strategy resolves cross-model gradient conflicts, transforming mutually repulsive gradients into synergistic optimization directions. Extensive simulated and real-world experiments demonstrate that JMOF outperforms state-of-the-art baselines against diverse black-box detectors. Crucially, JMOF exhibits substantial cross-vision-task generalization, generating attacks capable of simultaneously deceiving object detection and semantic segmentation or monocular depth estimation models. This research advances the generalization limits of physical adversarial attacks, providing a robust framework for evaluating visual AI vulnerabilities in real-world deployments.

The preservation and interpretation of cultural heritage increasingly rely on digital technologies, among which Knowledge Graphs (KGs) stand out for their ability to structure vast amounts of data. However, the construction and expansion of these KGs often face challenges due to the diverse and complex nature of cultural heritage information. In this paper, we propose a novel approach for extending KG resources in the domain of cultural heritage, which we applied to French data. First, we introduce a new knowledge graph in the domain of French cultural heritage, WJoconde, which is distinguished by its multimodality as it integrates both textual and image information of the entities. We further introduce three variants of WJoconde to facilitate downstream research, such as Knowledge Graph Completion (KGC). We also built a comprehensive benchmark for KGC methods on our dataset. Second, we propose a new framework for extending cultural heritage KGs using multi-modal approaches leveraging Large Language Models (LLMs) and Vision-Language Models (VLMs), which includes automated data extraction from unstructured resources combined with a special validation pipeline for grounding the output of both models, to further extend WJoconde. Our results show that by integrating the rich text and image information in cultural heritage data, we can efficiently enhance KGs with high reliability. We open-source all code and benchmark datasets with text and images, as well as the original data with an interactive access point

Open world image segmentation aims to achieve precise segmentation and semantic understanding of targets within images by addressing the infinitely open set of object categories encountered in the real world. However, traditional closed-set segmentation approaches struggle to adapt to complex open world scenarios, while foundation segmentation models such as SAM exhibit notable discrepancies between their strong segmentation capabilities and relatively weaker semantic understanding. To bridge these discrepancies, we propose WOW-Seg, a Word-free Open World Segmentation model for segmenting and recognizing objects from open-set categories. Specifically, WOW-Seg introduces a novel visual prompt module, Mask2Token, which transforms image masks into visual tokens and ensures their alignment with the VLLM feature space. Moreover, we introduce the Cascade Attention Mask to decouple information across different instances. This approach mitigates inter-instance interference, leading to a significant improvement in model performance. We further construct an open world region recognition test benchmark: the Region Recognition Dataset (RR-7K). With 7,662 classes, it represents the most extensive category-rich region recognition dataset to date. WOW-Seg attains strong results on the LVIS dataset, achieving a semantic similarity of 89.7 and a semantic IoU of 82.4. This performance surpasses the previous SOTA while using only one-eighth the parameter count. These results underscore the strong open world generalization capabilities of WOW-Seg. The code and related resources are available at https://github.com/AAwcAA/WOW-Seg-Meta.

Although large-scale visual foundation models (VFMs) achieve remarkable performance in semantic understanding, they still underperform in instance-aware dense prediction tasks. They exhibit different biases in representation: for instance, promptable segmentation models (e.g., SAM2) focus on fine-grained region boundaries, while self-supervised models (e.g., DINOv3) emphasize object-level structure. This observation highlights the potential of combining complementary features from different VFMs to enhance downstream dense prediction tasks. However, naive multi-VFM fusion seldom leads to reliable gains, and interpretable principles for leveraging their complementary features are still underexplored. In this work, we propose a metric-guided approach that effectively selects and aggregates complementary features from different VFMs based on explicit assessment scores. Specifically, we design a suite of label-free metrics in feature space across two aspects, Structural Coherence and Edge Fidelity, to assess features of VFM encoders. Guided by these scores, we identify complementary edge-strong and structure-strong encoder pairs, and integrate them via a master-auxiliary fusion scheme. This feature fusion requires no complex architectural changes and is trained only in a single stage. Our model shows consistent performance gains across multiple dense prediction tasks compared with the baselines, with better object-level semantics and more accurately localized boundaries. The code is available at {https://github.com/gyc-code/metric-guided-fusion}.

3D change detection from multi-view images is essential for urban monitoring, disaster assessment, and autonomous driving. However, existing methods predominantly operate in the 2D domain, where viewpoint variations are mistaken for physical changes and depth is unavailable. While visual geometry foundation models like VGGT rapidly produce dense point clouds from unposed images, independent per-epoch reconstruction encounters fundamental obstacles: unpredictable inter-epoch scale ambiguity, registration-change paradox where scene changes corrupt alignment, and pervasive edge-flying noise. To address these challenges, we present VGGT-CD, a training-free pipeline decoupling cross-temporal registration from dynamic-change interference. In the Coarse Stage, sparse keyframe joint inference establishes a unified metric space and yields an initial Sim(3) prior. In the Fine Stage, dense reconstructions are purified by isolating static-background correspondences. A closed-form centroid alignment refines the translation while locking scale and rotation, using a residual self-check to mathematically guarantee non-degradation. Evaluated on an 11-scene benchmark from the World Across Time dataset, VGGT-CD reduces Absolute Trajectory Error by 44% outdoors and 59% indoors. It completes registration over 6 times faster, producing high-purity 3D change maps without task-specific training.

Reconstructing coherent 3D geometry and appearance from unposed multi-view images is a fundamental yet challenging problem in computer vision. Most existing visual geometry foundation models predict explicit geometry by regressing pixel-aligned pointmaps, often suffering from redundancy and limited geometric continuity. We propose IVGT, an Implicit Visual Geometry Transformer that implicitly models continuous and coherent geometry from pose-free multi-view images. This formulation learns a continuous neural scene representation in a canonical coordinate system and supports continuous spatial queries at any 3D positions, retrieving local features to predict signed distance (SDF) values and colors using lightweight decoders. It allows direct extraction of continuous and coherent surface geometry, enabling rendering of RGB images, depth maps, and surface normal maps from arbitrary viewpoints. We train IVGT via multi-dataset joint optimization with 2D supervision and 3D geometric regularization. IVGT demonstrates generalization across scenes and achieves strong performance on various tasks, including mesh and point cloud reconstruction, novel view synthesis, depth and surface normal estimation, and camera pose estimation.

The paradigm of learning from automatic annotations driven by pre-trained experts and Foundation Models dominates data-hungry applications. However, it introduces a critical challenge: model-induced label noise. Unlike stochastic noise in classical robust learning, this noise stems from annotator inductive biases, manifesting as systematic errors tightly coupled with local feature manifolds. Existing methods relying on global transition matrices underfit these structural patterns, while learning instance-specific matrices remains mathematically intractable. We propose Model-Induced Noise Decoupling (MIND), a theoretically grounded framework addressing this dilemma. We demonstrate that the high-dimensional noise manifold can be decoupled into tractable, subspace-dependent components via Latent Manifold Disentanglement. Specifically, our Latent Decoupling Estimator (LDE) dynamically projects samples into latent structural clusters with consistent error modes, facilitating noise identifiability without ground-truth anchor points. To rigorously evaluate robustness, we adopt a hierarchical protocol: moving from controlled noise on CIFAR-100 to a structural stress test on large-scale real-world 3D datasets (S3DIS, ScanNet), where error patterns explicitly couple with geometric manifolds. Empirically, MIND significantly outperforms state-of-the-art methods on these complex benchmarks and effectively corrects zero-shot hallucinations from Vision-Language Models (e.g., OpenSeg), highlighting its potential as a robust distillation framework for Foundation Models.

Deep neural networks trained with different architectures, objectives, and datasets have been reported to converge on similar visual representations. However, what remains unknown is which visual properties models actually converge on and which factors may underlie this convergence. To address this, we decompose the object similarity structure of 162 diverse vision models into a small set of non-negative dimensions. To determine universal versus model-specific dimensions, we then estimate how often each dimension reappears across models. In contrast to model-specific dimensions, universal dimensions are more interpretable and more strongly driven by conceptual image properties, indicating the relevance of interpretability and semantic content as implicit factors driving universality across models. Differences in architecture, objective function, training data, model size, and model performance do not explain the emergence of universal dimensions. However, models with more universal dimensions also better predict macaque IT activity and human similarity judgments, suggesting that universality reflects representations relevant to biological vision. These findings have important implications for understanding the emergent representations underlying deep neural network models and their alignment with biological vision.

