Vision Foundation Models

Domain adaptation remains a central challenge in 3D vision, especially for multimodal foundation models that align 3D point clouds with visual and textual data. While these models demonstrate strong general capabilities, adapting them to downstream domains with limited data often leads to overfitting and catastrophic forgetting. To address this, we introduce ReFine3D, a regularized fine-tuning framework designed for domain-generalizable tuning of 3D large multimodal models (LMMs). ReFine3D combines selective layer tuning with two targeted regularization strategies: multi-view consistency across augmented point clouds and text diversity through synonym-based prompts generated by large language models. Additionally, we incorporate point-rendered vision supervision and a test-time augmentation mechanism with confidence-based aggregation to further enhance robustness. Extensive experiments across different 3D domain generalization benchmarks show that ReFine3D improves base-to-novel class generalization by 1.36%, cross-dataset transfer by 2.43%, robustness to corruption by 1.80%, and few-shot accuracy by up to 3.11%, outperforming prior state-of-the-art methods with minimal added computational overhead.

Accurate vision-based navigation in monocular endoscopy is difficult due to limited depth cues, weak tissue texture, non-rigid deformation, and substantial appearance variation across domains, all of which complicate pose estimation, depth prediction, and image-to-anatomy alignment. Although recent vision foundation models have shown promise, their learned representations often remain insufficiently geometry-consistent, hindering stable feature correspondence and limiting their reliability for downstream navigation tasks. We propose a unified framework for learning geometry-consistent and domain-robust image representations for monocular endoscopy. The framework combines a synthetic data pipeline that provides accurate geometric supervision with Hierarchy-Aware Geometry-Semantic Adaptation, a structured alternative to standard LoRA that inserts low-rank adapters selectively across the transformer hierarchy and couples them with layer-wise training objectives to encourage geometric correspondence in intermediate features and semantic consistency in deeper features. Experiments on public and proprietary datasets show improved geometric and semantic representation quality, leading to better performance on downstream navigation tasks including pose estimation and monocular depth estimation. The learned representations show favorable synthetic-to-real transfer on clinical bronchoscopy and provide a useful initialization for adaptation to sinus endoscopy and colonoscopy under limited supervision. The framework also shows favorable scaling with model size and training data. These results support hierarchy-aware, geometry-guided adaptation as a practical approach for endoscopic representation learning.

Vision Navigation Foundation Models (VNMs) promise end-to-end learned navigation policies capable of zero-shot deployment across diverse embodiments and environments. To maintain generality, many vision-based navigation models predict normalized actions. However, this normalization introduces a critical deployment vulnerability: applying different scaling factors to the same normalized trajectory alters its physical geometry, which degrades navigation performance and increases collision risks. We address this vulnerability by conditioning the model on normalized action histories alongside image observations, providing explicit context on the relationship between the model's predictions and the robot's actual physical displacement. Furthermore, current VNMs often struggle in visually repetitive environments that lack distinct features. To resolve this issue, we integrate a DINOv3 encoder, whose richer representations enable our model to capture both spatial and geometric dimensions between observations. VISTA generalizes robustly to out-of-distribution environments, achieving 100% goal prediction accuracy in zero-shot, real-world deployment in Outdoor, Forest and Office settings, and an average of 95% checkpoints crossed, demonstrating consistent path following in unseen environments.

JPEG, a lossy image compression technique designed for human viewers, has maintained its dominance for decades. However, in the era of artificial intelligence (AI), a substantial portion of image data, often compressed by JPEG, is and will continue to be consumed by deep neural networks (DNNs) instead of humans, thus creating a need to optimize JPEG for DNN inference performance. To this end, we propose learned JPEG compression for DNN vision (J4D), a novel training framework for determining JPEG encoding parameters to minimize compression rate while maximizing DNN inference performance. The major challenge of solving this optimization problem lies in representing the JPEG codec and compression rate in closed form. By incorporating a differentiable soft quantizer based on a probabilistic quantization scheme, we not only obtain a differentiable proxy for the JPEG codec, but are also able to compute the entropy of the coded source analytically, which is a close estimate of the actual compression rate. Equipped with both the differentiable JPEG codec and the information-theoretic rate estimator, we are then able to solve the aforementioned optimization problem with backpropagation. After training, the learned encoding parameters will be subsequently used in actual JPEG encoding based on probabilistic quantization. Extensive experimental results across multiple datasets and DNN architectures demonstrate that J4D consistently and significantly outperforms the default JPEG and other competitive JPEG codecs optimized for DNNs. Notably, compared to the default JPEG, J4D achieves an increase in accuracy by as much as 11.60% at the same rate, or a reduction of compression rate up to 80.05% at the same accuracy. Additionally, with the help of J4D, we show the potential to design universal JPEG encoding parameters for various DNN architectures for the first time.

Vision foundation models (VFMs) have demonstrated strong robustness and transferability across a wide range of visual tasks. However, each model typically encodes strong inductive biases shaped by its pre-training objective and data domain, resulting in fragmented yet complementary visual knowledge. As a result, a single model often struggles to capture the diverse visual representations required across multiple dense prediction tasks. To address this limitation, we propose TIGER (Task-Instruction-Guided Expert Routing), a framework that coordinates multiple heterogeneous VFMs for multi-task dense prediction. Instead of naively aggregating expert features, TIGER leverages natural-language task instructions to guide a routing network that assigns token-level expert weights conditioned on task semantics, enabling adaptive integration of complementary expert features. TIGER further introduces a counterfactual loss that aligns routing decisions with each expert's causal contribution by measuring prediction changes when experts are excluded, encouraging more reliable and interpretable routing. We evaluate TIGER on two multi-task dense prediction benchmarks, NYUD-v2 and Pascal Context, where it consistently outperforms recent multi-task learning baselines while keeping all VFMs frozen. These results demonstrate that combining instruction-guided expert routing with counterfactual causal alignment enables effective coordination of heterogeneous vision foundation models.

Vision models can achieve strong performance on classification tasks, but the internal representations supporting their predictions are often difficult to interpret. This work investigates whether sparse autoencoders can decompose intermediate representations of a vision model into interpretable features. We train a ConvNeXt classifier on the FGVC-Aircraft dataset, extract spatial activations from its final feature stage, and train a sparse autoencoder on these activations. The learned sparse features are analyzed using top-activating image patches, activation strength, and class selectivity. Qualitative visual inspection reveals that several features correspond to recognizable aircraft structures and visual patterns. We evaluate a subset of selected features using input-space and feature-space ablations, measuring how blurring image patches and suppressing sparse features affect class logits, classification margins, and prediction confidence. The results suggest that sparse autoencoders can reveal partially interpretable, class-relevant visual features associated with aircraft recognition, while also exposing limitations such as polysemanticity and coarse spatial localization.

Color fundus photography (CFP) is the mainstay of large-scale retinal screening, but its diagnostic capacity is limited by the lack of depth-resolved structure, which optical coherence tomography (OCT) provides yet is less accessible at population scale. We present EyeMVP, a cross-modal retinal foundation model that uses paired CFP--OCT pretraining to learn OCT-informed CFP representations while requiring only CFP at inference. Pretrained on 674,893 same-eye same-day CFP--OCT triples from 112,642 patients across eight hospitals, EyeMVP uses cross-modal masked reconstruction to enrich CFP features with OCT-associated supervision, and combines source-constrained cross-attention with CFP-derived structural masks to accommodate the non-aligned geometry of en-face CFP and cross-sectional OCT. Across 15 dataset-level settings spanning classification and segmentation, under both full-data and few-shot regimes, EyeMVP performs on par with or better than representative retinal foundation models, with consistent gains on macular and optic-nerve tasks; it attains AUROCs of 0.923 for macular edema and 0.867 for myopic macular schisis, two conditions poorly resolved in CFP. In an exploratory reader study, EyeMVP surpasses junior and intermediate ophthalmologists but not seniors on macular edema, while exceeding all groups on myopic macular schisis. These results indicate that cross-modal reconstruction can enrich CFP representations with OCT-associated supervision, offering a practical route to stronger CFP-based screening.

