Vision Foundation Models

Vision-based automation is an excellent candidate for reducing manual labor in greenhouse crop production and phenotyping. However, progress is constrained by the lack of annotated training data. Recent advances in vision-based foundational models have shown promising results in zero-shot generalization to novel domains, but their performance drops in complex agricultural environments. In this work, we present a sim-to-real framework for tomato plant segmentation that combines synthetic data generation with fine-tuning of a foundation model. We model a commercial cherry tomato greenhouse and use it to generate a large-scale synthetic dataset under diverse viewpoints, lighting conditions, and plant morphology. Subsequently, we fine-tune the Segment Anything Model 3 (SAM 3) on the synthetic dataset, specializing its text-conditioned segmentation behavior for greenhouse crop organs while retaining the general visual prior that makes zero-shot transfer possible. By evaluating our framework on multiple real-world greenhouse datasets, we demonstrate that combining synthetic data with SAM 3 fine-tuning significantly improves segmentation performance and model confidence. To support community benchmarking, we publicly release the procedural model, the generated synthetic dataset, and our fine-tuned SAM 3 weights.

Benchmarks for Geospatial Foundation Models (GFMs) increasingly rank models by aggregate score, but such rankings obscure why models differ: how much of the gap is architecture, how much is decoder capacity, and how much is a use-case-specific artefact? This study addresses that gap through a controlled comparison of two GFMs developed under European Space Agency's $Φ$-lab with contrasting design philosophies: THOR, which introduces a compute-adaptive architecture supporting variable patch sizes and unifies Sentinel-1, -2, and -3 data at their native resolutions; and TerraMind, a multimodal generative GFM pretrained with a dual-scale token/pixel objective that enables any-to-any cross-modal generation (Thinking-in-Modalities) to infer missing sensors at inference time. Rather than reporting a single leaderboard, we investigate the axes along which the two architectures actually differ - patch size, decoder complexity, finetuning regime, input modality, and model scale - across ten use cases spanning segmentation and regression in diverse domains, including climate disaster response, methane leak detection, snow monitoring, or sea ice mapping. We find that architectural design choices - patch size and decoder type in particular - explain more performance variance than model identity itself, that the two models embody complementary investment strategies (pretraining-time scale for TerraMind versus inference-time tokenisation for THOR), and that correctly interpreting results requires dataset-level characterisation. The resulting picture is not a single winner but a set of hypotheses and a diagnostic ablation methodology that we expect to generalise to future GFMs beyond THOR and TerraMind.

Recent geometric foundation models (e.g., Metric3D, Depth Anything and UniDepth) have substantially improved monocular depth estimation (MDE) in both cross-scene generalization and metric-scale prediction, yet these gains have not translated to tiny models. We bridge this gap with DepthART (Depth Anything Rethought for Tiny Models), which is a compact MDE model for on-device deployment across diverse scenes. We first identify two capacity-driven bottlenecks in tiny models: (i) overfitting to dataset-specific distribution bias and (ii) unstable metric adaptation under camera shift, where full fine-tuning easily damages transferable geometry. Accordingly, DepthART combines two simple but effective strategies: a bias-resistant data sampling scheme to reduce distribution bias under the same training budget, and a camera-conditioned fine-tuning protocol that freezes the distilled encoder and adjusts metric scale conditioned on intrinsics while better preserving cross-dataset generalization. Across datasets, DepthART consistently surpasses previous tiny baselines in both zero-shot generalization and metric accuracy (e.g., zero-shot $δ_1$=0.964 for DepthART-S on NYUD v2), and in some cases approaches heavy models. We further provide a scalable model family, with DepthART-S reaching 347/245 FPS (strict FP32) on an RTX A6000 at $224^2/448^2$, 102 FPS (TF32) on a Orin NX 8GB, and over 15 FPS (FP32) on a Jetson Nano 4GB.

This paper studies how to adapt a computer vision object detector to an unknown environment under both a robot navigation time and annotation budget constraint. Our approach selects informative robot trajectories and image samples to retrain the detector, explicitly targeting its failure cases. Formally, the approach is an embodied variant of batch active learning, where at each round an agent has a limited navigation budget to collect candidate samples and a limited annotation budget for the most relevant images. We leverage spatial consistency to identify images with inconsistent labels, which are likely to provide the greatest improvement to the vision model. We evaluate the approach using different active learning objectives on large scenes from the AI2-THOR simulator and on a real-world setup using a Boston Dynamics Spot robot with the real-time object detector YOLOv5. Through comparison against several baselines, our experimental results show that spatial inconsistency helps guide the agent and select relevant images without external supervision, achieving the highest detection accuracy at the end of the adaptation process under the same budget. The open-source project can be found at https://mkabouri.github.io/embodied-active-learning-od

We describe a model of perceptual inference in primary visual cortex (V1) equivalent to a minimal diffusion model whose function can be readily understood from its parameters. The model is based on sparse coding with a non-factorial prior over latent variables in the form of an unconstrained, pairwise interaction matrix, extending standard sparse coding inference to a general recurrent dynamical system. We efficiently train these recurrent dynamics using a denoising score-matching objective and implicit differentiation. After training on natural images, the learned interaction matrix mirrors the structure of horizontal connections in superficial layers of V1 that link neurons of similar orientation tuning. This model exhibits exceptionally good denoising performance, restoring image features such as extended contours amid extreme visual ambiguity, nearly matching the behavior of standard, black-box diffusion architectures in generalization regime. Owing to the model's simplicity, the network's Jacobian can be decomposed directly in terms of the interaction matrix between latent variables, revealing mechanistically how the recurrent dynamics assign high probability over a continuous family of natural structural deformations. Intriguingly, within this circuit, a large fraction of latent variables learn to disconnect from visual input altogether, essentially forming a hierarchical representation that appears to enforce global consistency among image features. Together, the model and results bridge two distinct domains: for neuroscience, it generates concrete, testable hypotheses regarding functional connectivity in recurrent neural circuits during perceptual inference tasks; for machine learning, it elucidates the internal mechanisms learned by diffusion models that allow them to generate infinitely many novel images from a finite training set.

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

Mild Cognitive Impairment is a critical early stage of cognitive decline that frequently precedes Alzheimer's disease, yet its automated detection from neuropsychological drawing tests remains fundamentally constrained by data scarcity, class imbalance, and diagnostic ambiguity near clinical boundaries. Existing methodologies attempt to bypass these constraints using computationally expensive, fully fine-tuned hybrid architectures that relegate spatial explainability to a post-hoc approximation rather than an intrinsic model property. We propose a parameter-efficient framework utilizing frozen DINOv2-Small model adapted via three modality-specific learnable prompt tokens while Operating with 1.19 million trainable parameters, each token serves as a query in a shared cross-attention layer over the source image patch tokens. Crucially, spatial explainability is achieved directly through these attention maps; as a structural consequence of the architecture. Then task-conditioned embeddings fused via an attention module to quantify modality-level importance per subject. To handle boundary ambiguity, a MoCA-adapted focal loss introduced that integrates continuous cognitive scores into the training target, loss modulation, and adaptive sample weighting, strictly generalizing standard soft-label approaches. Under stratified five-fold cross-validation, the proposed architecture yields an MCI-class F1 of 0.641 and an AUC of 0.795, outperforming the computationally heavier ResViT baseline by 0.110 in MCI-class F1.