The Segment Anything Model 2 (SAM2) has emerged as a foundation model for universal segmentation. Owing to its generalizable visual representations, SAM2 has been successfully applied to various downstream tasks. However, extending SAM2 to the RGB-D video salient object detection (RGB-D VSOD) task encounters three challenges including limited spatial modeling of linear LoRA, insufficient employment of SAM's multi-scale features, and dependence of initialization on explicit prompts. To address the issues, we present Multi-Modal Mixture-of-Experts with Memory-Augmented SAM (M$^4$-SAM), which equips SAM2 with modality-related PEFT, hierarchical feature fusion, and prompt-free memory initialization. Firstly, we inject Modality-Aware MoE-LORA, which employs convolutional experts to encode local spatial priors and introduces a modality dispatcher for efficient multi-modal fine-tuning, into SAM2's encoder. Secondly, we deploy Gated Multi-Level Feature Fusion, which hierarchically aggregates multi-scale encoder features with an adaptive gating mechanism, to balance spatial details and semantic context. Finally, to conduct zero-shot VSOD without manual prompts, we utilize a Pseudo-Guided Initialization, where a coarse mask is regarded as a pseudo prior and used to bootstrap the memory bank. Extensive experiments demonstrate that M$^4$-SAM achieves the state-of-the-art performance across all evaluation metrics on three public RGB-D VSOD datasets.

This work explores a simple yet powerful lightweight adapter design for feed-forward 3D Gaussian Splatting (3DGS). Existing methods typically apply complex, architecture-specific designs on top of the generic pipeline of image feature extraction $\rightarrow$ multi-view interaction $\rightarrow$ feature decoding. However, constrained by the scale bottleneck of 3D training data and the low-pass filtering effect of deep networks, these methods still fall short in cross-domain generalization and high-frequency geometric fidelity. To address these problems, we propose AdaptSplat, which demonstrates that without complex component engineering, introducing a single adapter of only 1.5M parameters into the generic architecture is sufficient to achieve superior performance. Specifically, we design a lightweight Frequency-Preserving Adapter (FPA) that extracts direction-aware high-frequency structural priors from the shallow features of a powerful vision foundation model backbone, and seamlessly integrates them into the generic pipeline via high-frequency positional encodings and adaptive residual modulation. This effectively compensates for the high-frequency attenuation caused by over-smoothing in deep features, improving the fitting accuracy of Gaussian primitives on complex surfaces and sharp boundaries. Extensive experiments demonstrate that AdaptSplat achieves state-of-the-art feed-forward reconstruction performance on multiple standard benchmarks, with stable generalization across domains. Code available at: https://github.com/xmw666/AdaptSplat.

We develop a foundation model using 1.2m high resolution satellite images of the Netherlands. By combining a Convolutional Neural Network and a Vision Transformer, the model captures both low- and high-frequency landscape features, such as fine textures, edges, and small objects as well as large terrain structures, elevation patterns, and land-cover distributions. Leveraging temporal data as input, the model learns from broader contextual information across time, allowing the model to exploit the temporal dependencies, such as topographic features, land-cover changes, and seasonal dynamics. These additional constraints reduce feature ambiguity, improve representation learning, and enable better generalization with fewer labeled samples. The foundation model is evaluated on multiple downstream tasks, ranging from use cases within the Netherlands to global benchmarking datasets. On the vegetation monitoring dataset of the Netherlands, the model shows clear performance improvements by incorporating temporal information instead of relying on a single time point. Despite using a smaller model and less pretraining data limited to the Netherlands, it achieves competitive results on global benchmarks when compared to state-of-the-art models. These results demonstrate that the model can learn rich, generalizable representations from limited data, achieving competitive performance on global benchmarks while using a fraction of the parameters of larger state-of-the-art remote sensing models. To maximize reproducibility and reuse, we made the scripts and the model accessible on GitHub.

Artificial intelligence models are increasingly scaled to improve predictive accuracy, yet it remains unclear whether scale improves the quality of post-hoc explanations. We investigate this relationship by evaluating 11 computer vision models representing increasing levels of depth and complexity within the ResNet, DenseNet, and Vision Transformer families, trained from scratch or pretrained, across three image datasets with ground-truth segmentation masks. For each model, we generate explanations using five post-hoc explainable AI methods and quantify mask alignment using two localisation metrics: Relevance Rank Accuracy (Arras et al., 2022) and the proposed Dual-Polarity Precision, which measures positive attributions inside the class mask and negative attributions outside it. Across datasets and methods, increasing architectural depth and parameter count does not improve explanation quality in most statistical comparisons, and smaller models often match or exceed deeper variants. While pretraining typically improves predictive performance and increases the dependence of explanations on learned weights, it does not consistently increase localisation scores. We also observe scenarios in which models achieve strong predictive performance while localisation precision is near zero, suggesting that performance metrics alone may not indicate whether predictions are based on the annotated regions. These results indicate that larger models do not reliably provide higher-quality explanations, and that explainability should therefore be assessed explicitly during model selection for safety-sensitive deployments.

Reliable displacement measurement is fundamental for structural health monitoring and digital engineering workflows, as it provides direct structural response information. Vision-based measurement has emerged as a promising approach for low-cost, non-contact displacement monitoring. However, its deployment often remains constrained by task-specific model training or on-site preparation, such as marker installation or manual camera calibration. This study presents a Vision Foundation Model-based framework for Structural Displacement Measurement (VFM-SDM) that integrates VFM-inferred camera parameter estimation and point tracking to reconstruct multi-directional structural displacements via triangulation without task-specific training or on-site preparation, enabling efficient non-contact deployment in real-world applications. Structural geometry constraints are incorporated to suppress physically implausible deviations and improve estimation consistency. A multi-modal field dataset collected from an in-service pedestrian bridge is introduced alongside a unified benchmarking protocol to support reproducible evaluation. Representative results show low amplitude errors (NRMSE$_{\text{range}}$: 0.11/0.12), strong temporal agreement (correlation coefficient: 0.86/0.88), and small peak-to-peak amplitude errors (RPPAE: 0.01/0.02) for vertical and lateral displacements, indicating robust performance under real-world conditions. The proposed framework advances automated, scalable displacement monitoring and lays the groundwork for VFM-enabled structural response measurements in digital twin and data-centric construction workflows.

Steering or intervening on model representations at inference time to correct predictions is essential for AI interpretability and safety, yet existing evaluation protocols are limited to ambiguous language modeling tasks. To address this gap, we introduce SwordBench, a benchmark for steering image representations of vision models across multiple backbones and concept removal tasks. Beyond a unified benchmarking suite, we propose new evaluation notions that uncover the second-order effects of orthogonalization among concept activation vectors for pragmatic steering. Specifically, cross-concept robustness measures the stability of concept detection performance across inputs orthogonalized against alternative concepts, and collateral damage quantifies whether steering inadvertently affects model performance on a downstream task for inputs lacking the bias. We find that although a linear support vector machine exhibits superior separability and orthogonality, it fails to achieve zero collateral damage, often trailing sparse autoencoders. In simpler regimes, both standard baselines and optimization-based methods fail to achieve perfect steering. The source code will be made available soon on GitHub.

Deep vision models degrade sharply in low-data regimes, particularly in medical imaging where labeled samples are scarce. We show this arises not merely from overfitting but from a geometric failure: finite-sample noise corrupts the embedding covariance, collapsing the eigengap and limiting the number of recoverable signal-bearing modes. We develop a spectral theory of finite-sample representation learning that quantifies the recoverable dimension K(N), the number of eigenmodes that can be stably estimated from N samples. Using perturbation theory and concentration bounds, we show that only modes with eigenvalues above the noise floor $\|\hatΣ - Σ\|_{\mathrm{op}} \sim \sqrt{D/N}$ are reliable, yielding a truncated Mahalanobis energy that governs classification performance. Under a power-law spectral model, this energy can be approximated by a truncated Riemann zeta function, linking eigenvalue decay to data efficiency and AUC. Within this framework, multimodal learning acts as spectral stabilization: vision-language models impose low-rank constraints that suppress noise-dominated directions and preserve the eigengap, increasing K(N) under data scarcity. Across MNIST and multi-disease neuroimaging, we show that multimodal training maintains more stable modes and improves class separation, even when unimodal models achieve comparable few-shot accuracy. These results identify spectral collapse as a fundamental bottleneck in low-data learning. We use truncated Mahalanobis energy and K(N) to diagnose encoder quality, and introduce zeta-based spectral filtering as a principled approach to improve data efficiency.

Spatial intelligence in vision-language models (VLMs) attracts research interest with the practical demand to reason in the 3D world.Despite promising results, most existing methods follow the conventional 2D pipeline in VLMs and use pixel-aligned representations for the vision modality. However, correspondence-based models with implicit 3D scene understanding often fail to achieve spatial consistency, and representation-based models with 3D geometric priors lack efficiency in vision sequence serialization. To address this, we propose a Proxy3D method with compact yet comprehensive 3D proxy representations for the vision modality. Given only video frames as input, we employ semantic and geometric encoders to extract scene features and then perform their semantic-aware clustering to obtain a set of proxies in the 3D space. For representation alignment, we further curate the SpaceSpan dataset and apply multi-stage training to adopt the proposed 3D proxy representations with the VLM. When using shorter sequences for vision information, our method achieves competitive or state-of-the-art performance in 3D visual question answering, visual grounding and general spatial intelligence benchmarks.