Self-supervised geospatial foundation models (GeoFMs) learn transferable representations from remote sensing data, but their downstream behavior is difficult to characterize. We study six representative GeoFMs spanning joint-embedding, reconstruction, and multimodal pretraining families, and evaluate transfer across classification, regression, and segmentation benchmarks under different label availability and downstream pipelines. We find that model rankings change across tasks and adaptation settings. Layerwise probing shows that, in most cases, task-relevant information is more accessible in intermediate transformer blocks compared to final-layer embeddings, and that GeoFMs exhibit distinct depthwise profiles. In segmentation case studies on PASTIS and Sen1Floods11, downstream adaptation settings such as decoder design and fine-tuning can be as impactful as the choice of GeoFM, and standard dense-prediction heads may be poorly aligned with how GeoFMs organize information over depth. Finally, CKA analysis on case studies shows that fine-tuning does not rewrite GeoFMs uniformly across depth, and the strongest changes are localized to the first linear layer of the MLP in ViT blocks. These results help explain why GeoFM rankings shift across benchmarks and motivate more representation-aware evaluation and adaptation strategies.

Vision-language models (VLMs) can answer image-based questions confidently, and often correctly, even when no image is provided. This mirage behavior inflates benchmark scores without reflecting visual grounding. Prior work treats this as a single failure mode. We argue it is two. Using Mirage Probes, a contrastive probing framework that pairs paraphrased question variants with matched mirage and non-mirage labels on the same image, we show that mirage behavior is linearly decodable from internal activations across residual stream, MLP, post-attention, and attention-head sites in two open-source VLMs. We demonstrate that a Naive Bayes text baseline cannot recover this signal, ruling out surface lexical confounds. Cross-benchmark separability patterns, together with a novel Prior Harnessing Index (PHI) measuring how much a model can answer from text alone, expose two distinct regimes: textual biases, where the model answers from language priors without engaging visual representations, and spurious images, where it constructs false visual content in latent space and answers as if grounded. The distinction has direct mitigation consequences: text-distribution cleaning can address the first regime but cannot reach the second, since spurious-image mirages live in the model's visual representations rather than its text. Faithful visual grounding will require interventions at the representational level.

Intersection-over-Union (IoU), as a pivotal metric for evaluating the spatial alignment between candidate proposals and ground-truth annotations, directly determines the quality of positive sample sets and the training efficacy of visual detection models. Through theoretical modeling and analysis, we uncover a non-sensitive region on the IoU response curve, within which samples yield nearly identical IoU scores despite distinct geometric overlaps. To overcome this limitation, we introduce a set of morphological similarity metrics covering area, shape, and aspect ratio, to refine the positive sample assignment process, thereby ensuring more discriminative and reliable matching. A supplementary matching score is derived via mean-based aggregation of these multidimensional similarities, compensating for the intrinsic limitation of IoU in representing structural correspondence. Theoretically, incorporating morphological similarity reshapes the response distribution of the matching function, yielding both effective directional gradients and polygon-like iso-response contours, which tightly confine high-response regions around each ground-truth instance and substantially enhance the precision of positive sample selection. Experiments based on the YOLOv9 framework demonstrate consistent performance gains on both NEUDET and GC10- DET datasets. Notably, the proposed approach is fully plug-and-play and incurs zero additional inference overhead, thereby ensuring deployment efficiency for industrial visual inspection.

Built on pretrained vision foundation models (VFMs), representation autoencoders (RAEs) have recently emerged as a promising approach for constructing semantically rich latent spaces for image generation. However, their reconstruction quality often remains suboptimal, largely because deep VFM representations do not preserve sufficient fine-grained visual detail. This limitation becomes even more severe after discretization, where missing low-level information is difficult to recover. In fact, we observe that shallow VFM features retain considerably richer local appearance and structural detail, which complements the high-level semantics carried by deep features used in existing RAEs. Motivated by this complementary property, we propose Ideal, an In-depth Alignment framework for discrete representation autoencoding. By jointly aligning quantized tokens with both shallow and deep VFM features, Ideal enables the resulting discrete visual tokens to preserve both visual fidelity and rich semantics. Extensive experiments demonstrate that Ideal yields superior reconstruction performance, achieving 0.61 rFID on ImageNet and outperforming the previous best method by 0.28. When used for autoregressive image generation, Ideal further produces a gFID of 1.89, establishing a new state of the art for autoregressive image generation.

Visual in-Context Learning (VICL) aims at making progress towards adaptive vision models, that can -- based on a few examples -- adapt to a new task at test-time. With the history of in-context learning in natural language processing research, where large, parameter-heavy models are in use, one pathway that current VICL methods take is model- and data-scaling as key ingredients. Yet, it is not clear, whether these ingredients are the key for in-context learning to take shape in vision models. To stress-test such large models, we challenge them with an extreme counterexample: we train a tiny visual in-context model with merely $1$ million parameters and a modest amount of $70,000$ images. We compare the results of this severely capacity capped tiny model to $7,000\times$ larger VICL models in different adaptive settings, (1) on image data with small distribution shifts, (2) on unseen task encodings and (3) on a completely new task, i.e., the setting VICL envisions. With the chasm of training resources between the tiny- and large models, our experiments showcase a lack in how adaptive capabilities are measured, with respect to how tasks are encoded, which tasks were used in pre-training and the choice of metrics. These gaps in current VICL benchmarking underscore a need for innovation in evaluation of adaptive capabilities.

Multi-task vehicle routing problems play a critical role in enhancing efficiency across various industries and service sectors. These problems consist of multiple variants that optimize routing costs while meeting diverse customer constraints. Existing multi-task VRP solvers solely utilize a graph-based modality, limiting their ability to address variants with multiple constraints. As a format to represent complex semantics, vision modality shows great potential for encoding diverse VRP constraints. This motivates us to learn patch-level semantics from the vision images, and then integrate them into a graph-based model to solve various VRP variants simultaneously. However, directly applying this approach to multi-task VRPs presents three challenges: 1) existing VRP images lack constraint representations, which are essential for multi-task VRPs, 2) the fixed receptive field of individual patches cannot effectively accommodate varying requirements across tasks, and 3) imbalanced pixel distribution among constraints may cause the model to overlook constraints with fewer pixels. In this paper, we propose a vision-assisted foundation model (VaFM) to address these challenges. In the vision modality, input images tailored to all constraints are encoded by a convolutional neural network. The obtained patch embeddings are fused with graph-based nodes to generate solutions, with an auxiliary task designed to address the pixel-imbalanced issue. The performance of VaFM is evaluated across 16 different VRP variants. The experimental results demonstrate the superiority of VaFM over state-of-the-art methods, especially for variants with complex constraints.

Generalized Few-Shot Semantic Segmentation (GFSS) has traditionally been approached as a representation-learning problem, requiring task-specific adaptation to incorporate novel classes from limited support examples. Recent foundation models, however, already exhibit strong open-vocabulary recognition and segmentation capabilities, raising a different question: can GFSS be solved through inference-time coordination of frozen semantic priors rather than parameter adaptation? We answer this question with Open-V, a training-free GFSS framework that combines Segment Anything (SAM3) Promptable Concept Segmentation (PCS) with a K-shot CLIP support centroid through calibrated per-pixel semantic arbitration. OpenV introduces no trainable components and supports arbitrary semantic categories at inference time. Beyond segmentation performance, our study contributes three broader findings. First, we show that support information can be incorporated through inference-time semantic grounding, and that its contribution increases as foundation-model text priors weaken on label-disjoint vocabularies. Second, we identify a reproducibility confound in foundationmodel segmentation, demonstrating that preprocessing and evaluation-space mismatches can silently distort reported performance. Finally, we validate Open-V across PASCAL5i, COCO-20i, and ADE-OW, showing that training-free coordination of foundation-model priors generalizes across both conventional GFSS and open-vocabulary evaluation settings. On PASCAL-5i (1-shot), Open-V attains base/novel/harmonic mIoU of 78.4/77.5/77.9, without GFSS-specific training surpassing the strongest trained baseline by +17.7 HM.

Remote-sensing foundation models (RSFMs) benefit from pretraining on imagery from multiple sensors and ground sampling distances (GSDs), but such exposure alone does not resolve scale mismatch during downstream adaptation. A fixed token-grid offset can correspond to different ground distances across sensors, making grid-based positional priors physically inconsistent. Meanwhile, heterogeneous spatial granularity means that compact urban regions and homogeneous landscapes may require different positional sensitivities even under the same GSD. Therefore, we propose {GeoRoPE}, a ground-aware, RoPE-compatible, and parameter-efficient spatial adaptation method for RSFMs. GeoRoPE recalibrates token-level positional interactions from two complementary aspects. First, \textit{Geo-Coordinate Calibration (GCC)} rescales raw token-grid offsets according to the ground distance represented by one token-grid step, producing geo-calibrated relative coordinates across GSDs. Second, \textit{Geo-Frequency Calibration (GFC)} adjusts the native RoPE frequency with a relation-specific factor, enabling position sensitive adaptation to scene-dependent spatial granularity. GeoRoPE is injected into pretrained RSFMs through a lightweight adapter, preserving the frozen spatial prior while adding geo-aware positional corrections. Experiments across multiple RSFMs, sensors, resolutions, and downstream tasks demonstrate that GeoRoPE improves cross-resolution robustness and scale-sensitive representation learning.