Relative pose estimation is a fundamental capability for collaborative perception and coordination in multi-robot systems. However, robots encountering each other in real-world environments often operate in short interaction windows and must operate under limited communication bandwidth with intermittent or missing visual overlap caused by occlusions or limited fields of view. Existing approaches typically rely on global reference frames, assume sustained view overlap, or incur prohibitive communication costs, thereby limiting their applicability to ephemeral collaborative perception. To address these challenges, we introduce communication-efficient relative pose estimation (CERPE), a system-level framework that coordinates vision foundation models to jointly estimate ego-motion and inter-robot relative pose. CERPE reduces unnecessary raw-observation exchange by using continuously shared fixed-size descriptors to gate event-triggered raw-image requests independently of pose estimation. Non-overlapping encounters are handled by propagating inter-robot relative poses through metrically scaled ego-motion, thus maintaining relative pose estimates even in the absence of visual overlap. Experiments in simulation and real-world robots show that CERPE improves 6-DoF relative pose estimation over selected baselines in ephemeral collaborative perception.

Do vision models see colors the way humans do? Existing evaluations of color representations usually compare them with geometric spaces such as CIELAB or with discrete color labels. These references capture perceptual distance or category membership, but not the graded way in which people organize colors. We evaluate color grounding against a fuzzy perceptual model with 86 graded categories fitted to human survey data. The framework can be applied to any image encoder and measures three complementary properties: category boundaries, category compactness, and graded alignment beyond what color geometry alone can explain. Across eleven Vision Transformer encoders, the category-level results are broadly similar, whereas graded alignment differs substantially. Masked Autoencoders achieve the strongest beyond-geometry alignment, with confidence intervals that do not overlap those of the other encoders. A layer-wise analysis further shows that masked reconstruction preserves this structure toward the output. On natural images, MAE represents surface color globally, while language-supervised models encode color more strongly in relation to the foreground object. These results show that human-like color grounding has several distinct aspects that should not be reduced to a single score.

LiDAR-based collaborative 3D perception in Vehicle-to-Everything (V2X) systems typically relies on fusing bird's-eye-view (BEV) features across agents. However, current BEV representations, typically extracted by LiDAR backbones trained from scratch, are geometry-dominated and lack general semantic priors, inherently limiting the efficacy of feature-level collaboration. Meanwhile, vision foundation models (VFMs) pretrained on large-scale image data have demonstrated strong capability in learning general-purpose and informative visual representations for 2D tasks, and have the potential to enhance agent-wise LiDAR BEV representations for collaboration. Despite this potential, adapting VFMs to LiDAR-based 3D detection remains challenging due to the substantial image-point cloud modality gap. To bridge this gap, we propose ViCo3D, a collaborative 3D object detection framework powered by VFMs. Specifically, ViCo3D adapts VFMs to LiDAR-based collaborative perception from three aspects: First, ViCo3D projects point clouds onto the BEV plane as three-channel images, enabling DINOv2 to extract BEV-space visual features from LiDAR inputs. Besides, to effectively integrate these DINOv2-derived features with LiDAR geometric features, ViCo3D introduces a multi-scale BEV fusion module within the single-agent encoder. In addition, ViCo3D adopts an ego-centric cross-agent fusion strategy to aggregate complementary information from multiple agents. Experiments on DAIR-V2X and V2XSet demonstrate that ViCo3D achieves state-of-the-art 3D detection performance. Remarkably, it delivers up to 1.8x greater collaborative gains than prior methods on DAIR-V2X. The code will be made public available for future investigation.

This work introduces a unified formulation for vision models, where diverse forms of visual information beyond natural images, such as masks, depth maps, and other structured visual signals, are all represented as RGB images, while general visual tasks can be converted into a common RGB-to-RGB image editing problem. In this paradigm, different types of visual information internally share the same encoding and decoding architecture and parameters as natural images, enabling a single model to transfer across tasks through a unified visual interface, in a way analogous to how language models operate over text. We refer to this formulation as RGB In and RGB Out (RINO). Built upon a generic image editing backbone without task-specific fine-tuning, RINO demonstrates robust and competitive zero-shot performance on both dense understanding tasks such as segmentation and depth estimation (where we unify outputs as RGB), and dense-conditioned generation tasks such as pose-to-image generation (where we unify inputs as RGB). We hope this study provides useful insights toward general unified vision-language systems, where diverse visual tasks can be expressed, interpreted, and solved through a shared visual language. Code is available at https://github.com/yangtiming/RINO.

Crack segmentation is essential for infrastructure inspection and structural health assessment, but existing high-performance methods typically require task-specific pixel-level annotations and training. Text-promptable vision foundation models enable zero-shot deployment, yet their final mask proposals are poorly suited to thin, fragmented, and low-contrast cracks, whose evidence may be suppressed, truncated, or over-expanded during mask generation. We find that language-conditioned semantic responses within the SAM3 decoder preserve more continuous and complete crack evidence than its final masks. Based on this observation, we propose Semantic-Edge Response Decoding (SERD), which interprets internal responses as a dense crack-likelihood field, calibrates them with a lightweight edge prior, and generates crack masks using a unified global threshold, without annotation or fine-tuning. Experiments on six public datasets show that SERD consistently improves over native SAM3 and outperforms the compared zero-shot and open-vocabulary segmentation methods, achieving an average Crack IoU of 61.14%, 4.63 points higher than SAM3. Further analyses show that most gains arise from directly decoding internal semantic responses, while edge calibration improves structural recovery and false-positive control without increasing end-to-end inference overhead. These results suggest that, for thin and non-compact targets, internal continuous responses can provide a more transferable interface than the final masks of foundation models. Code is available at: https://github.com/xauat-liushipeng/SERD

The analysis of satellite and aerial imagery has entered a new era with the advent of foundation models. This paper describes the concept of Geospatial Foundation Models (GeoFMs), which are artificial intelligence/machine learning (AI/ML) models pre-trained on massive geospatial datasets through varied methodologies. We first articulate the core paradigm shift that GeoFMs enable: a separation of duties, where large-scale model providers perform the computationally intensive pretraining, allowing domain experts to rapidly fine-tune or prompt these models for specific, mission-critical tasks. This approach democratizes access to state-of-the-art AI/ML while maintaining the security and confidentiality of the downstream task. We then explore the novel capabilities unlocked by different types of GeoFMs, distinguishing between the finetunable vision models produced by self-supervised techniques like masked auto-encoding, and the vision-language models produced by contrastive learning which enable zero-shot tasks like open-vocabulary image analysis. Next, we discuss the practical considerations for operationalizing GeoFMs, from performance-cost analysis to the broader MLOps ecosystem. To that end, we introduce a taxonomy of model adaptation strategies and propose a framework for domain experts to select the most cost-effective adaptation approach for their particular mission set. Finally, we present a forward-looking vision of Agentic Geospatial Reasoning, where Large Language Models act as intelligent orchestrators, leveraging GeoFMs as tools to answer high-level user queries in natural language and automate complex analytical workflows, moving the field from perception to cognition.