Underwater salient object detection (USOD) plays a vital role in marine vision tasks but remains fundamentally challenging due to severe visual degradation, such as selective absorption and medium scattering. Conventional pipelines typically adopt a sequential "enhance-then-detect" paradigm. However, isolating low-level visual restoration from high-level semantic perception often leads to semantic inconsistency, where the restored images may not be optimal for detection and can even introduce task-irrelevant noise. To break this sequential bottleneck, we propose UniV2D, a Unified Vision-to-Detection Network that jointly optimizes visual restoration and salient object detection within a mutually beneficial framework. Unlike traditional methods that rely on disjointed pipelines or rigid physical priors, UniV2D introduces a semantic-driven learning paradigm: high-level saliency semantics actively guide the restoration process, while the restored visual cues reciprocally enhance saliency perception. Specifically, UniV2D features a hierarchical dual-branch architecture. It first employs a self-calibrated decoder to predict initial saliency masks alongside a mask-aware restoration module to reconstruct image content. Subsequently, a saliency-guided refinement module equipped with cross-level modulation is utilized to align structural fidelity with semantic consistency. Extensive experiments across multiple benchmarks demonstrate that UniV2D significantly outperforms state-of-the-art methods in both quantitative and qualitative evaluations, establishing a new standard for joint underwater perception.

Concept-based explanations offer a promising approach for explaining the predictions of deep neural networks in terms of high-level, human-understandable concepts. However, existing methods either do not establish a causal connection between the concepts and model predictions or are limited in expressivity and only able to infer causal explanations involving single concepts. At the same time, the parallel line of work on formal abductive and contrastive explanations computes the minimal set of input features causally relevant for model outcomes but only considers low-level features such as pixels. Merging these two threads, in this work, we propose the notion of concept-based abductive and contrastive explanations that capture the minimal sets of high-level concepts causally relevant for model outcomes. We then present a family of algorithms that enumerate all minimal explanations while using concept erasure procedures to establish causal relationships. By appropriately aggregating such explanations, we are not only able to understand model predictions on individual images but also on collections of images where the model exhibits a user-specified, common behavior. We evaluate our approach on multiple models, datasets, and behaviors, and demonstrate its effectiveness in computing helpful, user-friendly explanations.

Part-based reasoning is a classical strategy to make a computer vision model directly focus on the object parts that are relevant to the downstream task. In the context of deep learning, this also serves to improve by-design interpretability, often by using part-centric attention mechanisms on top of a latent image representation provided by a standard, black-box model. This approach is based on a locality assumption: that the latent representation of an object part encodes primarily information about the corresponding image region. In this work, we test this basic assumption, measuring intra-object leakage in vision models using part-based attribute annotations. Through a comprehensive experimental evaluation, we show that modern pretrained vision transformers violate the locality assumption and exhibit a strong intra-object leakage, in which each part encodes information from the whole object, a visual metonymy that compromises the faithfulness of attention-based interpretable-by-design methods for part-based reasoning, ultimately rendering them uninterpretable. In addition, we establish an upper bound using a two-stage approach that prevents leakage by design. We then show that this inherently disentangled feature extraction improves attribute-driven part discovery on a variety of tasks, confirming the practical impact of intra-object leakage. Our results uncover a neglected issue affecting the interpretability of part-based representations, such as those in CBMs relying on part-centric concepts, highlighting that two-stage approaches offer a promising way to mitigate it.

The condition monitoring (CM) of synthetic fibre ropes (SFRs) used in offshore, maritime, and industrial settings demands more than a classifier: inspectors need continuous severity estimates, maintenance recommendations, anomaly flags, deterioration timelines, and automated reports, all from a single inspection image. We present DART (Damage Assessment via Rope Transformer), a vision-language foundation model that addresses the full rope inspection workflow through a unified multi-task architecture. DART extends the Joint-Embedding Predictive Architecture (JEPA) to the cross-modal domain by coupling a Vision Transformer (ViT-H/14) with Llama-3.2-3B-Instruct via a Severity-Conditioned Cross-Modal Fusion (SC-CMF) module. Three architectural innovations drive the model's versatility: (1) HD-MASK, a saliency-guided masking strategy that focuses self-supervised reconstruction on damage-dense patches; (2) per-class learnable severity gates that adaptively weight language grounding by damage category; and (3) a Contrastive Damage Disentanglement (CDD) loss that shapes the embedding space to simultaneously encode damage type, severity ordering, and cross-modal semantics. Trained once on 4,270 images spanning 14 fine-grained rope damage classes, the frozen DART backbone supports downstream tasks without any task-specific fine-tuning: damage classification (93.22 % accuracy, 91.04 % macro-F1, +38.5 pp over a vision-only baseline), continuous severity regression (Spearman rho = 0.94, within-1-ordinal accuracy 99.6 %), few-shot recognition (89.2 % macro-F1 at 20 shots). These results demonstrate that DART functions as a general-purpose CM backbone that goes well beyond classification, providing actionable inspection intelligence from a single shared representation.

Recent studies have shown that large generative models can solve vision tasks they were not explicitly trained for. However, existing evidence relies on closed-source models~(Veo~3, Nano Banana Pro) or requires task-specific instruction tuning, leaving open whether publicly available image-editing models possess zero-shot vision abilities out of the box. We conduct a systematic evaluation of three open-source image-editing models -- Qwen-Image-Edit, FireRed-Image-Edit, and LongCat-Image-Edit -- on dense visual prediction tasks \emph{without any fine-tuning}. We benchmark monocular depth estimation on NYUv2 and DIODE, surface normal estimation on NYUv2, and semantic segmentation on Cityscapes, covering both geometric and semantic scene understanding. Results show that open-source image-editing models exhibit non-trivial zero-shot visual understanding. On NYUv2 surface normals, FireRed-Image-Edit achieves a mean angular error of $17.69^\circ$, surpassing the fine-tuned Marigold ($20.86^\circ$) and matching the instruction-tuned Vision Banana ($17.78^\circ$) without any task-specific training. On NYUv2 depth estimation, LongCat-Image-Edit obtains $δ_1{=}0.822$ with affine alignment, and Qwen-Image-Edit leads on DIODE Indoor ($δ_1{=}0.868$). On Cityscapes semantic segmentation, Qwen-Image-Edit reaches 25.7 mIoU at the 19-class level and 49.5 mIoU at a coarser 7-category level. By comparing three independently trained editors, we test whether zero-shot vision ability is an emergent property of image-editing pretraining rather than a model-specific artifact. Code, evaluation scripts, and all results are publicly released to serve as a reproducible baseline for future work.

Solar photovoltaic (PV) deployment is expanding rapidly, yet detailed, up-to-date information on the spatial distribution and capacity of rooftop PV remains limited. This paper presents an open, scalable framework for detecting solar panels from open data and generating city-level solar power profiles. We leverage foundation vision AI models to detect solar panel geometries from open-source satellite imagery. This avoids manual data labeling and case-specific model training while maintaining robustness across heterogeneous imagery. Detected solar panels are converted into georeferenced polygons, yielding spatially explicit and incrementally extensible inventories. By integrating open weather data, we translate panel footprints into regional solar power profiles. The framework reduces dependency on proprietary imagery, manual labeling, and closed-source models, and offers a transparent and scalable approach for solar planning and analysis. We released the data and an API resulted from this work. For any user-specified building location, our API retrieves aerial imagery, detects rooftop solar panels, and returns georeferenced polygons. This empowers researchers and developers to scan user-defined areas to build solar panel maps and associated solar production profiles, thus facilitating advanced analysis like distributed solar production integration, local power flow optimization, energy tariff design, and infrastructure planning.

The advent of foundation models has heralded a new era in medical artificial intelligence (AI), enabling the extraction of generalizable representations from large-scale unlabeled datasets. However, current ophthalmic AI paradigms are predominantly constrained to single-modality inference, thereby creating a dissonance with clinical practice where diagnosis relies on the synthesis of complementary imaging modalities. Furthermore, the deployment of high-performance AI in resource-limited settings is frequently impeded by the unavailability of advanced three-dimensional imaging hardware. Here, we present the Ophthalmic multimodal Masked Autoencoder (OphMAE), a multi-imaging foundation model engineered to synergize the volumetric depth of 3D Optical Coherence Tomography (OCT) with the planar context of 2D en face OCT. By implementing a novel cross-modal fusion architecture and a unique adaptive inference mechanism, OphMAE was pre-trained on a massive dataset with of 183,875 paired OCT images derived from 32,765 patients. In a rigorous benchmark encompassing 17 diverse diagnostic tasks with 48,340 paired OCT images from 8,191 patients, the model demonstrated state-of-the-art performance, achieving an Area Under the Curve (AUC) of 96.9% for Age-related Macular Degeneration (AMD) and 97.2% for Diabetic Macular Edema (DME), consistently surpassing existing single-modal and multimodal foundation models. Crucially, OphMAE exhibits robust engineering adaptability: it maintains high diagnostic accuracy, such as 93.7% AUC for AMD, even when restricted to single-modality 2D inputs, and demonstrates exceptional data efficiency by retaining 95.7% AUC with as few as 500 labeled samples. This work establishes a scalable and adaptable framework for ophthalmic AI, ensuring robust performance across different tasks.