Different vision neural networks -- trained to classify, contrast, reconstruct, or match images to text -- should have correspondingly different internal representations. We report that they do not. After training, the top sixteen principal directions of variation inside thirteen modern vision encoders converge to the same sixteen-dimensional geometric object. We call this the cross-architecture substrate and study it with PCA, centred kernel alignment (CKA), and Pang 2026 calibration. The substrate transports across four visual domains (natural photographs, medical CT, satellite, microscopy) at median Procrustes-CKA 0.679, and across eight domains (adding sketches, depth, thermal infrared, astronomy) at 0.604, every pair >0.40. It survives Pang calibration globally (7.4x disc-vs-MAE separation, n=13,394) and locally (4.82-5.30, p<10^{-44}). It is not pixel statistics (0.263), not Gabor features (0.31), not a random projection (0.041), and emerges in the first 10% of training while accuracy keeps climbing. We deliver four applications: a label-free transferability filter beating LogME (3x faster, +0.15 Kendall-tau); a four-way domain detector (99.6% accuracy); a frozen low-shot probe (16 dims beat 768-dim DINOv2 by 3.78pp at N=50 labels per class); and a teacher-free distillation auxiliary matching trained-teacher KD on 33 pairs (7.56pp peak gain at 10% label fraction). The substrate does not cross modalities, does not help cross-paradigm distillation, and does not predict transfer quality (rho=0.08 against transfer accuracy).

The growing demand for transparency in automated decision-making has propelled eXplainable Artificial Intelligence (XAI) to the forefront of machine learning research. In computer vision, however, existing explanation methods often prioritize end-user accessibility at the expense of formal guarantees, leaving a critical gap between practical utility and theoretical rigor. In this paper, we address this gap by introducing OPTIMUS, a novel framework for generating concept-based visual explanations for deep classification models. OPTIMUS explanations take the form of visual heatmaps that not only remain interpretable to end users, but are grounded in the well-established theory of prime implicants, providing formal guarantees that have been largely absent from existing saliency-based methods. Specifically, OPTIMUS explanations satisfy two desirable properties: sufficiency, ensuring that the highlighted concepts provably guarantee the classifier's prediction, and minimality, ensuring that no strict subset of those concepts retains this guarantee. Together, these properties yield explanations that are both logically tight and visually coherent. We validate our approach on a visual classification benchmark, demonstrating that OPTIMUS heatmaps naturally and faithfully surface the decision-relevant concepts underlying model predictions.

Variations in illumination remain a major challenge for visual representation learning, as they induce substantial appearance changes both across and within environments. While existing approaches typically address this issue through data augmentations that encourage models to become invariant to lighting changes, such strategies do not explicitly model lighting information during learning. Inspired by theories of human vision, we propose a lighting-aware representation learning framework that incorporates illumination variation as an explicit training signal rather than a nuisance factor to be suppressed. Our method extends contrastive learning by introducing an auxiliary objective that captures illumination-dependent variation in rendered scenes, enabling the model to jointly learn representations that preserve semantic consistency while remaining sensitive to lighting-dependent visual structure. We evaluate the proposed model on image classification and object detection tasks across the ImageNet, ExDark, and PASCAL VOC benchmarks. Results demonstrate that the proposed lighting-aware training consistently improves downstream performance over standard contrastive learning baselines, while maintaining the same architecture and training budget. Furthermore, our approach shows promising performance in supervised learning frameworks and under settings involving simpler lighting variation, suggesting broad applicability beyond complex illumination scenarios. These results indicate its potential to enhance model robustness and adaptability in complex visual environments as well as in more conventional image processing tasks.

GUI agents - vision-based models that control desktops, web browsers, and mobile devices through graphical user interfaces - promise to automate a wide range of digital tasks. While million-scale datasets have enabled substantial progress on click-grounding, drag grounding (e.g. drag-and-drop, swipe, highlight) data remains an order of magnitude smaller and current models fall short on complex drag-based interactions. We introduce DragOn, a drag grounding benchmark and training dataset covering four domains: text highlighting, cell selection, element resizing and slider manipulation. The dataset comprises 286K training screenshots and 3.5M training tasks, plus a 2000-example held-out evaluation suite. We evaluate proprietary (GPT, Claude) and open-weight (Qwen, Kimi, Holo) models, as well as a Qwen VLM fine-tuned on our training data. Results suggest that our dataset could improve performance of state-of-the-art models on downstream computer-use tasks.

Longitudinal medical visual question answering (VQA) requires reasoning about anatomical differences between an image of a current time point and an image of a referred time point. We propose an attention-guided encoder-decoder for this task with chest X-rays. Instead of conventional direct contrast, we propose to include a lightweight affine registration module to reduce nuisance motion by co-registering the current image to the reference image with a small registration regularizer. The registered image pair is fed into the image encoder, followed by a frozen DINO-based mask generator and a trainable adaptive mask generator to produce masks applied to the original image pairs. The masked image pairs are again fed into the image encoder and concatenated with text features as the input to a multimodal transformer-based decoder to generate final answers. To facilitate learning stabilization and clarify the change signal, inspired by DINO-v3, we include additional auxiliary objectives, including a mask rebuilding loss, a pairwise Gram-style consistency loss, and a KoLeo uniformity loss, which enhances the geometry of the representation. On the Medical-Diff-VQA benchmark, the model delivers strong BLEU, ROUGE-L, CIDEr, and METEOR scores while offering intrinsic interpretability through the shared saliency mask. These results support saliency-conditioned generation with mild pre-alignment as a principled framework for longitudinal reasoning in medical VQA. Our training strategy also illustrates the potential of a paradigm in utilizing image foundation models in biomedicine: optimizing both supervised and unsupervised learning objectives simultaneously.

We propose a label-free approach to adapt powerful but generic vision foundation models to specialized scientific domains. Standard supervised fine-tuning is often ill-suited to these settings: labels are scarce, and task-specific training can collapse the model's generality and hurt robustness. We instead leverage metadata to adapt representations to new domains in a self-supervised manner. Our method, FINO, combines a standard self-supervised objective with flexible metadata guidance that handles both highly granular discrete metadata and continuous metadata. It encourages the representation to preserve informative factors while suppressing spurious ones. Across subcellular fluorescence microscopy, Earth observation, wildlife monitoring, and medical imaging, FINO consistently outperforms standard unsupervised domain adaptation and fully supervised adaptation. It also exceeds highly-specialized domain-specific state of the art, while using no task labels for backbone adaptation and only lightweight probes for supervision.

Adapting foundation models to detect generative artifacts via gradient-based updates compromises their intrinsic representations. Under optimization on limited samples, models overfit to local domain shortcuts. Fine-tuning massive weights on specialized data introduces erroneous inductive biases, inducing a measurable $\mathcal{L}_2$ norm perturbation in the high-dimensional feature space -- a phenomenon we formalize as anchor drift. Amplified by nonlinear activations, this drift impairs zero-shot forgery detection across unseen domains.We propose a gradient-free methodology reframing detection from binary classification to an out-of-distribution (OOD) anomaly measurement problem. Treating a frozen foundation model as a stable coordinate system, we establish an absolute natural anchor on the real visual manifold by analytically decoupling statistical and semantic deviations, derived from attention-weighted spatial moments and orthogonal projection of perceptual inconsistencies. Evaluated in an extreme zero-shot setting (trained on face forgeries, tested on universal Text-to-Image generations), our method significantly outperforms gradient-optimized paradigms. Backpropagation-free forward passes and linear solvers enable hardware-agnostic, edge-deployable calibration with minimal latency. Furthermore, the Sherman-Morrison formula unlocks instantaneous online learning against novel attacks and enables privacy-preserving federated collaboration via covariance delta transmission.