Mainstream visual encoders are pretrained on natural images and cannot be effectively applied to document images without document-oriented adaptation, as dense text and fine-grained character strokes demand character-level visual perception. We present MonkeyOCRv2, a visual-text pretrained model for document AI. First, we construct MonkeyDoc v2, to our knowledge the largest document-image pretraining corpus, comprising 113 million images spanning 17 languages. Second, we propose a pretraining strategy that jointly learns image-to-text generation and pixel-level document reconstruction: the former aligns visual representations with textual content, while the latter preserves character strokes and layout details. Extensive experiments are conducted on five representative document analysis tasks, including text recognition, formula recognition, text detection, document tampering detection, and overlapping text segmentation. Replacing the original encoders with MonkeyOCRv2 consistently improves performance across all five tasks. Finally, we validate its effectiveness as the vision encoder of multimodal large language models on the more challenging tasks of document parsing and document understanding. Kept frozen and paired with a lightweight language model, it yields a 0.7B document parsing model that sets a new open-source state-of-the-art on MDPBench, a recent benchmark spanning digital-born and photographed documents across 17 languages, surpassing the previous best 3B dots.mocr by 2.8% absolute with a vision encoder roughly 11$\times$ smaller. The frozen encoder also powers a document understanding model that outperforms counterparts built on CLIP, DINO, and SAM across eight benchmarks under identical training settings. These results suggest that document-oriented visual pretraining can serve as a foundation for document intelligence in its own right.

Foundation models such as CLIP have enabled open-vocabulary object detectors that generalise to novel categories via vision-language similarity. However, the confidence scores these detectors produce are not reliable localization probability estimates: they conflate visual scale and semantic query specificity with the true detection signal. Through controlled experiments on COCO across three foundation-model-based detectors (GroundingDINO, OWL-ViT, YOLO-World), with the scale-bias finding further replicated on LVIS (1,203 categories) using GroundingDINO, we show that s=cos(v,t) is a biased mixture of two effects. Scale bias (alpha = +0.064, r = 0.579, p = 1.29 x 10^-58) systematically inflates scores for large objects. Semantic bias (beta = -0.705, p = 5.23 x 10^-41) suppresses scores for generic queries. Both biases are structurally inevitable from CLIP's image-level pretraining. Threshold adjustment cannot remove them: oracle per-scale thresholding yields Delta F1 = +0.001 for small objects versus +0.102 for large. A parameter-free temperature scaling correction improves small-object Recall@10 by 19.6% (p < 0.01) without retraining. This comes at a modest, measurable cost to pooled-ranking precision, so the bias is partially, not freely, reversible at inference time. These findings reveal a fundamental limitation of adapting image-level foundation models to region-level detection tasks.

Cross-modal distillation from Vision Foundation Models (VFMs) to LiDAR backbones has recently emerged as a self-supervised pretraining strategy that reduces reliance on dense point-wise annotation for 3D scene understanding. However, existing distillation pipelines typically treat the VFM as a frozen feature source and train a heterogeneous 3D backbone to match fixed image embeddings, forcing the student to bridge both the modality gap and the cross-architecture gap between dense ViT token representations and sparse 3D encoders. We propose TOLiD, a self-supervised pretraining method for LiDAR representation learning that addresses this gap by coupling a LiDAR backbone with a student Vision Transformer (ViT) initialized from a frozen VFM teacher and applying supervision over compatible patch-token representations. TOLiD converts the set of point features within each image patch frustum into a token using Frustum Pooling followed by Frustum Attention, and performs token-level distillation with visibility masking. For LiDAR-only deployment, we lift token features back to per-point representations using masked bilinear sampling to avoid patches that have limited LiDAR points. We extensively evaluate TOLiD on five heterogeneous LiDAR datasets and four cross-sensor adaptation pairs, demonstrating improved transfer with frozen backbones and lightweight heads.

Foundation models are increasingly used as image feature extractors for mammography, but their robustness under external domain shift remains unclear. We benchmark 15 foundation-model backbones across breast density, BI-RADS severity, and cancer status using a unified frozen-backbone linear-probe protocol, training on 3 source datasets and evaluating on 12 task-compatible out-of-distribution (OOD) datasets after label harmonization. Mammography-specific vision-language models (Mammo-FM and MaMA) provide the strongest mean OOD performance, but robustness is not explained by mammography exposure alone. DINOv3 remains a competitive vision-only baseline, and mammography-adapted pretraining does not consistently improve generalization. Dataset-level analysis further shows that even leading models show heterogeneous performance across datasets. Feature-space inspection reveals that useful representations can preserve clinical signal while retaining dataset and acquisition structure. These findings highlight dataset-level OOD evaluation as a central criterion for assessing mammography representations. Our code is publicly available: https://github.com/biomedia-mira/mammo-ood.

The rapid progress of large foundation models has been driven predominantly by pretraining on large-scale text corpora. However, many forms of knowledge are conveyed through visual representations, where figures, typeset equations, and page layouts carry rich information that cannot be faithfully or completely captured by text alone. Yet current pretraining approaches discard these visual cues by converting visually rich sources, such as documents and web pages, into plain text for learning language intelligence. This paper challenges the default assumption that language models must be trained on text-only representations and shows that Visual Pretraining is a scalable learner for foundation model intelligence. To this end, we conduct a systematic study of unsupervised visual pretraining paradigms that directly leverage visual documents without text extraction. Across multiple backbones and benchmarks, visual pretraining on the same underlying corpora consistently outperforms text-only pretraining, offering an efficient pathway to scalable language intelligence.

The deployment of large-scale foundation models, such as the Segment Anything Model 3 (SAM 3), promises a transition toward open-vocabulary, training-free computer vision. However, their capacity to generalize out-of-distribution to the complex, top-down geometric structures of Earth Observation imagery remains largely unquantified. Driven by SAM 3's performance disparities in highly specialized domains, we present a comprehensive, multi-task empirical evaluation across remote sensing scene classification, object detection, and instance segmentation under strict zero-shot and one-shot constraints. To achieve this, we introduce a structural adaptation of SAM 3 by repurposing its decoupled binary presence head into a standalone zero-shot classifier. Furthermore, by systematically isolating textual and visual prompt modalities across five configurations, we explicitly diagnose the alignment mechanics within the model's multimodal decoder. Our findings reveal severe cross-modal interference: while visual prompts successfully align the decoder to complex remote sensing geometry, textual prompts inject misaligned, ground-level semantic bias, actively degrading coordinate regression. To benchmark these capabilities without resource-intensive training, we formulate a novel training-free proxy evaluation protocol for Generalized Zero-Shot tasks (scene classification and instance segmentation). Ultimately, our results demonstrate that SAM 3 avoids the overfitting commonly seen in legacy domain-adapted models, achieving high Harmonic Mean scores in segmentation tasks. However, it remains fundamentally constrained by sub-pixel resolution limits and overhead semantic blind spots, charting a definitive mandate for parameter-efficient geospatial fine-tuning of its multimodal decoder.