Cross-view Referring Multi-Object Tracking (CRMOT) aims to track multiple objects specified by natural language across multiple camera views, with globally consistent identities. Despite recent progress, existing methods rely heavily on costly frame-level spatial annotations and cross-view identity supervision. To reduce such reliance, we explore CRMOT under weak supervision by leveraging the capabilities of foundation models. However, our empirical study shows that directly applying foundation models such as SAM2 and SAM3, even with task-specific modifications, fails to accurately understand referring expressions and maintain consistent identities across views. Yet, they remain effective at producing reliable object tracklets that can serve as pseudo supervision. We therefore repurpose foundation models as pseudo-label generators and propose a two-stage framework for weakly supervised CRMOT, using only object category labels as coarse-grained supervision. In the first stage, we design an Affinity-guided Cross-view Re-prompting strategy to refine and associate SAM3-generated tracklets across cameras, producing reliable cross-view pseudo labels for subsequent training. In the second stage, we introduce ViewSAM, a CRMOT model built upon SAM2 that explicitly models view-aware cross-modal semantics. By formulating view-induced variations as learnable conditions, ViewSAM bridges the gap between view-variant visual observations and view-invariant textual expressions, enabling robust cross-view referring tracking with only approximately 10% additional parameters. Extensive experiments demonstrate that ViewSAM achieves SOTA performance under weak supervision and remains competitive with fully supervised methods.

Salient Object Detection (SOD) remains an essential yet underexplored task in the era of large-scale vision models. Although foundation models like SAM exhibit strong generalization, their potential for SOD is not fully realized, and training or fully fine-tuning them is computationally expensive and prone to overfitting under limited data. To overcome these challenges, we introduce GLASSNet, a Global-Local feature decoding framework that uses SAMv2 as a frozen encoder paired with a lightweight, spatially aware convolutional adapter-reducing learnable encoder parameters by over 97%. To enhance saliency quality, GLASSNet employs a dual-decoder architecture: one decoder captures global, long-range semantics with an expanded receptive field, while the other captures fine local details such as edges and textures. Fusing these complementary cues yields saliency maps that combine global coherence with local precision, producing accurate final masks. Extensive experiments on standard SOD and camouflaged object detection benchmarks show that GLASSNet surpasses state-of-the-art methods, demonstrating the power of frozen foundation models combined with targeted adaptation and global-local decoding.

Vision foundation models have attracted significant attention for their ability to leverage large-scale unlabeled visual data. This advantage is particularly important in remote sensing, where data acquisition is costly and annotation often requires expert knowledge. Recent electro-optical vision foundation models aim to learn domain-specific representations from remote sensing imagery, but it remains unclear whether they are more effective than strong generalist vision foundation models under retrieval-based evaluation. In this study, we conduct a controlled comparison between representative EO-specific and generalist vision foundation models for remote sensing image retrieval. Using the same datasets, retrieval protocol, and evaluation metric, we evaluate both in-domain performance and cross-scene generalization. Our results show that strong generalist vision foundation models are competitive with, and in some cases outperform, existing EO-specific models. Moreover, EO-specific models often suffer from substantial degradation under cross-scene evaluation, while generalist models show more stable transfer. These findings suggest that EO pretraining alone does not guarantee stronger retrieval-oriented remote sensing representations. We discuss the limitations of current EO-specific pretraining strategies and highlight the need for future EO vision foundation models to better exploit the physical, spatial, spectral, and geographic characteristics of remote sensing imagery.

Vision Foundation Models (VFMs) pretrained on large-scale RGB data have demonstrated remarkable representation quality, yet their applicability to multispectral imaging spanning Near-Infrared (NIR), Short-Wave Infrared (SWIR), and Long-Wave Infrared (LWIR) remains largely unexplored. These spectral modalities offer complementary sensing capabilities critical for robust perception in adverse conditions, but present a fundamental domain gap relative to RGB-centric pretrained models. We present SpectraDINO, a multispectral VFM that bridges this spectral gap by extending DINOv2 ViT backbones to beyond-visible modalities through lightweight, per-modality bottleneck adapters, while preserving the rich representations of the frozen RGB backbone. We introduce a multi-stage teacher-student training protocol in which a frozen DINOv2 teacher guides a spectral student via cosine distillation, symmetric contrastive loss, patch-level alignment, and a novel neighborhood-structure-preservation loss. This staged curriculum enables strong cross-modal alignment without catastrophic forgetting of RGB priors. We evaluate SpectraDINO on multispectral object detection and semantic segmentation across challenging NIR, SWIR, and LWIR benchmarks using widely adopted fusion strategies. SpectraDINO achieves state-of-the-art performance across most benchmarks, validating its effectiveness as a general-purpose backbone for spectral generalization. The code and weights for model variants are available at https://github.com/Yonsei-STL/SpectraDINO.

Scene-level point cloud self-supervised learning (PC-SSL) has demonstrated potential in enhancing the generalization capability of 3D vision models. Despite the advances in the field through existing methods, the sample-independent modeling paradigm still poses significant limitations in terms of maintaining consistent semantic representations across scenes. This challenge hinders the construction of a unified and transferable semantic space. To address this issue, we propose a PC-SSL framework based on cross-sample semantic propagation (CSP), in which samples within a batch are serialized into continuous input and processed by a state-space model to enable semantic state propagation. This mechanism explicitly models the dynamic dependencies across samples in the state space, allowing the network to establish cross-sample semantic consistency in the latent space and achieve global semantic alignment. Since serialization-based pretraining requires batch-level input organization, we further introduce an asymmetric semantic preservation distillation (SPD) during finetuning to achieve structural alignment of semantic transfer and eliminate inconsistencies caused by batch dependency. The proposed SPD ensures stable transfer of pretrained semantics through a heterogeneous input mechanism and a semantic feature alignment constraint. This enables the model to maintain structured semantic consistency and robustness under single-scene testing conditions. Extensive experiments on multiple benchmark datasets demonstrate that our method consistently outperforms state-of-the-art methods in both performance and semantic consistency.

Modern vision models achieve remarkable accuracy, but explaining where evidence arises, what the model encodes, and how internal computations assemble that evidence remains fragmented. We introduce an iERF-centric framework that unifies local, global, and mechanistic interpretability around a single analysis unit: the pointwise feature vector (PFV) paired with its instance-specific Effective Receptive Field (iERF). On the local side, Sharing Ratio Decomposition (SRD) expresses each PFV as a mixture of upstream PFVs via sharing ratios and propagates iERFs to construct class-discriminative saliency maps. SRD yields high-resolution, activation-faithful explanations, is robust to targeted manipulation and noise, and remains activation-agnostic across common nonlinearities. For the global view, we introduce Concept-Anchored Feature Explanation (CAFE), which utilizes the iERF as a semantic label, grounding abstract latent vectors in verifiable pixel-level evidence. With CAFE, we address the challenge of non-localized sparse autoencoder latents--especially in Transformers, where early self-attention mixes distant context. To answer how representations are composed through depth, we propose the Interlayer Concept Graph with Interlayer Concept Attribution (ICAT), which quantifies concept-to-concept influence while isolating layer pairs; an interlayer insertion, deletion protocol identifies Integrated Gradients as the most faithful instantiation. Empirically, across ResNet50, VGG16, and ViTs, our framework outperforms baselines in both fidelity and robustness, successfully interprets dispersed SAE features, and exposes dominant concept routes in correct, misclassified, and adversarial cases. Grounded in iERFs, our approach provides a coherent, evidence-backed map from pixels to concepts to decisions.