Foundation models have driven rapid progress in computer vision, yet the two dominant paradigms, vision-language foundation models (VLMs) and vision-only foundation models (VFMs), remain only partially compatible. VLMs offer language-grounded semantic alignment but are often visually coarse, while VFMs learn discriminative perceptual geometry but lack semantic grounding. We propose GPUA (Geometry-Preserving Unsupervised Alignment), a framework that integrates the complementary strengths of VFMs and VLMs. Inspired by cross-lingual alignment, GPUA treats VFM features as a visual language and learns an orthogonal mapping that translates the VFM space into the VLM semantic space, preserving geometry and narrowing the modality gap without labels or model parameter updates. GPUA is task-agnostic and requires only feature-level access to pretrained models. Experiments across diverse benchmarks demonstrate improved cross-model compatibility and strong gains in downstream zero-shot recognition and segmentation with negligible overhead. Code is available at https://github.com/Yuteam14/GPUA

Previous work has evaluated physics reasoning in foundation models using synthetic or semi-synthetic scenes and visual question-answering tasks. However, these benchmarks emphasize high-level events and lack the visual fidelity required to assess true low-level Newtonian understanding. We introduce NewtPhys, a 4D physically annotated dataset built from multiview images of real-world scenes with physics-grounded simulations. The dataset provides dense, fine-grained annotations across timesteps -- including 3D forces and amodal per-pixel quantities covering physics, tracking, semantics and geometry -- bridging the gap between simplistic synthetic setups and realistic visual complexity. Using NewtPhys, we systematically evaluate 56 VLMs, including 54 open-weight models and 2 closed-source frontier models, and 10 VFMs and reveal limitations in low-level physics reasoning. Beyond benchmarking, our dataset enables future research in physics-grounded vision and the development of next-generation physics-aware evaluations. Code and datasets are available at https://astra-vision.github.io/NewtPhys.

Real-time vision demands models that are accurate, efficient, and simple to deploy across diverse hardware. The YOLO family has become widely deployed for this reason, yet most YOLO detectors still rely on non-maximum suppression at inference, carry heavy detection heads due to Distribution Focal Loss, require long training schedules, and can leave the smallest objects without positive label assignments. We present Ultralytics YOLO26, a unified real-time vision model family that addresses these limitations through coordinated architecture and training advances. YOLO26 uses a dual-head design for native NMS-free end-to-end inference and removes DFL entirely, yielding a lighter head with unconstrained regression range. Its training pipeline combines MuSGD, a hybrid Muon-SGD optimizer adapted from large language model training; Progressive Loss, which shifts supervision toward the inference-time head; and STAL, a label assignment strategy that guarantees positive coverage for small objects. Beyond detection, YOLO26 introduces task-specific head and loss designs for instance segmentation, pose estimation, and oriented detection, producing consistent gains across tasks and scales. The family spans five scales (n/s/m/l/x) and supports detection, instance segmentation, pose estimation, classification, and oriented detection in a single pipeline, with an open-vocabulary extension, YOLOE-26, for text-, visual-, and prompt-free inference. Across all scales, YOLO26 achieves 40.9-57.5 mAP on COCO at 1.7-11.8 ms T4 TensorRT latency, advancing the accuracy-latency Pareto front over prior real-time detectors, while YOLOE-26x reaches 40.6 AP on LVIS minival under text prompting. Code and models are available at https://github.com/ultralytics/ultralytics.

Existing human attention (saliency) modeling methods persist as highly fragmented across modalities, scenes, and task formulations. Consequently, even with increasing model capacity and data scale, current models predominantly remain scene-dependent and task-specific, failing to practically generalize in real-world applications. To address the fundamental limitations, we present the Attend to Anything Model (AAM), a multi-modal foundation model that unifies attention modeling across various image, video, and audio-visual tasks and scenes. AAM reformulates attention as a cognitive entailment relationship organized in a general-to-specific hierarchy, implemented through language prompts with hierarchical embeddings in hyperbolic space. Furthermore, to unify static image and dynamic video attention, we adopt a fluid-dynamics perspective, formulating video-frame attention as a diffusive temporal evolution governed by the Fokker--Planck equation. Extensive experiments on 16 benchmarks demonstrate that AAM consistently outperforms state-of-the-art methods by an average of 6% across various scenarios, while achieving approximately a 4$\times$ speedup in video inference. Overall, these results demonstrate that AAM provides a principled foundation for future research on attention and saliency-related tasks. The dataset and code will be available at https://github.com/wz-zhao/Attend-to-Anything.

Unifying the complementary strengths of diverse Vision Foundation Models (VFMs) into a single efficient model is highly desirable but challenged by the negative transfer inherent in monolithic distillation. To address these feature conflicts, we introduce \textbf{PRISM}, a novel dual-stream Mixture-of-Experts (MoE) framework that synergizes VFMs via modular specialization. We propose a two-stage paradigm: (1) expertise deconstruction, where a teacher-conditional router guides experts to specialize in distinct representational subspaces to mitigate interference, followed by (2) dynamic recomposition, where the router learns to assemble these experts into tailored computational pathways for downstream tasks. Experiments on PASCAL-Context and NYUD-v2 show that \textbf{PRISM} establishes a new state of the art, validating that sparse, emergent specialization is a scalable approach for integrating diverse visual knowledge.

Gaussian splatting has shown strong potential for 3D reconstruction and novel view synthesis. However, most existing methods require accurate camera parameters and per-scene optimization, while feed-forward methods with pixel-aligned Gaussian primitives often suffer from artifacts and non-uniform primitives. In this paper, we propose $\text{VG}^2$GT, a Voxel-Gaussian Splatting Visual Geometry-Grounded Transformer. $\text{VG}^2$GT leverages a frozen pretrained visual foundation model (VFM), incorporates a multi-scale differentiable voxel module to enhance geometric understanding, and directly splits and regresses Gaussian primitive parameters from voxel features. During training, depth maps are supervised through stochastic solid volume rendering, enabling geometrically accurate Gaussian scene reconstruction while keeping the visual foundation model fully frozen. This design enables $\text{VG}^2$GT to be seamlessly plugged into any patch-feature-based VFM, while substantially reducing the required training cost. $\text{VG}^2$GT outperforms current state-of-the-art methods on widely used DTU, Replica, TAT, and ScanNet datasets.

Vision-Language Models (VLMs) exhibit emerging spatial reasoning capabilities, yet they remain unreliable on tasks requiring precise spatial understanding, such as viewpoint reasoning, directional comparison, and distance estimation. In multi-view images and monocular videos, relevant spatial cues are often sparse and distributed across redundant observations, making them difficult to organize and exploit. Reconstruction-based Vision Foundation Models (VFMs) offer a natural way to aggregate such observations into explicit spatial memory, such as point clouds. However, simply exposing reconstruction models as free-form tools is brittle, VLMs may invoke tools incorrectly, skip required spatial transformations, or misuse intermediate results. We propose \textbf{Reasmory}, a framework that formulates spatial reasoning as structured program execution over reconstructed spatial memory. Reasmory constructs explicit 3D memory, augments it with semantically grounded 3D object instances, and introduces a lightweight Domain-Specific Language (DSL) that constrains how VLMs query objects and cameras, transform viewpoints, and render observations during reasoning. Generated programs are parsed and validated before execution, enabling more reliable interaction with spatial memory than unconstrained tool use. Experiments on multi-view image and video spatial reasoning benchmarks show consistent gains of 6--18% over strong baselines, including GPT-5-mini and Gemini-3-flash, indicating that explicit 3D memory is most useful when accessed through constrained, validated operations rather than free-form tool calls.

Water research in Brazil largely overlooks the widespread damming of small streams for agricultural uses such as watering cattle, farm-scale hydropower, irrigation, and aquaculture. These ubiquitous dams and their reservoirs can alter water temperature, stream connectivity, aquatic habitats, greenhouse gas emissions, and evaporative water losses. Mapping small reservoirs is challenging because it requires reliably detecting small water bodies and distinguishing artificial reservoirs from natural lakes. As a result, most regional and global datasets exclude them. To address this gap, we trained a deep learning computer vision model to accurately segment small ($< 1 km^2$), stream-fed, surface water reservoirs in Brazil leveraging data from Landsat 5-9. Applying our model from 1984 to 2025, we created annual reservoir maps for the entire country to evaluate how their count, size, and distribution have changed over time. The number of detected reservoirs grew nearly fourfold from 263,913 to 996,245, while their total surface area increased from 3510 $km^2$ to 8550 $km^2$. To our knowledge, this is the first country-wide annual dataset representing the evolution of small reservoirs over four decades. The publicly available annual maps highlight the extent and cumulative impacts of the small stream impoundments across Brazil, providing actionable insights for managing freshwater ecosystems and water resources.