Foundation models are reshaping computational pathology, yet their capabilities remain shaped by pretraining objectives, data sources, and spatial scales, fragmenting complementary expertise across separate backbones. Here we present ALICE, a unified foundation model trained through multi-stage agglomerative distillation that sequentially distills eight vision-only, vision-language, and slide-level teacher models into dedicated modules of a single backbone. ALICE is pretrained on 24,985,184 tile-level pathology images and 155,604 high-resolution images, and evaluated across 21 task scenarios, 96 downstream tasks, and 48 data sources, spanning region-of-interest tissue analysis, vision-language multimodal evaluation, and whole-slide clinical assessment. In all three evaluation settings, ALICE achieved the best average rank among task-matched pathology foundation models. These results demonstrate that agglomerative distillation can consolidate complementary capabilities from specialized models into a unified backbone for broad computational pathology applications. The model is available at https://github.com/WonderLandxD/ALICE.

A fundamental challenge in 3D reconstruction and robotic localization is co-visibility: determining which image pairs share overlapping visible surfaces, particularly in scenarios with minimal overlap. We demonstrate that VGGT implicitly encodes co-visibility as an emergent behavior: without any supervision for this task, its internal representations exhibit a clear hierarchical structure mirroring that of large language models, i.e. early layers build a 3D-aware scene representation, while late layers act as dedicated co-visibility reasoners. In particular, we identify layer L17 as a negative anchor that consistently routes non-co-visible pairs for this backbone, regardless of the evaluation setting, providing task-grounded evidence of layer specialization in a geometry-grounded foundation model. Building on this, we introduce Co-VGGT, which freezes VGGT and trains only a lightweight layer-wise mixture-of-experts head (less than 7.5M parameters) to classify co-visibility from RGB alone, treating each layer as a specialized expert whose geometric abstraction is adaptively weighted per input pair. On the Co-VisiON benchmark, Co-VGGT surpasses the human annotation baseline and improves over prior work by more than 25% pairwise and 10% multiview. Pairwise predictions are well-calibrated (ECE=0.030), enabling direct use as edge weights in visibility graphs for downstream SfM and SLAM pipelines without post-hoc correction. Code and data are available.

A striking feature of the human visual system is that it ingests visual information through a series of local foveated glimpses, rather than a single global computation. This makes human vision distinctly different from most popular computer vision models in use today, which input images globally and in a single shot. A natural question therefore is whether local, sequential vision models may provide any fundamental computational benefits in addition to being biologically more plausible than global models. In this work, we investigate this question from the perspective of visual state tracking and length generalization. Inspired by recent studies of length generalization in language models, we study the behavior of vision models trained on simple vision tasks that require the aggregation of local information across an image. Our experiments reveal that, similar to language models, vision models can learn to exploit global shortcuts and thereby fail to generalize over task length or complexity. We also show that recurrent vision policies based on strictly local perception can mitigate these failures, thereby allowing models to generalize on these tasks. Our results show that local attention may be an essential overlooked requirement for robust compositional generalization.

Driven by next-token prediction, NLP shifted from task-specific models into powerful generalist foundation models. What, then, is the equivalent catalyst needed to achieve a general-purpose model in computer vision? In this paper, we contend that large-scale text-to-video generation serves as a strong pre-training paradigm for computer vision, providing the necessary spatiotemporal priors, vision-language alignment, and scalability required for general visual intelligence. We introduce GenCeption, which leverages a pre-trained video generative diffusion backbone to define a feed-forward perception model, capable of performing various vision tasks steered by text instructions. Empirical results demonstrate that GenCeption achieves state-of-the-art performance across a diverse suite of tasks, including depth, surface normal, and camera pose estimation, expression-referring segmentation, and 3D keypoint prediction, often matching or surpassing specialized models (e.g. DepthAnything3, SAM3, D4RT, VGGT-Omega, Sapiens, David, Genmo, and Lotus-2). Furthermore, the video generative pretrained backbone outperforms alternative pretraining paradigms (e.g., V-JEPA, and Video MAE) under comparable settings. Importantly, GenCeption exhibits preliminary data and model scaling properties along with exceptional data efficiency, where it achieves comparable performance with leading models like D4RT and VGGT-Omega with 7 to 500 less training data. Finally, GenCeption also exhibits intriguing emergent behaviors: a model trained exclusively on synthetic human videos generalizes to real-world footage and out-of-distribution object categories (e.g., animals and robots). These findings suggest that video generation is not merely a synthesis tool, but a foundational path toward generalist vision intelligence for the physical world. Project page: https://genception.github.io

Monocular depth estimation has seen remarkable progress through foundation models achieving robust zero-shot generalization, yet their computational demands place them far beyond the reach of embedded and mobile platforms. Lightweight alternatives exist, but have been developed almost exclusively within single-domain, self-supervised paradigms, failing silently under domain shift. We present ZipDepth, a compact monocular depth network that bridges this gap by combining an efficient reparameterizable encoder-decoder with large-scale knowledge distillation from a foundation model over a large multi-domain training set. Comprising just 6.1M parameters, ZipDepth runs at real-time rates from server GPUs to power-constrained devices, achieving the best trade-off between zero-shot accuracy and deployment efficiency among lightweight models across five benchmarks, taking a significant step towards the accuracy of foundation models with 50x more parameters.

In computational vision science, Convolutional Neural Networks (CNNs) have emerged as a popular model of biological vision because of the alignment they can exhibit with neural and behavioral data in humans and animals. However, it remains unclear to what extent this alignment persists for visual tasks that extend beyond the canonical object recognition paradigm based on well defined semantic content. In this study, we diverge from the common object-centric view by focusing on another aspect of vision: texture perception. We consider textures of different complexity generated with three different algorithms from the same source images. Using a rank-based statistic, we quantify the information encoded in the internal representations of a CNN and three Vision Transformers (ViTs), and we compare the similarity of these representations to those inferred from human psychophysics data. We find that the representation of textures is aligned in different ViTs, but not between the ViTs and the CNN; that ViTs form similar representations for textures of different complexity; that human performance in recognizing textures can be better predicted from ViTs representations rather than CNN representations. Taken together, these results suggest that ViTs may capture more faithfully than CNNs how texture patterns are visually processed by humans, and that the representations of texture stimuli in computational models may be driven by the network architecture.

Foundation models (FMs) have transformed machine learning from isolated task-specific model development toward general-purpose models pretrained on broad data and adapted to multiple downstream tasks. Earth observation (EO) is an important domain for this paradigm because satellite and airborne archives are large, high-revisit, and increasingly multimodal, while reliable field labels are often sparse. Remote sensing foundation models (RSFMs) cannot be transferred reliably/optimally without domain-specific adaptation. This is because EO data are governed by measurement physics and operational decision constraints. This chapter reviews the design principles arising from these domain-specific constraints. It first defines the FMs paradigm in remote sensing (RS), then synthesizes the current model landscape, pretraining objectives, architecture designs, downstream adaptation and trustworthiness requirements. The chapter also incorporates recent benchmark evidence showing that no single geospatial foundation model is universally best and that inconsistent evaluation remains a major issue to fair comparison and reliable deployment. In addition, two brief environmental monitoring case studies; physics-informed spectral targeted masking for harmful algal bloom prediction and reinforcement learning for adaptive environmental monitoring station selection to illustrate the FMs domain-guided principles in practice. This chapter posits that next-generation RSFMs should be evaluated not only by benchmark accuracy, but also by modality-aware transfer and physically plausible representations for trustworthy EO decisions.