The demand for high-resolution subsurface imaging and continuous Earth monitoring has driven rapid growth in active and passive seismic data from dense geophone deployments, distributed acoustic sensing (DAS) arrays, and large-scale 2D and 3D surveys. This expansion makes complex noise suppression increasingly challenging, especially when signal fidelity must be preserved. Conventional supervised deep learning methods are often task-specific, require large paired datasets, and can suffer from domain shift under new acquisition conditions. Foundation models offer a promising alternative, but pre-training seismic foundation models from scratch requires massive domain-specific data and substantial computation. We propose an efficient framework that repurposes general-purpose Vision Foundation Models (VFMs) for geophysical tasks through Parameter-Efficient Fine-Tuning. The architecture uses a pre-trained VFM, a DINOv3 encoder, adapted with Low-Rank Adaptation (LoRA) to enable effective feature adaptation with few additional parameters. To improve robustness under unseen field conditions without ground truth, we introduce a kurtosis-guided unsupervised test-time adaptation module that updates only LoRA parameters during inference. This module self-calibrates the model to site-specific noise by identifying information-rich regions via kurtosis and performing self-training without labeled data. Experiments on public exploration seismic images and DAS vertical seismic profiling data from the Utah FORGE site show that the framework matches or outperforms domain-specific models. Tests on unseen cross-site data from a land survey in China and the Groß Schönebeck geothermal site in Germany further demonstrate strong generalization and effective signal-noise separation. These results highlight the potential of adapting pre-trained VFMs to data-intensive problems in exploration seismology.

AI-generated images are becoming increasingly realistic and diverse, posing significant challenges for generalizable detection. While Vision Foundation Models (VFMs) provide rich semantic representations and frequency-based methods capture complementary artifact cues, existing approaches that combine these modalities still suffer from limited generalization, with notable performance degradation on unseen generative models. We attribute this limitation to two key factors: frequency shortcut bias toward easily distinguishable cues associated with specific generators and cross-domain representation conflict between high-level semantics and low-level frequency patterns. To address these issues, we propose a Frequency-aware Gated Injection Network (FGINet) to improve generalization. Specifically, we design a Band-Masked Frequency Encoder (BMFE) that applies cross-band masking in the frequency domain to reduce reliance on generator-specific patterns and encourage more diverse and generalizable representations. We further introduce a Layer-wise Gated Frequency Injection (LGFI) mechanism to progressively inject frequency cues into the VFM backbone with adaptive gating, aligning with its hierarchical abstraction and alleviating representation conflict. Moreover, we propose a Hyperspherical Compactness Learning (HCL) framework with a cosine margin objective to learn compact and well-separated representations. Extensive experiments demonstrate that FGINet achieves state-of-the-art performance and strong generalization across multiple challenging datasets.

Detectors often suffer from degraded performance, primarily due to the distributional gap between the source and target domains. This issue is especially evident in single-source domains with limited data, as models tend to rely on confounders (e.g., illumination, co-occurrence, and style) from the source domain, leading to spurious correlations that hinder generalization. To this end, this paper proposes a novel Basis-driven framework for domain generalization, namely \textbf{\textit{Bridge}}, that incorporates causal inference into object detection. By learning the low-rank bases for front-door adjustment, \textbf{\textit{Bridge}} blocks confounders' effects to mitigate spurious correlations, while simultaneously refining representations by filtering redundant and task-irrelevant components. \textbf{\textit{Bridge}} can be seamlessly integrated with both discriminative (e.g., DINOv2/3, SAM) and generative (e.g., Stable Diffusion) Vision Foundation Models (VFMs). Extensive experiments across multiple domain generalization object detection datasets, i.e., Cross-Camera, Adverse Weather, Real-to-Artistic, Diverse Weather Datasets, and Diverse Weather DroneVehicle (our newly augmented real-world UAV-based benchmark), underscore the superiority of our proposed method over previous state-of-the-art approaches. The project page is available at: https://mingbohong.github.io/Bridge/.

Recent methods demonstrate that large-scale pretrained models, such as CLIP vision transformers, effectively detect AI-generated images (AIGIs) from unseen generative models when used as feature extractors. Many state-of-the-art methods for AI-generated image detection build upon the original CLIP-ViT to enhance this generalization. Since CLIP's release, numerous vision foundation models (VFMs) have emerged, incorporating architectural improvements and different training paradigms. Despite these advances, their potential for AIGI detection and AI image forensics remains largely unexplored. In this work, we present a comprehensive benchmark across multiple VFM families, covering diverse pretraining objectives, input resolutions, and model scales. We systematically evaluate their out-of-the-box performance for detecting fully-generated AI-images and AI-inpainted images, and discover that the best model outperforms the original CLIP by more than 12% in accuracy, beating established approaches in the process. To fully leverage the features of a modern VFM, we propose a simple redesign of the classifier head by utilizing tunable attention pooling (TAP), which aggregates output tokens into a refined global representation. Integrating TAP with the latest VFMs yields substantial performance gains across several AIGI detection benchmarks, establishing a new state-of-the-art on two challenging benchmarks for in-the-wild detection of AI-generated and -inpainted images.

Monocular depth estimation (MDE) is a fundamental yet inherently ill-posed task. Recent vision foundation models (VFMs), particularly DINO-based transformers, have significantly improved accuracy and generalization for dense prediction. Prior works generally follow a unified paradigm: sampling a fixed set of intermediate transformer layers at uniform intervals to build multi-scale features. This common practice implicitly assumes that geometric information is uniformly distributed across layers, which may underutilize the structural 3D cues encoded in VFMs. In this study, we present a systematic layer-wise analysis of DINOv3, revealing that 3D information is distributed non-uniformly: deeper layers exhibit stronger depth predictability and better capture inter-sample geometric variation. Motivated by this, we introduce a Last-Layer-Centric Feature Recombination (LFR) module to enhance geometric expressiveness. LFR treats the final layer as a geometric anchor and adaptively selects complementary intermediate layers according to a minimal-similarity criterion. Selected features are fused with the last-layer representation via compact linear adapters.Extensive experiments show that LFR module consistently improves MDE accuracy and achieves state-of-the-art performance. Our analysis sheds light on how geometric knowledge is organized within VFMs and offers an efficient strategy for unlocking their potential in dense 3D tasks.

Industrial object detection systems typically rely on large annotated datasets, which are expensive to collect and challenging to maintain in industrial scenarios where the inventory of objects changes frequently. This work addresses the challenge of few-shot object detection in such industrial scenarios, where only a limited number of labeled samples are available for newly introduced objects. We present a detection framework that leverages vision foundation models to recognize objects with minimal supervision. The method constructs class prototypes from a small set of reference samples by extracting feature representations. For a given query scene during inference, object regions are generated using a segmentation model, and feature embeddings are extracted and matched with class prototypes using similarity matching. We evaluate the detection method on three established industrial datasets from the Benchmark for 6D Object Pose Estimation benchmark following the official 2D object detection evaluation protocol. We demonstrate competitive detection performance, improving AP by 6.9% compared to the state-of-the-art training-free detection methods. Furthermore, the presented method is able to onboard new objects using only a few reference images, without requiring any CAD models or large annotated datasets. These properties make the approach well-suited for real-world industrial applications.

To usher in the next round of client AI innovation, there is an urgent need to enable efficient, lossless inference of high-accuracy large language models (LLMs) and vision language models (VLMs), jointly referred to as xLMs, on client systems. To address this, we present pipelined sharding, a novel, benchmark-profile-guided CPU-GPU hybrid scheduling technique to achieve efficient, VRAM-constrained inference for both dense and mixture-of-experts (MoE) LLMs. Using a combination of model sharding at the sub-layer level, CPU offloading, pipelined copy-compute, and prioritized tensor placement in VRAM, it optimizes both time-to-first-token (TTFT) and tokens per second (TPS) metrics, while flexibly adapting to system and inference conditions. For efficient, high-accuracy VLM inference, we combine pipelined sharding with a llama$.$cpp implementation of three well-understood prior ideas (jointly called VLMOpt), namely, vision tensor CPU offloading, flash attention, and vision and language model VRAM overlap avoidance. These enhancements are targeted at improving client xLM inference in future releases of two important NVIDIA products - the In-Game Inferencing software development kit (IGI SDK) and the Cosmos-Reason1 (CR1) physical AI reasoning VLM. Highlights from our rigorous evaluation spanning multiple models and client systems include: for interactive use, TTFT improves by up to 6.7x and TPS by up to 30x for LLMs, and CR1 inference's VRAM demand is down by 10x, while in batched mode, throughput improves by up to 8.2x, all compared to their respective aggressive baselines. This paper is accepted at the 9th MLSys Conference (Industry Track), 2026. Code and artifact available at: https://github.com/deepshnv/pipeshard-mlsys26-ae

While preference optimization is crucial for improving visual generative models, how to effectively scale this paradigm remains largely unexplored. Current open-source preference datasets contain conflicting preference patterns, where winners excel in some dimensions but underperform in others. Naively optimizing on such noisy datasets fails to learn preferences, hindering effective scaling. To enhance robustness against noise, we propose Poly-DPO, which extends the DPO objective with an additional polynomial term that dynamically adjusts model confidence based on dataset characteristics, enabling effective learning across diverse data distributions. Beyond biased patterns, existing datasets suffer from low resolution, limited prompt diversity, and imbalanced distributions. To facilitate large-scale visual preference optimization by tackling data bottlenecks, we construct ViPO, a massive-scale preference dataset with 1M image pairs at 1024px across five categories and 300K video pairs at 720p+ across three categories. State-of-the-art generative models and diverse prompts ensure reliable preference signals with balanced distributions. Remarkably, when applying Poly-DPO to our high-quality dataset, the optimal configuration converges to standard DPO. This convergence validates dataset quality and Poly-DPO's adaptive nature: sophisticated optimization becomes unnecessary with sufficient data quality, yet remains valuable for imperfect datasets. We validate our approach across visual generation models. On noisy datasets like Pick-a-Pic V2, Poly-DPO achieves 6.87 and 2.32 gains over Diffusion-DPO on GenEval for SD1.5 and SDXL, respectively. For ViPO, models achieve performance far exceeding those trained on existing open-source preference datasets. These results confirm that addressing both algorithmic adaptability and data quality is essential for scaling visual preference optimization.