Measuring structured object understanding in vision foundation models remains challenging due to inconsistent evaluation protocols and limited part-level supervision. Semantic correspondence (SC) evaluates this capability by testing whether object parts can be matched across instances and categories under large variations in appearance, viewpoint, and geometry. To enable a systematic SC evaluation, we introduce SOCO, a new benchmark for Semantic Object Correspondence that introduces a taxonomy of correspondence types and provides consistent, functionally meaningful keypoint annotations across 100 categories and over 1M correspondence pairs. In addition, SOCO includes keypoint language descriptions, enabling the evaluation of large vision-language models (LVLMs) and their fine-grained part-level understanding. Comprehensive experiments reveal that (i) vision foundation backbones encode strong semantic structure but transfer correspondences poorly across related categories and only partially capture object-part position, (ii) LVLMs are stronger at text-prompted part localization than at visual-reference cross-image matching, exposing a gap between language-grounded localization and fine-grained visual correspondence, and (iii) correspondence performance predicts performance on dense downstream tasks, including segmentation, tracking, 3D pose estimation, and 3D detection, more strongly than ImageNet classification. Together, these findings position SOCO as a benchmark for structured, part-level representation quality in vision and multimodal foundation models.

Deploying computer vision models in Warehouse Facilities traditionally requires extensive resources for camera mounting, image collection, annotation, training, and deployment - a process often needing repetition in each new environment due to camera mounting constraints and environmental variability. This paper explores an innovative approach to streamline this process by conducting the standard procedure solely in a laboratory setting, focusing on vertical material handling systems and anomaly detection in forks of the systems. Through extensive experimentation, we have found that combining optimal camera placement, strategic image triggering, careful model selection and model ensemble enables effective generalization from laboratory conditions to diverse warehouse facilities environments, potentially transforming warehouse automation implementation by simplifying warehouse facilities deployment to just camera mounting, image collection, and model deployment, thereby saving significant resources and time typically spent on image annotation and model retraining. This is an experimental research study and not a production deployment.

Vision-based approaches have become the dominant paradigm for traversability estimation in unstructured outdoor environments, typically adapting vision foundation models (VFMs) via semantic segmentation supervision. However, this paradigm faces three fundamental challenges that undermine its reliability: the task-agnostic design of VFMs, the ambiguity of traversability annotations, and the discrepancy between semantic labels and physical safety. We propose Vision-to-Traversability Adaptation (ViTA), a framework that adapts VFMs for reliable traversability estimation, instantiated on SAM2. ViTA injects task-specific knowledge through learnable traversability prompts while preserving the VFM's cross-domain generalization. To handle annotation ambiguity, we introduce Perspective-Diversified Training, which estimates semantic uncertainty to suppress confident predictions at ambiguous boundaries. To bridge the semantic-traversability discrepancy, we distill geometric knowledge during training, enabling slope and elevation reasoning from RGB images alone at inference. The semantic and geometric outputs are fused into a continuous traversability score that reflects both semantic uncertainty and geometric risk. Evaluations across diverse domains, including challenging real-world off-road datasets, demonstrate that ViTA achieves state-of-the-art IoU and Precision with substantial false-positive reduction and strong cross-domain generalization.

Current vision-language models (VLMs) typically stitch together separate image encoders and language decoders via multi-stage alignment, a modular framework that inevitably fragments pixel-level signals across frames and scatters early pixel-word interactions. In parallel, native VLMs, despite impressive performance on single images, remain largely unexplored in multi-image, video understanding, and spatial intelligence. Hence, we introduce NEO-ov, a native foundation model that learns cross-frame and pixel-word correspondence end-to-end, without any external encoders, auxiliary adapters, or post-hoc fusion. By eliminating module boundaries entirely, NEO-ov enables fine-grained and unified spatiotemporal modeling to emerge natively inside the model. Notably, NEO-ov largely narrows the gap to modular counterparts while excelling at fine-grained visual perception, validating that native "one-vision" architectures are not only feasible but competitive at scale. Beyond empirical performance, we unveil systematic architectural analyses and detailed training recipes to facilitate subsequent native multimodal modeling. Our code and models are publicly available at: https://github.com/EvolvingLMMs-Lab/NEO.

Vision Foundation Models (VFMs) have demonstrated impressive representational capabilities. However, adapting them to downstream tasks via full fine-tuning incurs prohibitive computational and storage overhead. Parameter-Efficient Fine-Tuning (PEFT) has emerged as a compelling alternative, aiming to achieve performance parity with full fine-tuning at minimal training costs. Nonetheless, applying PEFT to VFMs for dense prediction tasks remains challenging due to the structural and distributional gaps. To bridge these gaps, we propose \textbf{S}cale-\textbf{I}ntegrated \textbf{G}lobal \textbf{M}odulation \textbf{A}dapter (\textbf{SIGMA}), a novel lightweight PEFT method, which consists of two modules: scale-adaptive fusion and semantic modulation. Specifically, the scale-adaptive fusion module is utilized to bridge structural gaps by enhancing the extraction of multi-granularity visual information. Furthermore, SIGMA introduces semantic modulation on the fusion features to perform global feature alignment to further eliminate the distribution gap. This design facilitates unified spatial and distributional adaptation, requiring only 1.72% trainable parameters relative to the VFM backbone. Comprehensive experiments across various downstream dense tasks and multiple VFM backbones demonstrate that SIGMA achieves consistent and superior performance over state-of-the-art PEFT methods.

Unsupervised visual object tracking is a challenging task that requires following arbitrary targets in videos without training on ground-truth annotations. Despite considerable progress, existing state-of-the-art unsupervised trackers often struggle in scenarios that demand fine-grained understanding of semantic and visual structural information within video frames. Text-to-image diffusion models are well known for their ability to generate images that accurately reflect the semantics and structures described in the input prompt, demonstrating a strong grasp of visual semantics and structures. Building on this capability, we approach the unsupervised tracking from a new perspective by exploiting the rich semantic knowledge encoded in pretrained text-to-image diffusion models. To adapt the diffusion models, which are originally developed for image generation, to the tracking task, we reinterpret the models as a bridge between text and image modalities. This connection is realized through the cross-attention mechanism: when both text and an image are input into the models, they highlight the regions of the image that are semantically aligned with the text in the cross-attention maps. We therefore learn a prompt that represents the tracking target and activates its corresponding region in the cross-attention map for each frame, which enables object tracking with the diffusion model. Specifically, our method Diff-Tracking is composed of two main components: an initial prompt learner and an online prompt updater. The initial prompt learner generates a prompt that captures the target object in the first frame, allowing the diffusion model to identify the target. The online prompt updater refines the prompt based on motion information, enabling consistent tracking across video frames. We evaluate our approach on six challenging tracking datasets demonstrate the effectiveness of our approach.

RGB-based imitation learning requires many demonstrations to generalize to unseen objects or scenes, motivating research into intermediate representations to improve generalization for robotic manipulation. Visual foundation models enable one-shot extraction of keypoints to provide such representation. However, it remains unclear how to integrate them into imitation learning optimally and when they outperform alternative representations. We combine approaches from previous works on keypoint imitation learning (KIL) and investigate several design choices to provide practical guidelines. Using over 2000 real-world rollouts, we also assess the generalization capabilities of KIL to unseen objects and scene variations. KIL achieves a 75% overall success rate across five tasks, significantly outperforming the RGB baseline (47%) and performing on par with S2-diffusion (73%). Finally, we explore the limitations of the foundation models used for keypoint extraction and extend KIL to tasks with multiple object instances. Our results confirm KIL as a data-efficient approach for robot learning, though it does not outperform alternative representations and inherits limitations of the foundation models used for keypoint extraction. All rollout videos, demonstrations, and results are available at https://kil-manipulation.github.io/.

We introduce JetViT, a novel family of hybrid-architecture Vision Transformer (ViT) models that match the accuracy of state-of-the-art full-attention vision foundation models while achieving substantially higher inference efficiency on high-resolution images. At the core of our approach is Post-Training Attention Search, a post-training acceleration framework that converts pre-trained full-attention ViTs into efficient hybrid-attention variants by identifying and replacing redundant full-attention blocks with linear or window-attention blocks. By inheriting the MLP and attention weights from the base model, Post-Training Attention Search efficiently explores the architectural design space through three key steps: (1) optimizing the linear-attention block design; (2) finding the best combination of linear-attention and window-attention blocks; and (3) identifying and preserving critical full-attention blocks. We evaluate JetViT on two representative high-resolution vision foundation models, DINOv3 and DepthAnythingV2. On the NVIDIA H100 GPU, JetViT achieves up to 1.79x higher throughput and up to 44.81% lower latency without sacrificing accuracy. We will release our code and accelerated ViT models soon.