Traditionally, continual learning has assumed access to labeled data, yet many real-world applications -- such as lifelong robotics -- require models to adapt continuously from unlabeled streams. This has led to the development of continual self-supervised learning (CSSL), a rapidly growing area that lacks a dedicated, systematic review. In this work, we present a comprehensive survey of CSSL for vision, with connections to emerging vision-language settings. First, we analyze existing evaluation protocols and highlight inconsistencies that hinder fair comparison. We then examine why self-supervised objectives exhibit improved robustness to catastrophic forgetting, relating this to task-agnostic representations and smoother loss landscapes. Next, we organize existing methods into a unified taxonomy based on their forgetting-mitigation strategies, including distillation, replay, regularization, architectural approaches, model merging, and objective-level adaptation. Finally, we identify open challenges such as scalability and the need for fast adaptability. We argue that advancing CSSL requires moving beyond small-scale benchmarks towards continual pre-training paradigms for large-scale systems.

Vision foundation models (VFMs) are increasingly being developed for radiological imaging, yet their definition, development and evaluation remain heterogeneous. We conducted a PRISMAScR scoping review of peer-reviewed studies published between January 2017 and March 2026 describing foundation models trained exclusively on radiological imaging data. Sixty-seven studies were included and mapped across three pillars: data scale and heterogeneity, architectural and pretraining scalability, and downstream transferability and generalization. Datasets primarily covered brain MRI, thoracoabdominal CT, and chest X-ray, ranging from fewer than 100,000 samples to multi-million-image cohorts. Transformer-based architectures and self-supervised pretraining predominated, particularly masked image modeling, contrastive learning and multi-stage approaches. Evaluation focused mainly on segmentation and classification, whereas cross-center, cross-scanner, anatomical and modality-shift validation was inconsistently reported. Alignment with FUTURE-AI principles was uneven. Overall, radiology-specific VFMs show promising transferability, but clinical translation remains constrained by limited data representativeness, heterogeneous benchmarks, incomplete reporting and insufficient deployment-oriented evaluation.

Pre-trained Vision Foundation Models (VFMs) provide strong visual representations for diverse downstream tasks. The key challenge of VFM adaptation stems from the prohibitive costs of full fine-tuning and catastrophic forgetting. To address this, Low-Rank Adaptation (LoRA) has emerged as the prevailing paradigm for Parameter-Efficient Fine-Tuning (PEFT). However, LoRA is typically designed for transformer self-attention layers parameterized by 2D matrices. Since convolutional kernels inherently couple spatial and channel information within a 4D tensor, forcing them into a monolithic 2D matrix disrupts the inherent spatial topology. In this paper, we propose Low-Rank Convolutional Adaptation (LoCA), a convolution-aware PEFT framework that addresses spatial-channel entanglement by decoupling channel and spatial adaptation. LoCA introduces a low-rank channel adaptation for dense cross-channel mixing and refines spatial bases extracted from pre-trained kernels via Singular Value Decomposition (SVD). Experimental results show that LoCA preserves pre-trained spatial priors and achieves competitive or state-of-the-art performance across fine-grained classification, domain-generalized semantic segmentation, and generative benchmarks.

Recent progress in large-scale self-supervised learning has improved dense geometric prediction, but it remains unclear whether such scaling yields inference-time physical plausibility checks. We propose Scrambled Edges, a controlled counterfactual that injects salient edge-like cues while violating surface continuity, illumination coherence, and occlusion ordering. With energy-matched and structure-matched controls, we isolate the effect of unsupported edge evidence from high-frequency energy and edge sparsity. Across CNN/ViT/SSL depth predictors on NYU Depth v2 and KITTI, Scrambled Edges induce up to 3.2x larger deviation from clean predictions than energy-matched noise; additional diffusion and flow-matching depth estimators show attenuated but still significant collapse. The resulting Geometric Collapse propagates globally: even with oracle knowledge of the corrupted region, output-level repair recovers only 47%, with substantial error outside the mask. These findings provide controlled behavioral evidence that current dense predictors lack reliable mechanisms to quarantine physically unsupported edge cues, motivating explicit plausibility scoring and selective cue integration.

Rendering time series as chart images for CNN-based classification has become increasingly common in time-series classification (TSC). However, it remains unclear whether models learn underlying temporal patterns or rely on encoding-specific visual cues introduced by chart design. We present VEIL: a systematic study examining how chart encodings influence learned representations through complementary analyses of similarity, transferability, and attribution. Attention-guided training appears to mitigate this effect when encoding sensitivity is consistently identified across diagnostics, but provides limited or negative benefit when such signals are absent. These findings position VEIL within the broader question of how machines perceive visualizations -- extending graphical perception from human readers to vision models -- and show that visualization design choices shape learned representations in ways that warrant treating chart-based TSC as a representation and measurement problem rather than a simple modeling decision.

Dense spatial perception is essential for physical intelligence, where visual systems are expected to recover structured, metric, and actionable representations from pixel observations. Modern visual foundation models tend to prioritize semantic invariance, often at the expense of detailed spatial understanding. In this work, we study vision pretraining through a boundary-centric lens, motivated by the premise that boundaries and shape discontinuities offer essential cues for perceiving geometric properties. Concretely, we propose masked boundary modeling, a self-supervised paradigm that dynamically learns sub-pixel boundary representations and subsequently leverages the discovered boundary-bearing tokens as masked targets to facilitate dense visual token learning. By scaling this framework, we develop LingBot-Vision and demonstrate its efficacy across a diverse set of downstream vision tasks with DINOv3 as a strong baseline. Remarkably, LingBot-Vision drives the progression from LingBot-Depth 1.0 to LingBot-Depth 2.0 for depth completion, and thereby yields enhanced depth estimation, a key pillar for embodied artificial intelligence. Our findings reveal that boundary modeling goes beyond simple line segments and instead serves as a scalable pretraining principle for learning spatially structured visual representations.

Test-Time Domain Adaptation (TTDA) aims to adapt Deep Neural Networks to distribution shifts using only streaming, unlabeled test data in real time. Current methods for semantic segmentation tasks suffer from critical limitations. Entropy minimization techniques require costly backpropagation, risking catastrophic forgetting and producing noisy segmentation boundaries. Memory-bank methods, while backpropagation-free, exhibit slow adaptation, requiring numerous samples to converge and struggle to handle continuous domain shifts. We introduce TestMate, a novel, real-time, and backpropagation-free TTDA framework that overcomes these issues. TestMate leverages generalization capability of a lightweight Visual Foundation Model to guide the adaptation. We use a zero-shot instance segmentation YOLOv8-seg based model to generate unlabeled mask proposals for objects and their parts at multiple scales in real time. These proposals are fused with the primary model via a heuristic, size-ordered competitive scheme, where small, high-confidence regions dominate and refine predictions in surrounding larger, less certain areas. This paremeter-free mechanism enables immediate adaptation from the first frame, inherently avoids catastrophic forgetting and effectively preserves fine object details and boundaries, even for small objects. TestMate can be used as a standalone, efficient refinement module or seamlessly integrated into existing TTDA methods to significantly boost their performance. We demonstrate state-of-the-art results across two benchmark datasets, proving TestMate's effectiveness in three distinct adaptation tasks: TTDA, Source-Free Domain Adaptation (SFDA), and online-TTDA. Code is available.