Chest computed tomography (CT) is central to the detection and management of thoracic disease, yet the growing scale and complexity of volumetric imaging increasingly exceed what can be addressed by scan-level prediction alone. Clinically useful AI for CT must not only recognize disease across the whole volume, but also localize abnormalities and provide interpretable visual evidence. Existing vision-language foundation models typically compress scans and reports into global image-text representations, limiting their ability to preserve spatial evidence and support clinically meaningful interpretation. Here we developed EXACT, an explainable anomaly-aware foundation model for three-dimensional chest CT that learns spatially resolved representations from paired clinical scans and radiology reports. EXACT was pre-trained on 25,692 CT-reports pairs using anatomy-aware weak supervision, jointly learning organ segmentation and multi-instance anomaly localization without manual voxel-level annotations. The resulting organ-specific anomaly-aware maps assign each voxel a disease-specific anomaly score confined to its corresponding anatomy, jointly encoding lesion extent and organ-level context. In retrospective multinational and multi-center evaluations, EXACT showed broad and consistent improvements across clinically relevant CT tasks, spanning multi-disease diagnosis, zero-shot anomaly localization, downstream adaptation, and visually grounded report generation, outperforming existing three-dimensional medical foundation models. By transforming routine clinical CT scans and free-text reports into explainable voxel-level representations, EXACT establishes a scalable paradigm for trustworthy volumetric medical AI.

To humans, a robin seems more like a bird than a bird seems like a robin, but does this asymmetry also hold for machine vision? Humans and modern vision models can match each other in accuracy while making systematically different kinds of errors, differing not in how often they fail, but in who gets mistaken for whom. We show these directional confusions reveal distinct inductive biases invisible to accuracy alone. Using matched human and deep neural network responses on a natural-image categorization task under 12 perturbation types, we quantify asymmetry in confusion matrices and link its organization to the geometry of the information--error trade-off - how efficiently, and how gracefully, a system generalizes under distortion. We find that humans exhibit broad but weak asymmetries across many class pairs, whereas deep vision models show sparser, stronger directional collapses into a few dominant categories. Robustness training reduces overall asymmetry magnitude but fails to recover this human-like distributed structure. Generative simulations further show that these two asymmetry organizations shift the trade-off geometry in opposite directions even at matched accuracy, explaining why the same scalar asymmetry score can reflect fundamentally different generalization strategies. Together, these results establish directional confusion structure as a sensitive, interpretable signature of inductive bias that accuracy-based evaluation cannot recover.

Real-world weather, illumination, and imaging variations often induce severe domain shifts, degrading single-source detectors in unseen environments. Existing single-domain generalized object detection (SDGOD) methods mainly rely on data augmentation or domain-invariant learning, while largely overlooking how domain shift disrupts detector prediction stability. Through analytical experiments, we find that performance degradation is mainly dominated by increasing missed detections. Further analysis shows that this phenomenon stems from reduced cross-domain stability in DETR-style detectors: domain shift disrupts encoder-side object-background and inter-instance relations, and further weakens the semantic-spatial binding between decoder queries and real objects. Motivated by this, we find that vision foundation models (VFMs) still preserve stable relational structures and object responses under severe shifts, making them suitable cross-domain stability priors to compensate for detector degradation. To this end, we propose VFM$^{4}$SDG, a dual-prior learning framework for SDGOD, which introduces a frozen VFM into encoder representation learning and decoder query modeling. Specifically, we propose Cross-domain Stable Relational Prior Distillation to distill stable object-background and inter-instance relations from the VFM into the encoder, compensating for relational degradation. Meanwhile, we propose Semantic-Contextual Prior-based Query Enhancement, which injects category semantic prototypes and global object context into queries before they enter the decoder layer, enhancing semantic-spatial query-object binding stability. Extensive experiments show that VFM$^{4}$SDG significantly outperforms existing advanced methods on standard SDGOD benchmarks and two mainstream DETR-based detection frameworks, demonstrating its effectiveness, robustness, and generality.

Vision Graph Neural Networks (ViGs) represent an image as a graph of patch tokens, enabling adaptive, feature-driven neighborhoods. Unlike CNNs with fixed grid biases or Vision Transformers with global token interactions, ViGs rely on dynamic graph convolution: at each layer, a feature-dependent graph is built via k-nearest-neighbor (kNN) search on current patch features, followed by message passing. This per-layer graph construction is the main bottleneck, consuming 50--95% of graph convolution time on CPUs and GPUs, scaling as $O(N^2)$ with the number of patches $N$, and creating a sequential dependency between graph construction and feature updates. We introduce GraphLeap, a simple reformulation that removes this dependency by decoupling graph construction from feature update across layers. GraphLeap performs the feature update at layer $\ell$ using a graph built from the previous layer's features, while simultaneously using the current layer's features to construct the graph for layer $\ell+1$. This one-layer-lookahead graph construction enables concurrent graph construction and message passing. Although using prior-layer features can introduce minor accuracy degradation, lightweight fine-tuning for a few epochs is sufficient to recover the original accuracy. Building on GraphLeap, we present the first end-to-end FPGA accelerator for Vision GNNs. Our streaming, layer-pipelined design overlaps a kNN graph construction engine with a feature update engine, exploits node- and channel-level parallelism, and enables efficient on-chip dataflow without explicit edge-feature materialization. Evaluated on isotropic and pyramidal ViG models on an Alveo U280 FPGA, GraphLeap achieves up to $95.7\times$ speedup over CPU and $8.5\times$ speedup over GPU baselines, demonstrating the feasibility of real-time Vision GNN inference.

Vision Mamba, as a state space model (SSM), employs a zero-order hold (ZOH) discretization, which assumes that input signals remain constant between sampling instants. This assumption degrades temporal fidelity in dynamic visual environments and constrains the attainable accuracy of modern SSM-based vision models. In this paper, we present a systematic and controlled comparison of six discretization schemes instantiated within the Vision Mamba framework: ZOH, first-order hold (FOH), bilinear/Tustin transform (BIL), polynomial interpolation (POL), higher-order hold (HOH), and the fourth-order Runge-Kutta method (RK4). We evaluate each method on standard visual benchmarks to quantify its influence in image classification, semantic segmentation, and object detection. Our results demonstrate that POL and HOH yield the largest gains in accuracy at the cost of higher training-time computation. In contrast, the BIL provides consistent improvements over ZOH with modest additional overhead, offering the most favorable trade-off between precision and efficiency. These findings elucidate the pivotal role of discretization in SSM-based vision architectures and furnish empirically grounded justification for adopting BIL as the default discretization baseline for state-of-the-art SSM models.

While Vision-Language Models (VLMs) have advanced high-level reasoning in autonomous driving, their ability to ground this reasoning in the underlying physics of ego-motion remains poorly understood. We introduce EgoDyn-Bench [Project page: (https://tum-avs.github.io/EgoDyn-Bench-Website/), Code: (https://github.com/TUM-AVS/EgoDyn-Bench), Dataset: (https://huggingface.co/datasets/fnc1901/EgoDyn-Bench)], a diagnostic benchmark for evaluating the semantic ego-motion understanding of vision-centric foundation models. By mapping continuous vehicle kinematics to discrete motion concepts via a deterministic oracle, we decouple a model's internal physical logic from its visual perception. Our large-scale empirical audit spanning 20$+$ models, including closed-source MLLMs, open-source VLMs across multiple scales, and specialized VLAs, identifies a significant Perception Bottleneck: while models exhibit logical physical concepts, they consistently fail to accurately align them with visual observations, frequently underperforming classical non-learned geometric baselines. This failure persists across model scales and domain-specific training, indicating a structural deficit in how current architectures couple visual perception with physical reasoning. We demonstrate that providing explicit trajectory encodings substantially restores physical consistency across all evaluated models, revealing a functional disentanglement between vision and language: ego-motion logic is derived almost exclusively from the language modality, while visual observations contribute negligible temporal signal. This structural finding provides a standardized diagnostic framework and a practical pathway toward physically aligned embodied AI. Ego-motion - Physical Reasoning - Foundation Models