Modern deep learning offers powerful tools for automated retinal screening, but it remains unclear how different visual model families compare in realistic multi-disease settings and under domain shift. In this work, we benchmark twelve architectures across four model families: convolutional neural networks, vision transformers, hybrid CNN-transformer backbones, and vision-language models, using the Retinal Fundus Multi-disease Image Dataset (RFMiD). We evaluate two tasks: binary screening for any retinal disease and multi-label classification across 28 disease classes. Using standardized training, calibration, and evaluation protocols, we report AUC, F1, precision, recall, and sensitivity at a clinically relevant operating point with specificity near 80%. On RFMiD, all architectures perform well on binary screening, with AUC above 84%, but attention-based models perform best. SwinTiny and the hybrid CoAtNet0 and MaxViTTiny models achieve the strongest binary screening results and improve macro and micro F1 in the multi-label setting. Vision-language models, including CLIP ViT-B/16 and SigLIP-Base384, are competitive with CNN baselines but do not surpass the best transformer and hybrid backbones. In external validation on Messidor-2 for referable diabetic retinopathy, AUC ranges from 66.8% to 84.7%, with hybrid and transformer models again showing strong performance. These results provide a reproducible reference for model selection in multi-disease retinal screening and guide future automated screening tools for clinical deployment.

Light sheet fluorescence microscopy (LSM) enables high-resolution, three-dimensional (3D) imaging of biological specimens, providing rich volumetric data for studying cellular organization, pathology, and vascular networks. However, the size, dimensionality, and annotation burden of LSM data make supervised deep learning approaches costly and difficult to scale. Additionally, despite the abundance of unannotated LSM volumes, foundation models for this modality remain underexplored due to computational challenges and the complexity of volumetric representation learning. In this work, we introduce a 3D foundation model for LSM data, pretrained on a large curated collection of 3D images spanning multiple organisms, stains, and imaging protocols. We learn transferable volumetric representations by jointly optimizing for masked reconstruction and image-text alignment. The pretrained backbone drastically reduces the annotation burden, enabling efficient, few-shot adaptation for varied downstream tasks. We evaluate this approach on downstream segmentation, classification, and deblurring. Our results demonstrate consistent improvements over baselines, (1) when measured using standard evaluation metrics and (2) when rigorously assessed by domain experts. This highlights the potential of foundation model pretraining to reduce annotation requirements while improving performance across diverse LSM analysis tasks. Pretrained model weights and code for pretraining and finetuning are publicly available: https://github.com/AdinaScheinfeld/lsm_fm_public_repo.git.

Ultrasound video segmentation is clinically valuable yet difficult due to speckle noise, weak boundaries, and rapid anatomical deformation. Recent promptable foundation models enable point-guided segmentation, but their direct deployment in ultrasound remains unreliable: a single point provides insufficient spatial context to resolve scale ambiguity, and greedy memory updates amplify early errors into severe temporal drift. We present EchoPilot, a training-free framework for ultrasound video segmentation under sparse first-frame interaction, requiring only a single point click and an anatomical category name. EchoPilot orchestrates a frozen medical vision-language model (VLM) for semantic localization, a vision foundation model (VFM) for dense geometric feature extraction, and a promptable video segmentor for mask prediction and propagation. To resolve initialization ambiguity, we propose Scale-Space Semantic Prompting, which first selects an optimal contextual view via a parameter-free S.E.E.D. (Semantic Energy-Entropy Density) criterion, and then synthesizes geometrically precise auxiliary point prompts from dense foundation features without additional user interaction. To reduce propagation drift, a Reliability-Gated Memory update is further introduced to selectively freeze the segmentor's memory bank under uncertain predictions, preventing error accumulation. We also contribute the first dynamic fetal placenta ultrasound video segmentation dataset with 671 annotated frames. Across three ultrasound video datasets, EchoPilot achieves state-of-the-art performance under the sparse-interactive setting, consistently outperforming training-free baselines and finetuned specialists.

Robotic perception in unstructured environments remains challenging despite the zero-shot capabilities of foundation models such as SAM. This work attributes performance degradation to non-uniform representation shifts across transformer layers: shallow layers exhibit substantial domain gaps (CKA < 0.5), whereas deep layers transfer effectively (CKA > 0.7). Based on this observation, we propose RepSAM, a representation-guided parameter-efficient fine-tuning (PEFT) framework for adapting foundation models to robotic vision. RepSAM employs a theoretically grounded CKA-guided rank allocation strategy combined with a multi-modal fusion module for robust handling of challenging robotic scenarios, including transparent objects and cluttered scenes. Experimental evaluation across six benchmarks and robotic manipulation tasks demonstrates that RepSAM achieves 97.9% of full fine-tuning performance (89.0% vs. 90.9% mIoU) while reducing trainable parameters by 158x (from 632M to 4.0M). RepSAM outperforms DoRA by 7.9% mIoU with just 4 hours of training on a single A100 GPU (a 96x reduction from full fine-tuning, which takes 384 GPU-hours). These improvements are statistically significant (p < 0.01) and translate to a 12.0% absolute improvement in robotic manipulation success rates over the LoRA (RGB) baseline.

Self-supervised pre-training paradigm has gained increasing prominence for learning transferable representations in medical imaging, yet existing methods for ultrasound (US) images operate at the image or frame level, overlooking the anatomical context for clinical-aligned representation learning. In this work, we propose an anatomy-anchored ultrasound self-supervision framework ANAUS that shifts representation learning from generic visual regions to clinically meaningful anatomical structures. Utilizing a learnable latent prompt engine alongside a one-time domain adaptation on existing public image-mask pairs, we empower the LP-SAM module to achieve annotation-free anatomy delineation at scale. Building upon this anatomical grounding, we propose a dual-policy self-supervised learning paradigm consisting of inter-view semantics-aware anatomy-separating alignment and contextual core-region prediction to enhance representation learning. Specifically, the former enforces feature invariance within identical anatomical regions while promoting discriminability across distinct structures; the latter compels the model to reconstruct corrupted regions, thereby capturing fine-grained structural details. Extensive evaluations on six public datasets demonstrate that ANAUS consistently outstrips current state-of-the-art methods while maintaining the computational efficiency essential for clinical deployment. Code is available at https://github.com/zhcz328/ANAUS.

The rapid evolution of generative models has enabled the creation of hyper-realistic facial deepfakes, exposing a critical vulnerability in modern digital forensics: the inability of detectors to generalize to unseen manipulation techniques. Traditional networks suffer from representation collapse, overfitting to localized artifact fingerprints of specific training generators. This work investigates whether modern Vision Foundation Models can serve as generalizable, out-of-the-box feature extractors capable of tracking forensic anomalies across entirely unseen generative manifolds. We conduct a systematic cross-domain evaluation comparing three foundational learning paradigms: fully supervised macro-semantic features (RoPE-ViT), pure self-supervised geometric features (DINOv3), and multi-teacher agglomerative representations (NVIDIA C-RADIOv4-H). By deploying frozen backbones subjected to downstream linear probing, we map the performance limitations of these architectures on the challenging DF40 benchmark. Our empirical findings expose the intrinsic trade-offs between pre-training paradigms and parameter scale, proving that while foundation models retain high discriminative capabilities for entire face synthesis, localized face editing techniques expose fundamental boundaries in linear probe evaluation structures. Source code and model weights are available in http://github.com/mribrahim/deepfake

Building upon the SAM2 vision foundation model for downstream segmentation, this study introduces Boundary Enhanced Depth (BED)-SAM2. The SAM2 Hiera encoder architecture is modified to directly encode monocular depth information from RGB images, thereby providing geometric cues that enhance object boundary delineation and facilitate the extraction of camouflaged object shapes. BED-SAM2 demonstrates competitive state-of-the-art performance across multiple salient and camouflaged object detection tasks with as few as five training epochs.