The deployment of large-scale foundation models like Segment Anything Model (SAM) on resource-constrained Earth observation platforms is hindered by prohibitive computational costs and the domain shift between natural and remote sensing imagery. To address these challenges, we propose \textit{Geo}spatial \textit{S}egment \textit{A}nything \textit{M}odel-Lite (GeoSAM-Lite), a lightweight, prompt-free segmentation framework designed for efficient onboard remote sensing segmentation. GeoSAM-Lite incorporates two core innovations: (1) Geospatial-Domain Initialization (Geo-Init), a domain-aware pre-training strategy that distills geospatial priors from a specialized teacher to bridge the domain gap; and (2) Feature Fusion Layers (FFL), which recalibrate spatial features and restore high-frequency boundary cues to overcome the capacity bottlenecks of lightweight backbones. Experiments across representative datasets, with a primary focus on cloud scenarios to evaluate performance under extreme scale variations and complex boundaries, demonstrate that GeoSAM-Lite achieves competitive accuracy while reducing parameters by 92.8% compared to the heavyweight RSAM-Seg. By establishing a superior Pareto frontier between efficiency and fidelity, GeoSAM-Lite offers a practical solution for real-time segmentation on edge devices.

This paper challenges the assumption that vision-language models (VLMs) require fixed patch-based 2D vision features. Analyzing fine-tuned vision encoders, we find that representations become increasingly abstract and less spatially coherent during VLM training. Notably, models trained with image-text alignment (such as SigLIP2) develop a small number of specialized tokens that effectively summarize global image content. Building on this, we introduce RADIO1D, which compresses images into a compact, variable-length 1D token sequence using multi-teacher knowledge distillation and an autoencoder design. The resulting representations exhibit strong hierarchical summarization, enabling accurate scene understanding - even with a single token - and support improved composition-aware image retrieval. In VLMs, RADIO1D provides flexible accuracy-efficiency tradeoffs through adjustable token counts, delivering competitive performance on diverse multimodal benchmarks with lower computational overhead and better accuracy.

Modern pretrained vision models achieve strong accuracy but demand substantial GPU memory for fine-tuning, making edge deployment impractical. This paper compares five parameter-efficient fine-tuning (PEFT) methods (Full FT, LoRA, AdaLoRA, QLoRA, BitFit) on Transformers- (ViT-Small, TinyViT) and Mamba-based vision backbones (Vim-Small, MambaVision-T) under an on-device VRAM budget (e.g., 2 GB), together with three gradient-checkpointing strategies (none, static, and a proposed memory-budget-aware adaptive algorithm); and we evaluate three families of foundation-model baselines: zero-shot contrastive vision language models (OpenCLIP, SigLIP), self-supervised vision backbones with lightweight evaluation protocols (DINOv2), and autoregressive VLMs for prompt-based classification (PaliGemma, MobileVLM, SmolVLM). Experiments on CIFAR-100 and DTD report accuracy, training time, energy, and the NetScore family of multi-objective metrics, which we extend with two deployment-aware variants. QLoRA and BitFit cut energy 20-30% at a 1-2% accuracy cost; the adaptive algorithm reduces peak memory 43-79% with 9-30% energy overhead. DINOv2 surpasses fine-tuned models on CIFAR-100 (0.917 vs. 0.897) at a fraction of the energy, while small autoregressive VLMs remain uncompetitive.

Foundation video diffusion models are increasingly viewed as world simulators for embodied agents, yet their pretraining on internet-scale generic video leaves them poorly aligned with real-world deployment domains. We study parameter-efficient adaptation of a pretrained foundation video world model to retail scenes: when synchronized egocentric and exocentric video of the same activity are available, which viewpoint of training data produces the strongest adapted model? We introduce RetailSMV (Retail Synchronized Multi-View), a corpus of 32,105 captioned retail clips from five supermarkets with synchronized ego/exo capture from the store-staff perspective (stocking, arranging, weighing, managing supply carts, scanning at checkout), rather than the customer-centric framing of prior retail video corpora, and train three matched Low-Rank Adaptation (LoRA) configurations of Cosmos3-Nano (egocentric-only, exocentric-only, combined) under identical hyperparameters. On a 200-clip held-out test set evaluated with seven complementary metrics under a strict paired statistical protocol, exocentric-only adaptation matches or exceeds combined adaptation on six of seven point estimates and is significantly better on LPIPS, PSNR, and DreamSim, despite training on only 15,985 exocentric clips (versus 32,105 for combined). A symmetric paired comparison further shows that adding exocentric data to egocentric-only training helps while adding egocentric data to exocentric-only training hurts. The absolute adaptation gap is largest at the shortest rollout time, identifying the near-horizon prediction window as the regime in which adaptation is most beneficial.

Algorithm development for radioisotope identification in mobile urban search scenarios face significant challenges from non-uniform backgrounds, momentary source encounters, and severe class imbalance between rare threat signatures and background measurements. We present a machine learning-based approach to this problem that converts list-mode gamma-ray data into two-dimensional waterfall spectrograms and applies computer vision architectures to the resulting images. Rather than treating waterfalls as conventional images, we employ a representation where consecutive time spectra can form input channels, similar to RGB channels in color images. This representation encodes both spectral and temporal information, enabling neural networks to more effectively learn patterns that distinguish source signatures from background fluctuations. We evaluate three architectures, a multilayer perceptron (MLP), convolutional neural network (CNN), and vision transformer (ViT), on the Radiological Anomaly Detection and Identification (RADAI) benchmark dataset. At a false positive rate of less than one false alarm per hour, our CNN outperforms the previous-best non-negative matrix factorization (NMF) method across all global metrics, achieving true detection, classification, and identification rates of 0.4334, 0.3965, and 0.2950 respectively, compared to 0.4151, 0.3611, and 0.2625 for NMF. At lower false positive rate constraints, the neural network approaches show comparable but ultimately lower performance than NMF, indicating opportunities for further research.

Foundation models have transformed vision and language processing by providing rich, reusable representations that transfer across diverse tasks. Sheet music, as a visual encoding of musical language, lacks such a strong domain-specific backbone. We introduce MuSViT (Music Score Vision Transformer): the first foundation vision model for sheet music representation -- a ViT encoder pre-trained via Masked Autoencoders on 9.7 million pages from the IMSLP. To handle the complexity of real-world scores, we adopt a two-stage curriculum: a synthetic warm-up on typeset scores followed by large-scale training on the full IMSLP corpus. We evaluate MuSViT on four downstream tasks -- full-page and staff-level music score recognition, music symbol detection, and score difficulty classification -- under two scenarios: linear probing (frozen encoder) and fine-tuning. Under linear probing, MuSViT consistently outperforms modern vision encoders, revealing that general-purpose representations, regardless of scale, fall systematically short on the structured symbolic properties of musical notation. Under fine-tuning, MuSViT generally improves upon task-specific state-of-the-art methods. An additional embedding-transcription consistency analysis reveals that MuSViT encodes symbolic musical structure directly in its representation space -- unlike other encoders, whose embeddings do not correlate with music notation content. These results establish MuSViT as a foundation backbone for sheet music understanding.