We propose Semantic-Fast-SAM (SFS), a semantic segmentation framework that combines the Fast Segment Anything model with a semantic labeling pipeline to achieve real-time performance without sacrificing accuracy. FastSAM is an efficient CNN-based re-implementation of the Segment Anything Model (SAM) that runs much faster than the original transformer-based SAM. Building upon FastSAM's rapid mask generation, we integrate a Semantic-Segment-Anything (SSA) labeling strategy to assign meaningful categories to each mask. The resulting SFS model produces high-quality semantic segmentation maps at a fraction of the computational cost and memory footprint of the original SAM-based approach. Experiments on Cityscapes and ADE20K benchmarks demonstrate that SFS matches the accuracy of prior SAM-based methods (mIoU ~ 70.33 on Cityscapes and 48.01 on ADE20K) while achieving approximately 20x faster inference than SSA in the closed-set setting. We also show that SFS effectively handles open-vocabulary segmentation by leveraging CLIP-based semantic heads, outperforming recent open-vocabulary models on broad class labeling. This work enables practical real-time semantic segmentation with the "segment-anything" capability, broadening the applicability of foundation segmentation models in robotics scenarios. The implementation is available at https://github.com/KBH00/Semantic-Fast-SAM.

Face Anti-Spoofing (FAS) remains challenging due to the requirement for robust domain generalization across unseen environments. While recent trends leverage Vision-Language Models (VLMs) for semantic supervision, these multimodal approaches often demand prohibitive computational resources and exhibit high inference latency. Furthermore, their efficacy is inherently limited by the quality of the underlying visual features. This paper revisits the potential of vision-only foundation models to establish a highly efficient and robust baseline for FAS. We conduct a systematic benchmarking of 15 pre-trained models, such as supervised CNNs, supervised ViTs, and self-supervised ViTs, under severe cross-domain scenarios including the MICO and Limited Source Domains (LSD) protocols. Our comprehensive analysis reveals that self-supervised vision models, particularly DINOv2 with Registers, significantly suppress attention artifacts and capture critical, fine-grained spoofing cues. Combined with Face Anti-Spoofing Data Augmentation (FAS-Aug), Patch-wise Data Augmentation (PDA) and Attention-weighted Patch Loss (APL), our proposed vision-only baseline achieves state-of-the-art performance in the MICO protocol. This baseline outperforms existing methods under the data-constrained LSD protocol while maintaining superior computational efficiency. This work provides a definitive vision-only baseline for FAS, demonstrating that optimized self-supervised vision transformers can serve as a backbone for both vision-only and future multimodal FAS systems. The project page is available at: https://gsisaoki.github.io/FAS-VFMbenchmark-CVPRW2026/ .

Human perception generalizes well across different domains, but most vision models struggle beyond their training data. This gap motivates multi-dataset learning, where a single model is trained on diverse datasets to improve robustness under domain shifts. However, unified training remains challenging due to inconsistencies in data distributions and label semantics. Mixture-of-Experts (MoE) models provide a scalable solution by routing inputs to specialized subnetworks (experts). Yet, existing MoEs often fail to specialize effectively, as their load-balancing mechanisms enforce uniform input distribution across experts. This fairness conflicts with domain-aware routing, causing experts to learn redundant representations, and reducing performance especially on rare or out-of-distribution domains. We propose GEM (Global Expert Mapping), a planner-compiler framework that replaces the learned router with a global scheduler. Our planner, based on linear programming relaxation, computes a fractional assignment of datasets to experts, while the compiler applies hierarchical rounding to convert this soft plan into a deterministic, capacity-aware mapping. Unlike prior MoEs, GEM avoids balancing loss, resolves the conflict between fairness and specialization, and produces interpretable routing. Experiments show that GEM-DINO achieves state-of-the-art performance on the UODB benchmark, with notable gains on underrepresented datasets and solves task interference in few-shot adaptation scenarios.

Frame-wise semantic segmentation of indoor lidar scans is a fundamental step toward higher-level 3D scene understanding and mapping applications. However, acquiring frame-wise ground truth for training deep learning models is costly and time-consuming. This challenge is largely addressed, for imagery, by Visual Foundation Models (VFMs) which segment image frames. The same VFMs may be used to train a lidar scan frame segmentation model via a 2D-to-3D distillation pipeline. The success of such distillation has been shown for autonomous driving scenes, but not yet for indoor scenes. Here, we study the feasibility of repeating this success for indoor scenes, in a frame-wise distillation manner by coupling each lidar scan with a VFM-processed camera image. The evaluation is done using indoor SLAM datasets, where pseudo-labels are used for downstream evaluation. Also, a small manually annotated lidar dataset is provided for validation, as there are no other lidar frame-wise indoor datasets with semantics. Results show that the distilled model achieves up to 56% mIoU under pseudo-label evaluation and around 36% mIoU with real-label, demonstrating the feasibility of cross-modal distillation for indoor lidar semantic segmentation without manual annotations.

In this work, we introduce DifFoundMAD, a parameter-efficient D-MAD framework that exploits the generalisation capabilities of vision foundation models (FM) to capture discrepancies between suspected morphs and live capture images. In contrast to conventional D-MAD systems that rely on face recognition embeddings or handcrafted feature differences, DifFoundMAD follows the standard differential paradigm while replacing the underlying representation space with embeddings extracted from FMs. By combining lightweight finetuning with class-balanced optimisation, the proposed method updates only a small subset of parameters while preserving the rich representational priors of the underlying FMs. Extensive cross-database evaluations on standard D-MAD benchmarks demonstrate that DifFoundMAD achieves consistent improvements over state-of-the-art systems, particularly at the strict security levels required in operational deployments such as border control: The error rates reported in the current state-of-the-art were reduced from 6.16% to 2.17% for high-security levels using DifFoundMAD.

Automated polyp segmentation is critical for early colorectal cancer detection and its prevention, yet remains challenging due to weak boundaries, large appearance variations, and limited annotated data. Lightweight segmentation models such as U-Net, U-Net++, and PraNet offer practical efficiency for clinical deployment but struggle to capture the rich semantic and structural cues required for accurate delineation of complex polyp regions. In contrast, large Vision Foundation Models (VFMs), including SAM, OneFormer, Mask2Former, and DINOv2, exhibit strong generalization but transfer poorly to polyp segmentation due to domain mismatch, insufficient boundary sensitivity, and high computational cost. To bridge this gap, we propose \textit{\textbf{LiteBounD}, a \underline{Li}gh\underline{t}w\underline{e}ight \underline{Boun}dary-guided \underline{D}istillation} framework that transfers complementary semantic and structural priors from multiple VFMs into compact segmentation backbones. LiteBounD introduces (i) a dual-path distillation mechanism that disentangles semantic and boundary-aware representations, (ii) a frequency-aware alignment strategy that supervises low-frequency global semantics and high-frequency boundary details separately, and (iii) a boundary-aware decoder that fuses multi-scale encoder features with distilled semantically rich boundary information for precise segmentation. Extensive experiments on both seen (Kvasir-SEG, CVC-ClinicDB) and unseen (ColonDB, CVC-300, ETIS) datasets demonstrate that LiteBounD consistently outperforms its lightweight baselines by a significant margin and achieves performance competitive with state-of-the-art methods, while maintaining the efficiency required for real-time clinical use. Our code is available at https://github.com/lostinrepo/LiteBounD.

This paper presents webmcu-vision-web, a single-file, zero-install browser application for end-to-end TinyML vision model training and deployment on the Seeed Studio XIAO ESP32-S3 Sense (XIAO ML Kit, $15--40 USD). Acting as a browser-based companion to the on-device Arduino firmware of Paper 1 [1], it provides a private, fully local machine learning pipeline, from firmware flashing through image collection, CNN training, weight export, and live activation visualization, without any software installation beyond a Chromium-based browser. The system targets educators, small businesses, and researchers who need to train task-specific visual classifiers under their exact deployment conditions. Key capabilities include: in-browser firmware flashing via esptool-js; an SD card file browser with image preview and inline editing; config.json live-sync for zero-recompile hyperparameter adjustment; webcam and ESP32 OV2640 camera image capture; TensorFlow.js CNN training completing a three-class run (~30 images per class, 20 epochs) in approximately 1 minute browser-side versus 9 minutes on-device, enabling a complete collect-train-deploy cycle in under 10 minutes; weight export as myWeights.bin and myWeights.h; confusion matrix; and a live Conv2 activation heatmap streamed from the ESP32 during inference. No data leaves the local machine at any stage. A five-run consistency evaluation on the three-class reference problem (0Blank, 1Cup, 2Pen) demonstrates stable convergence with mean accuracy and standard deviation reported; all artefacts are released at the repository link below. The repository is a living template for LLM-assisted adaptation to new hardware and tasks. All source code is MIT-licensed at https://github.com/webmcu-ai/webmcu-vision-web.