This study presents a closed-loop robotic strawberry harvesting system that combines a robust vision module, simulation-trained deep reinforcement learning (DRL) control, and ROS-based realrobot execution. For perception, we propose HRAttnEdge-YOLO26-seg, a modified YOLO26-seg architecture that incorporates a high-resolution P2 branch, segmentation-path attention, and edgesupervised prototype learning to improve instance segmentation in cluttered scenes. For control, we train a target-conditioned Proximal Policy Optimization (PPO) policy in Isaac Lab to produce smooth joint-position commands for a UR10e manipulator and deploy it on a UR10e robot for targetfruit reaching and harvesting. This simulation-based approach reduces hardware dependency, lowers development cost, and allows scalable policy training without exhaustive physical trials before real deployment. The proposed vision model demonstrated the highest overall performance among the evaluated methods. On both self-collected and public datasets, the model showed a 10 to 14% improvement in segmentation performance. In controlled in-house tests, the PPO controller produced stable and dynamically smoother motion than a inverse kinematics (IK)-based MoveIt baseline. In greenhouse trials, the proposed integrated system harvested 281 strawberries, achieving 96.6% reaching success, 91.3% grasp-and-pull success, and 84.3% overall harvesting success. These results illustrate that task-specific perception combined with simulation-trained PPO can serve as a practical and resource-efficient alternative to conventional planner-dependent reaching in manipulation, enabling reliable closed-loop robotic harvesting in complex agricultural environments.

Training high-capacity vision models from scratch requires substantial computational resources. To improve training efficiency of a wide target model, existing growth methods often assume the availability of narrower models, obscuring the true computational cost of the entire pipeline. We propose an efficient training protocol, RBDC, that builds wide models by coupling in a parameter-free block-diagonal way narrower, independently trained models in a recursive way. This allows a flexible allocation of the training budget available across all the models involved. Evaluated with vision transformers (DeiT) and convolutional networks (ResNet) on ImageNet, our RBDC training protocol shows a much better efficiency than models trained from scratch with the standard protocol, yielding 30% FLOPs reduction at similar test accuracies. It also achieves higher performances at same training FLOPs than training protocols from the model growth literature. Finally, we show that our models can serve as better backbones than their original counterparts for downstream object detection and instance segmentation tasks.

Vision foundation models are widely used as frozen backbones across many downstream tasks, making them a single point of failure under adversarial attack. We study multi-level Floyd-Steinberg error-diffusion dithering as a lightweight, model-agnostic input transformation that disrupts adversarial perturbations while preserving semantic content. Unlike prior work, which was limited to binary dithering, grayscale CIFAR-10, and a single small model trained from scratch, we evaluate across six tasks (classification, segmentation, depth estimation, retrieval, captioning, visual question answering), two model families (DINOv2, PaliGemma), and three attacks of increasing strength (PGD, MI-FGSM, SIA), as well as an adaptive attacker using a straight-through estimator. Our results show that Floyd-Steinberg dithering at intermediate quantization levels, especially when combined with post-processing blur, exceeds or matches all tested baselines, including diffusion-based denoising, with substantially less degradation on clean inputs.

Gaze following requires both scene understanding and gaze reasoning to localize the gaze target of an in-scene person. Recently, vision foundation models (VFMs) have demonstrated strong performance on this task, enabling simpler architectures while outperforming prior methods. However, we observe a key limitation of VFM-based approaches: while VFMs substantially improve scene understanding, they contribute little to gaze reasoning. As a result, existing methods often rely on semantically salient objects rather than true gaze cues, leading to degraded performance when targets are not salient. To address this, we propose a novel training mechanism to enhance gaze reasoning in VFMs for gaze following. Our method includes: (1) a head-conditioned local LoRA, which enables localized adaptation to preserve scene token learning while improving head token learning for gaze reasoning; and (2) an out-of-cone penalty, which injects gaze cues into head tokens while aligning them with scene tokens. Experiments on the GazeFollow and VAT datasets demonstrate that our method achieves state-of-the-art performance, with particularly strong improvements when gaze targets are not semantically salient. Our findings offer valuable insights for advancing future gaze following research. We will release the code once the paper is accepted.

Frozen Vision Foundation Models (VFMs) with lightweight classification heads are increasingly used in medical imaging because they offer efficient and reproducible deployment. Yet noisy-label learning methods for this frozen-feature regime remain poorly understood, and most existing methods still rely on a small-loss assumption inherited from end-to-end training. We present a controlled benchmark of eight noisy-label methods across five medical datasets, three backbones, two noise types, and five noise rates (150 conditions, 6,000 training runs), evaluated with balanced accuracy. The benchmark shows that there is no universal winner: Friedman ranking over the 150 conditions yields $χ^2 = 333.2$ ($p = 4.77 \times 10^{-68}$), ELR wins the most conditions (49/150), while CUFIT attains the best mean rank (2.51). The practical cost of method choice grows sharply with noise severity, from 4.5pp on clean data to 18.8pp at asymmetric 40% noise. To explain these benchmark-level patterns, we revisit the small-loss assumption in a representative high-risk regime. Under frozen DINOv2 features, clean and noisy loss distributions overlap by 53--61%, and matched-rate clean-sample detection shows that prediction agreement is markedly more stable than loss ranking under asymmetric noise (3pp vs.\ 13pp precision drop). On ISIC2019 with asymmetric 40% noise, Co-Teaching reaches 68% overall accuracy while collapsing to 35.1% balanced accuracy with zero recall on three minority classes. Together, these results recast noisy-label learning for frozen VFMs as a regime-aware method-selection problem rather than a search for a single dominant algorithm. We conclude with evidence-based guidance and a low-regret feature-space selector for practical recommendation.

Real-world deployment of AI vision models is both fueled and limited by the data available for training and testing. Real datasets are sparse and uneven: long-tailed or unbalanced distributions hinder generalization, and the low number of samples in low density regions makes it hard to run evaluations. Synthetic data can fill these gaps, providing us with a way to sample the input space more continuously and improve data coverage for benchmarks. Focusing on the autonomous driving safety-critical case of pedestrian detection in the dark, we show how synthetic low-light samples can be used to better characterize the performance of a state-of-the-art object detection model as a function of the scene illumination. We use a synthetic RAW image augmentation technique to generate low-light samples that match the noise model of the camera sensor. Performance metrics on real and synthetic low-light data are similar, indicating that the AI model finds it hard to distinguish between them.

Pretrained vision foundation models deliver strong performance across tasks with limited fine-tuning. However, their Vision Transformer (ViT) backbones impose high inference costs, limiting deployment on resource-constrained devices. In this work, we accelerate large-scale pretrained ViTs while preserving their feature extraction capabilities by exploiting the intrinsic convolution-like behavior of some attention heads. Specifically, we introduce an efficient depthwise convolution-based layer that serves as a drop-in replacement for these heads. Additionally, we propose simple strategies to identify which heads can be replaced and introduce a fine-tuning procedure that recovers downstream task performance. Across both image classification and segmentation tasks, our method achieves 17-20% percent inference speedup with minimal performance degradation. We validate the approach through detailed derivations, extensive experiments, and efficiency benchmarks. The reference implementation is publicly available.

Modern vision models achieve strong performance on standard benchmarks, yet their aggregate accuracy reveals little about which scene properties drive their predictions. Existing robustness benchmarks provide important stress tests, but typically manipulate global 2D image properties, rely on entangled real-world variation, or cover only a limited set of 3D objects and scene parameters. We introduce MAPS (Manifolds of Artificial Parametric Scenes), a scalable instrument for controlled attribution of vision model behavior to scene parameters. MAPS comprises 2,618 curated photorealistic 3D meshes validated for recognizability across 560 ImageNet classes and provides a Blender-based rendering pipeline for on-demand image generation under continuous variation of nine independent scene-factors spanning background, camera, and lighting, extensible to other factors. To showcase its applicability, we use MAPS to evaluate 20 convolutional and transformer-based models by quantifying their reliance on these scene factors through regression-based sensitivity analysis. We find a near-universal failure axis across all tested architectures: camera distance and elevation consistently dominate recognition failure regardless of ImageNet accuracy. However, the full sensitivity structure reveals that modern CNNs and transformers cluster together, distinct from older architectures, suggesting that fine-grained architectural design choices, rather than the coarse CNN-versus-transformer distinction, are the stronger determinant of sensitivity profiles.

Foundation models like the Segment Anything Model (SAM) demonstrate impressive zero-shot generalization but frequently degrade under diverse real-world illumination, particularly for instance segmentation. In this work, we address this limitation by developing \textit{Lighting Convolutional-Attention (\lca{})}, an adapter module that enhances segmentation robustness without fine-tuning the heavy backbone. \lca{} employs a dual-branch architecture to process RGB features alongside contrast maps, enabling physically motivated sensitivity to structural changes rather than illumination artifacts. We optimize \lca{} through a pairwise training strategy, introducing a targeted loss term that explicitly penalizes discrepancies between clean images and their corresponding illumination variants. To evaluate and support this architecture, we conduct a comprehensive empirical study across multiple existing benchmarks and present a novel Unity-based synthetic dataset specifically designed to accurately replicate complex real-world lighting conditions. Extensive experimental results demonstrate that our approach successfully bridges the domain gap, delivering superior lighting-robust segmentation.