Foundation models are increasingly integrated into embodied intelligence systems, but directly assigning them structured prediction tasks requires precise geometric and numerical estimation, where specialized models often remain stronger. This capability mismatch raises a key question: should foundation models replace task-specific predictors, or should they collaborate through tasks better aligned with their strengths? We propose FAT, a foundation-model-augmented task-specific reasoning framework that treats collaboration as task decomposition rather than model replacement. FAT decomposes structured prediction into specialist prediction, information-space reconstruction, and foundation-model proxy reasoning. The specialist generates geometrically and physically valid hypotheses in the native output space, while the foundation model performs a bounded proxy task, such as selection or verification, over reconstructed multimodal candidates. We instantiate this principle as ProxySelect with a vision--language model. Across 2D object detection, 3D object detection, trajectory prediction, and semantic segmentation, ProxySelect consistently improves specialized baselines and substantially outperforms direct foundation-model regression at lower computational cost. These results suggest a general collaboration principle: specialized models preserve task-specific structure, while foundation models refine their hypotheses through contextual proxy reasoning.

Vision foundation models such as SAM 3 can provide transferable object-level structure across diverse surgical video conditions, but segmentation outputs do not explicitly encode the action-conditioned semantics that define functional surgical landmarks. Estimating instrument extent and geometry differs from localizing the tip or anchor relevant to clipping, grasping, or dissecting. We investigate vision foundation model-enabled sparse action-aware landmark localization, using zero-shot, point-prompted structural masks to provide dense instrument-level context without manual pixel-level mask annotations. We propose a lightweight refinement framework that uses SAM 3 as a structural prior. A coarse multi-frame network predicts tip and anchor prompts, generating non-oracle masks that are fused with visual and heatmap features to refine functional landmark predictions. We compare direct mask-augmented supervision, prediction-derived mask-prior refinement, and auxiliary mask supervision to examine how vision foundation model-derived structure should enter a precision-oriented localization system. Experiments on 7,867 clips from 60 surgical videos spanning YouTube, Cholec80, HeiChole, SurgVU, and CRCD evaluate the approach under heterogeneous conditions. Without manual pixel-level mask annotations for training, the proposed model achieves overall F1 scores of 72.4% for tip and 58.0% for anchor localization. Directly imposing masks on heatmap targets biases learning toward broad tool regions, whereas prediction-derived priors and auxiliary supervision provide effective intermediate structural guidance for action-dependent landmark prediction.

A faithful 3D world representation should account for layered geometry, where a single camera ray may contain multiple visible and geometrically valid surfaces. Monocular depth estimation, however, reduces this structure to one scalar depth per pixel. Transparent scenes make this ambiguity measurable: the same ray can pass through foreground glass and observe the background, turning the supervised target into a convention of annotation, data, and training rather than a scene-intrinsic truth. A learned predictor exposes this convention as its depth-layer preference. We introduce MultiDepth-3k (MD-3k), a sparse two-layer ordinal benchmark for measuring depth-layer preference and multi-layer spatial relationship accuracy (ML-SRA). On MD-3k, leading depth foundation models exhibit diverse layer preferences under standard RGB input, showing that the same layered geometry can be resolved differently across models. We further find that Laplacian Visual Prompting (LVP), a training-free spectral input transformation, can substantially change the reported layer for certain frozen models. The strongest RGB/LVP pair, DAv2-L, reaches 75.5% ML-SRA. These results suggest that depth foundation models may express complementary geometric hypotheses that standard RGB inference leaves unexpressed. We invite the community to rethink depth supervision and evaluation through an ambiguity-aware lens, where multiple valid 3D interpretations are treated as geometric structure to be measured, preserved, and expressed.

Can a vision model truly see an object, or does it only fit surface-level visual cues? Following Wittgenstein's view that the limits of language are the limits of the world, we view a model's recognition ability as bounded by the descriptive system it has learned. In current vision models, this system is often realized through learned feature representations that exploit local statistical cues. We therefore ask whether a model can still classify correctly when such local cues provide no stable basis for distinction. We formalize this question with syntactic distance, which measures class separability through the symmetry of the operations mapping one class to the other: positive distance exposes exploitable local features, whereas zero distance requires global semantics rather than local rules. We construct a visual self-referential task in maximum-variance binary noise: positive samples contain a closed square, while negative samples contain an otherwise identical square with one flipped boundary pixel. The two classes differ in global semantics but have zero syntactic distance, making local statistical shortcuts unreliable. Experiments on ResNets and Vision Transformers reveal a consistent phase-transition phenomenon, with accuracy collapsing to random guessing once the image scale crosses a critical point and does not recover within the tested range. Larger training sets and models only delay this collapse, while globally attentive ViTs reach it earlier. These results reveal a structural capability boundary of current architectures on global-concept tasks, suggesting that general intelligence may require creating new language, not reusing an existing one.

Panoramic images capture the full visual sphere in a frame, offering context unavailable to conventional cameras. Yet this completeness has an unavoidable geometric cost: the 2-sphere cannot be faithfully mapped to the plane, and every projection introduces distortions that challenge standard vision architectures. This survey traces panoramic scene understanding from projection-based adaptation and distortion-aware engineering to sphere-native modeling, reflecting increasing commitment to spherical geometry. Foundation models form a fourth family, geometry-aware tokenization, which adapts the input interface while reusing perspective-pretrained weights. We review these approaches across five task families: dense prediction, unified multi-task understanding, open-world perception, vision-language reasoning, and dynamic video analysis. Across tasks, the same shift toward spherical geometry recurs. In practice, however, the field has converged not on the strongest sphere-native operators, which are exactly rotation-equivariant but cannot reuse perspective-pretrained backbones and thus have not scaled, but on a compatibility-preserving middle ground combining moderate geometric awareness with large pretrained models. This commitment is uneven: deepest in dense prediction and shallowest in dynamic perception, where methods are spatially sphere-aware yet temporally planar. Foundation-model adaptation has advanced panoramic depth fastest, while layout, surface-normal, and video-level understanding remain largely unexplored. No panoramic foundation model has yet been pretrained on spherical data. We identify five evaluation gaps: spherical-area-weighted metrics, seam-consistency tests, polar-robustness stratification, cross-projection generalization, and standardized open-world protocols. We conclude with a six-point roadmap toward general-purpose panoramic intelligence.

Prototype-based neural networks aim to provide intrinsic interpretability by grounding predictions in a small set of part prototypes. However, modern vision backbones typically operate in normalized, directional embedding spaces where each semantic part exhibits substantial intra-class variability. As a result, point prototypes often become redundant or unstable, hurting both explanation quality and robustness. We propose vMFProto, a distributional part-prototype framework that models each class as a mixture of von Mises-Fisher components on the hypersphere. Each prototype learns its own concentration, capturing part-specific variability, and we use entropic optimal transport (OT) to obtain structured patch-to-prototype assignments. A two-stage training schedule performs OT-driven prototype discovery followed by end-to-end refinement with patch-level distillation and distribution-aware diversity regularization. Experiments with frozen DINO backbones show that vMFProto achieves leading consistency and distinctiveness on CUB-200-2011 and competitive classification accuracy across CUB, Stanford Dogs, and Stanford Cars. Qualitative results confirm that vMFProto yields localized, non-redundant part evidence.