This report presents the NTIRE 2026 Rip Current Detection and Segmentation (RipDetSeg) Challenge, which targets automatic rip current understanding in images. Rip currents are hazardous nearshore flows that cause many beach-related fatalities worldwide, yet remain difficult to identify because their visual appearance varies substantially across beaches, viewpoints, and sea states. To advance research on this safety-critical problem, the challenge builds on the RipVIS benchmark, evaluating both detection and segmentation. The dataset is diverse, sourced from more than $10$ countries, with $4$ camera orientations and diverse beach and sea conditions. This report describes the dataset, challenge protocol, evaluation methodology, final results, and summarizes the main insights from the submitted methods. The challenge attracted $159$ registered participants and produced $9$ valid test submissions across the two tasks. Final rankings are based on a composite score that combines $F_1[50]$, $F_2[50]$, $F_1[40\!:\!95]$, and $F_2[40\!:\!95]$. Most participant solutions relied on pretrained models, combined with strong augmentation and post-processing design. These results suggest that rip current understanding benefits strongly from the robust general-purpose vision models' progress, while leaving ample room for future methods tailored to their unique visual structure.

With the rapid development of generative models and multimodal content editing technologies, the key challenge faced by synthetic image detection (SID) lies in cross-distribution generalization to unknown generation sources. In recent years, visual foundation models (VFM), which acquire rich visual priors through large scale image-text alignment pretraining, have become a promising technical route for improving the generalization ability of SID. However, existing VFM-based methods remain relatively coarse-grained in their adaptation strategies. They typically either directly use the final layer representations of VFM or simply fuse multi layer features, lacking explicit modeling of the optimal representational hierarchy for transferable forgery cues. Meanwhile, although directly fine-tuning VFM can enhance task adaptation, it may also damage the cross-modal pretrained structure that supports open-set generalization. To address this task specific tension, we reformulate VFM adaptation for SID as a joint optimization problem: it is necessary both to identify the critical representational layer that is more suitable for carrying forgery discriminative information and to constrain the disturbance caused by task knowledge injection to the pretrained structure. Based on this, we propose I2P, an SID framework centered on intrinsic importance perception. I2P first adaptively identifies the critical layer representations that are most discriminative for SID, and then constrains task-driven parameter updates within a low sensitivity parameter subspace, thereby improving task specificity while preserving the transferable structure of pretrained representations as much as possible.

Language-aligned vision foundation models (VFMs) enable versatile visual understanding for always-on contextual AI, but their deployment on edge devices is hindered by strict latency and power constraints. We present AdaVFM, an adaptive framework for efficient on-device inference of language-aligned VFMs that dynamically adjusts computation based on scene context and task complexity. Our key insight is that the effect of model size reduction on performance is task-dependent in vision applications, motivating a runtime-adaptive execution strategy. AdaVFM integrates neural architecture search (NAS) into the language-aligned VFM backbone to enable lightweight subnet execution during runtime. A multimodal large language model (LLM) deployed on the cloud enables runtime control with a context-aware agent. This synergy allows efficient model adaptation under diverse conditions while maintaining strong accuracy. Extensive experiments on zero-shot classification and open-vocabulary segmentation demonstrate that AdaVFM achieves state-of-the-art accuracy-efficiency trade-offs, surpassing prior baselines by up to $7.9\%$ in acc@1 on IN1K and $5.2\%$ mIoU on ADE20K over the best models of comparable VFM sizes. For models with similar accuracy, AdaVFM further reduces average FLOPs by up to $77.9\%$.

A vision foundation model outputs an embedding vector for an image, which can be affected by common editing operations (e.g., JPEG compression, brightness, contrast adjustments). These common perturbations alter embedding vectors and may impact the performance of downstream tasks using these embeddings. In this work, we present the first systematic study on foundation models' robustness to such perturbations. We propose three robustness metrics and formulate five desired mathematical properties for these metrics, analyzing which properties they satisfy or violate. Using these metrics, we evaluate six industry-scale foundation models (OpenAI, Meta) across nine common perturbation categories, finding them generally non-robust. We also show that common perturbations degrade downstream application performance (e.g., classification accuracy) and that robustness values can predict performance impacts. Finally, we propose a fine-tuning approach to improve robustness without sacrificing utility.

Retail theft costs the global economy over $100 billion annually, yet existing AI-based detection systems require expensive custom model training on proprietary datasets and charge $200-500/month per store. We present Paza, a zero-shot retail theft detection framework that achieves practical concealment detection without training any model. Our approach orchestrates multiple existing models in a layered pipeline - cheap object detection and pose estimation running continuously, with an expensive vision-language model (VLM) invoked only when behavioral pre-filters trigger. A multi-signal suspicion pre-filter (requiring dwell time plus at least one behavioral signal) reduces VLM invocations by 240x compared to per-frame analysis, bounding calls to <=10/minute and enabling a single GPU to serve 10-20 stores. The architecture is model-agnostic: the VLM component accepts any OpenAI-compatible endpoint, enabling operators to swap between models such as Gemma 4, Qwen3.5-Omni, GPT-4o, or future releases without code changes - ensuring the system improves as the VLM landscape evolves. We evaluate the VLM component on the DCSASS synthesized shoplifting dataset (169 clips, controlled environment), achieving 89.5% precision and 92.8% specificity at 59.3% recall zero-shot - where the recall gap is attributable to sparse frame sampling in offline evaluation rather than VLM reasoning failures, as precision and specificity are the operationally critical metrics determining false alarm rates. We present a detailed cost model showing viability at $50-100/month per store (3-10x cheaper than commercial alternatives), and introduce a privacy-preserving design that obfuscates faces in the detection pipeline. The source code is available at https://github.com/xHaileab/Paza-AI.

Over the recent years, the field of robotics has been undergoing a transformative paradigm shift from fixed, single-task, domain-specific solutions towards adaptive, multi-function, general-purpose agents, capable of operating in complex, open-world, and dynamic environments. This tremendous advancement is primarily driven by the emergence of Foundation Models (FMs), i.e., large-scale neural-network architectures trained on massive, heterogeneous datasets that provide unprecedented capabilities in multi-modal understanding and reasoning, long-horizon planning, and cross-embodiment generalization. In this context, the current study provides a holistic, systematic, and in-depth review of the research landscape of FMs in robotics. In particular, the evolution of the field is initially delineated through five distinct research phases, spanning from the early incorporation of Natural Language Processing (NLP) and Computer Vision (CV) models to the current frontier of multi-sensory generalization and real-world deployment. Subsequently, a highly-granular taxonomic investigation of the literature is performed, examining the following key aspects: a) the employed FM types, including LLMs, VFMs, VLMs, and VLAs, b) the underlying neural-network architectures, c) the adopted learning paradigms, d) the different learning stages of knowledge incorporation, e) the major robotic tasks, and f) the main real-world application domains. For each aspect, comparative analysis and critical insights are provided. Moreover, a report on the publicly available datasets used for model training and evaluation across the considered robotic tasks is included. Furthermore, a hierarchical discussion on the current open challenges and promising future research directions in the field is incorporated.

Visual Geometry Foundation Models (VGFMs) demonstrate remarkable zero-shot capabilities in local reconstruction. However, deploying them for kilometer-level Simultaneous Localization and Mapping (SLAM) remains challenging. In such scenarios, current approaches mainly rely on linear transforms (e.g., Sim3 and SL4) for sub-map alignment, while we argue that a single linear transform is fundamentally insufficient to model the complex, non-linear geometric distortions inherent in VGFM outputs. Forcing such rigid alignment leads to the rapid accumulation of uncorrected residuals, eventually resulting in significant trajectory drift and map divergence. To address these limitations, we present CAL2M (Calibration-free Assistant-eye based Large-scale Localization and Mapping), a plug-and-play framework compatible with arbitrary VGFMs. Distinct from traditional systems, CAL2M introduces an "assistant eye" solely to leverage the prior of constant physical spacing, effectively eliminating scale ambiguity without any temporal or spatial pre-calibration. Furthermore, leveraging the assumption of accurate feature matching, we propose an epipolar-guided intrinsic and pose correction model. Supported by an online intrinsic search module, it can effectively rectify rotation and translation errors caused by inaccurate intrinsics through fundamental matrix decomposition. Finally, to ensure accurate mapping, we introduce a globally consistent mapping strategy based on anchor propagation. By constructing and fusing anchors across the trajectory, we establish a direct local-to-global mapping relationship. This enables the application of nonlinear transformations to elastically align sub-maps, effectively eliminating geometric misalignments and ensuring a globally consistent reconstruction. The source code of CAL2M will be publicly available at https://github.com/IRMVLab/CALM.