How interpretable are the features of leading vision models? The question is increasingly pressing as these models move from research benchmarks into high-stakes deployments, yet existing methods cannot answer it reliably. We close this gap with a framework for measuring and comparing the human interpretability of vision models, built around two complementary psychophysics protocols: (1) localizability -- can an observer predict where a feature fires on a novel image? -- and (2) nameability -- can an observer accurately describe what the feature represents? Features are recovered via sparse autoencoders, and a chance-anchored scoring function places every model on a common scale. Applying the framework to six vision transformers -- two supervised ViTs and four foundation models (DINOv2, DINOv3, CLIP, SigLIP) -- we collected more than $15{,}000$ behavioral responses, analyzing the $13{,}400$ responses from the $377$ participants who passed our pre-specified quality checks. Foundation models are consistently less interpretable than their supervised counterparts, and the gap is not a capability tradeoff: interpretability does not correlate with downstream task performance on any benchmark we examine. What does correlate is the locality of a feature's activations and coarse-grained semantic alignment with humans -- models with focal activations and representations that reflect the world's broad categorical structure produce more interpretable features, whereas fine-grained perceptual alignment does not. The two protocols yield strongly correlated rankings and share the same predictors, establishing interpretability as an independent, measurable dimension of representation quality -- and, surprisingly, one on which every foundation model we tested falls below the supervised baselines that came before. Capability alone cannot close that gap; locality and coarse-grained alignment can.

Artificial vision models are often evaluated against the human visual cortex by measuring how accurately their internal representations predict brain responses. However, prediction accuracy alone does not indicate which dimensions of the target brain's response space are recovered. Here, we introduce a unified framework for evaluating both model-brain and brain-brain alignment by identifying the response dimensions recovered by prediction. Using repeated fMRI measurements, we first identify target-brain response dimensions that can be reproducibly predicted across independent trial splits. We then predict target-brain responses from either another subject's brain responses or a vision model's internal representations, and quantify how strongly each of these reproducible response dimensions is recovered. Applying this framework to a subset of the Natural Scenes Dataset, in which eight subjects viewed the same natural images during fMRI, we find that the early-to-intermediate visual-cortex responses contain a low-dimensional set of reproducible dimensions. Brain-to-brain comparisons identify which of these dimensions are consistently recoverable from other subjects' brains, providing a diagnostic human reference rather than only a scalar benchmark. In some cases, pretrained and randomly initialized models achieve similar prediction accuracy while showing distinct recovery profiles across these response dimensions. These results show that prediction accuracy alone can mask model-brain mismatches. By making explicit which reproducible brain response dimensions are recovered by prediction, our framework provides a more diagnostic evaluation of alignment between artificial vision models and the human visual cortex.

Machine unlearning in Vertical Federated Learning (VFL) has attracted growing interest, yet existing methods certify forgetting solely using output-level metrics. We challenge these works by introducing Mirage, a representation-level auditing framework that comprises four complementary diagnostics: Linear probe recovery (LPR), centered kernel alignment (CKA), feature separability scoring, and layer-wise recovery analysis. Extensive experiments across seven datasets and seven baseline methods following recent VFL unlearning protocols reveal three key findings: (1) Forgetting gap: methods that pass output-level certification still retain substantial class structure in their representations, with LPR exceeding the retrained baseline by up to 15.4 points; CKA shows that these models remain structurally closer to the original than to the retrained reference, while separability scores indicate persistent geometric discrimination. (2) Unlearning trilemma: no existing method simultaneously achieves high utility, output-level forgetting, and representation-level forgetting. (3) Class-sample asymmetry: class-level forgetting leaves strong representational traces (LPR exceeding 96 percent on several datasets), whereas sample-level forgetting is indistinguishable from chance (LPR is approximately 50 percent); layer-wise analysis further shows that residual class information persists across network depths. These findings call for representation-aware evaluation standards in federated unlearning research. Code is publicly available at https://github.com/YuZhenyuLindy/Mirage.

Vision foundation models (VFMs) have achieved strong performance across various vision tasks. However, it still remains challenging to apply VFMs for cross-domain few-shot segmentation (CD-FSS), which segments objects of novel classes under domain shifts using only a few labeled exemplars. The challenge is mainly driven by two factors: (1) limited labeled exemplars per novel class relative to the scale of VFM pre-training, making the model prone to overfitting during retraining, and (2) target-domain shifts underrepresented during pre-training, inducing cross-domain inconsistency and layer-wise sensitivity. To address these issues, we propose Hierarchical Exemplar Representation Adaptation (HERA), a three-stage select-regularize-calibrate VFM-based segmentation framework that learns effectively from limited labels and adapts to novel domains without source-data retraining. We first design Hierarchical Layer Selection (HLS) to adaptively identify the most informative VFM layer using a data-dependent Exemplar Transfer Risk (ETR) computed for each candidate layer. Then, Prior-Guided Regularization (PGR) regularizes interactions on the selected representation, yielding well-structured local signals for the subsequent stage. Furthermore, Pixelwise Adaptive Calibration (PAC) combines the selected representation with the refined interaction maps to calibrate pixel-wise predictions, producing consistent masks. Together, these stages form a hierarchical select-regularize-calibrate pipeline that guides frozen VFM features in new domains while fine-tuning less than 2.7% of parameters at test time. Extensive experiments show that HERA surpasses the state of the art by more than 4.1 mIoU across multiple CD-FSS benchmarks.

Monocular SLAM historically suffers from scale ambiguity and tracking failure in dynamic environments. While recent vision foundation models (VFMs) provide remarkable zero-shot depth priors, naively integrating these deterministic predictions ignores predictive uncertainty and frame-to-frame scale inconsistencies. We propose PRISM-SLAM, a real-time framework that rigorously integrates VFM priors into a structured Bayesian factor graph to achieve scale-aware, metric-consistent localization and mapping. Specifically, we introduce a Plücker Ray-Distance Factor to anchor monocular observations in absolute space within a globally consistent metric coordinate system, mathematically resolving scale drift by making the metric scale Fisher-identifiable. To handle environmental dynamics, we derive an epistemic uncertainty proxy from temporal depth consistency and formulate a Dynamic Scene Uncertainty Gating (DSUG) mechanism. This soft-gating approach probabilistically down-weights dynamic distractors without incurring the heavy computational overhead associated with traditional semantic segmentation masks. By employing a multi-process architecture that asynchronously processes VFM inference and geometric tracking, PRISM-SLAM provides verified metric output at 30 FPS using solely RGB input, bridging the gap between foundation models and real-world robotic applications. Evaluated on the TUM RGB-D and 7-Scenes benchmarks, PRISM-SLAM achieves a metric $SE(3)$ Absolute Trajectory Error (ATE) nearly identical to its oracle-aligned $Sim(3)$ error. This demonstrates that our system can produce deployment-ready metric trajectories by delivering robust metric SLAM solutions without any post-hoc scale correction. Project page: https://prismslam-cmd.github.io/prismslam_pr/

Vision foundation models have demonstrated strong transferability across diverse visual recognition tasks and are increasingly considered for biometric applications. Their suitability for iris Presentation Attack Detection (PAD), particularly under realistic open-set operating conditions, remains insufficiently examined. This work presents a systematic failure analysis of general-purpose vision foundation models for open-set iris PAD using periocular imagery. Five representative foundation models are evaluated under three open-set protocols that explicitly separate different sources of distribution shift: unseen Presentation Attack Instruments (PAIs), unseen datasets captured with different sensors and cross-spectral transfer from near-infrared (NIR) to visible spectrum (VIS) imagery. Both frozen feature representations and parameter-efficient task adaptation using Low-Rank Adaptation (LoRA) are assessed within a unified experimental framework. The results indicate that foundation models can transfer across datasets with similar sensing characteristics, but fail to generalise reliably to unseen attack instruments and degrade sharply under cross-spectral evaluation. While LoRA improves performance in certain cross-dataset settings, it frequently amplifies failure under attack-level and spectral shifts. Additional validation experiments using segmented iris inputs, full backbone fine-tuning, joint cross-dataset and cross-PAI shifts, and reverse VIS to NIR transfer further confirm that these failures are not simply artefacts of periocular input, weak adaptation, or one-directional spectral evaluation. These findings show that strong closed-set or cross-dataset performance should not be treated as evidence of robust open-set security, and highlight the need for PAD representations that maintain sensitivity to presentation artefacts while remaining stable under realistic deployment variation.