Sparse autoencoders (SAEs) have become a leading tool for interpreting the representations of vision foundation models, decomposing their polysemantic activations into a larger set of sparse, more monosemantic features. The Top-$k$ SAE, a now-standard variant, enforces sparsity architecturally through its activation function, retaining only the $k$ most active latents per input. Because it was designed precisely to avoid the $\ell_1$ penalty used by earlier SAEs and its known drawbacks, it has not been combined with an explicit sparsity regularizer. Yet the Top-$k$ SAE retains limitations of its own, and we hypothesize that a sparsity penalty acting before the selection could sharpen each latent's selectivity and make the code more interpretable, without reintroducing the drawbacks of the $\ell_1$ penalty. We introduce two sparsity regularizers compatible with the Top-$k$ architecture, both acting on the activations before the Top-$k$ selection: an $\ell_1$ penalty on the unselected (off-support) units, and a scale-invariant $\ell_1/\ell_2$-ratio penalty that concentrates the code onto fewer effective units. Both penalties are applied only to the batch-active units, those selected by the Top-$k$ operator at least once within the batch. Across two datasets, three vision foundation models, and a range of $k$, both regularizers consistently improve monosemanticity at no cost to reconstruction quality. The $\ell_1/\ell_2$ penalty further concentrates information into fewer latents, making reconstruction more robust to the inference-time choice of $k$ and improving small-budget linear probing. Our central finding is that hard architectural sparsity and soft sparsity regularization are complementary rather than mutually exclusive.

AI safety is evaluated by how reliably a model detects the hazards it is told to find, yet accidents often arise from the hazard no one specified. We show that conditioning a language or vision model on a narrow task suppresses its reporting of co-present, safety-critical signals it can otherwise report, a machine analogue of human inattentional blindness, produced by a different mechanism. Across radiology and driving text scenarios and chest-radiograph vision tasks, the ordinary focused instructions under which such systems are deployed suppressed reporting by up to 0.92 in report rate relative to the same models when unconstrained, and an explicit exclusive instruction abolished reporting entirely in radiology. Suppression appeared in every model tested, did not diminish with scale, persisted in a reasoning model, and varied more by model family than by size. We name this dissociation the Inattentional Gap and argue that it decouples measured benchmark safety from real-world safety: a system can score near-perfectly on the hazards an evaluation specifies while remaining blind to those that cause harm. Probing the mechanism, we localize the proximal trigger to output scope and find System-1-style task capture without reliable intrinsic oversight in the sampled systems. Oversight could, however, be supplied externally: routing each narrow report to an independent open-ended critic restored every omitted finding, demonstrating that the gap is both measurable and mitigable. We propose reporting-complete evaluation, scoring what a system fails to report alongside what it is asked to find, as a requirement for safety-critical deployment.

Visual prompt tuning has emerged as a parameter-efficient fine-tuning approach for adapting large-scale Vision Transformers (ViTs) to downstream tasks. As its learnable prompts are applied in input and feature spaces, prior to jointly going through attention in transformer layers, the most commonly used scheme for fusing image and prompt tokens is concatenation or addition. In this paper, we aim to study a fundamental yet essential problem in visual prompt tuning: whether a single fusion scheme tends to yield better results, and whether that would be beneficial to develop a hybrid fusion scheme. To this end, we formulate the task as a bi-level optimization problem, and solve it leveraging differentiable architecture search. In this context, the learnable prompts and their fusion schemes are jointly optimized. To enrich the search space in the architecture search, we propose two additional fusion schemes, namely, affine transformation and cross-attention, in addition to concatenation and addition. Extensive experiments on 34 datasets spanning VTAB-1k, FGVC, and HTA show consistent gains over prompt-tuning baselines. With a frozen ViT backbone, our method delivers a favorable accuracy--latency--parameter trade-off compared with VPT-Deep and recent variants. Our findings reveal that how prompts fuse with image tokens plays a significant role in visual prompt tuning, and a hybrid fusion fashion can more effectively leverage layer semantics of ViTs, contributing a novel perspective for visual prompt-tuning research.

The computational complexity of Transformers scales quadratically with the number of tokens, which significantly constrains the efficiency of vision models, particularly recent ViT-based foundation models in dense prediction tasks. Instance segmentation, a typical dense visual prediction task in the remote sensing field, faces similar challenges. In this paper, inspired by the recent advances of knowledge distillation in large language models, we introduce RS4D - a new remote sensing instance segmentation method with linear computational complexity, which addresses the inefficiency of long sequence modeling through distilled state space modeling (SSM). We propose an adaptive noise and masking knowledge distillation training method for pre-training lightweight SSM backbones, which effectively compresses knowledge from the vast self-attention space into a compact, dense linear state space. We also design a remote sensing image instance segmentation architecture based on this lightweight visual encoder, where we explore variants of three different backbones and two segmentation heads. Extensive experiments are conducted on multiple benchmark datasets, including SSDD, WHU, and NWPU. Compared to ViT-based approaches, our proposed SSM backbone achieves an 8x reduction in parameters and a 9x reduction in FLOPs while maintaining comparable or superior accuracy to both ViT- and CNN-based instance segmentation methods. The implementation codes have been publicly available at https://github.com/QinzheYang/RS4D.

Remote sensing object detection has advanced rapidly with the development of large-scale benchmarks and modern detection architectures. However, existing datasets and detectors remain fragmented. Most benchmarks focus on limited categories, fixed spatial resolutions, or a single sensor, while detectors still struggle to work across different sensors and categorical systems. In this paper, we introduce LEVIRDet-159, the largest and most comprehensive remote sensing object detection dataset to date, with 159 categories, 2.56 million bounding boxes, and 700k fine-grained annotations under a multi-level taxonomy. In each key scale dimension, LEVIRDet-159 exceeds the corresponding largest existing remote sensing object detection dataset, containing approximately (7x) more images, (6x) more object instances, and (4x) more categories. Based on this dataset, we design LEVIRDetNet, a scale-hierarchy-aware detection foundation model for universal remote sensing object detection. LEVIRDetNet couples online visual Ground Sampling Distance (GSD) prediction, GSD-conditioned query modulation and allocation, and a hierarchy-aware detection head for mixed-granularity remote sensing supervision. Under stringent evaluation settings, LEVIRDetNet demonstrates strong cross-domain generalization. Even without target-domain training or fine-tuning, it achieves state-of-the-art detection performance on 9 external benchmarks, improving the strongest fully supervised competing methods by 5.02 mAP on average under each benchmark's primary metric. We hope this study will facilitate the development of strongly generalizable remote sensing object detection across diverse category systems, spatial resolutions, and sensor platforms. The dataset and trained models will be released at https://qinzheyang.github.io/LEVIRDet/, accompanying the final paper.

Recent advances in stereo matching have achieved remarkable accuracy, but often rely on large models, heavy computation, or additional foundation-model priors, making them difficult to deploy on resource-constrained platforms. In contrast, efficient stereo models offer faster inference but are commonly considered less capable of strong zero-shot generalization. In this paper, we challenge this assumption by introducing Lite Any Stereo V2 (LAS2), an ultra-fast model series designed for efficient zero-shot stereo matching. LAS2 is developed from both architecture and training perspectives. Architecturally, we revisit efficient stereo design under practical deployment settings and propose a 2D-only cost aggregation framework, optimized for real inference latency rather than theoretical MACs alone. For training, we develop a three-stage strategy that combines synthetic supervision, self-distillation, and real-world knowledge distillation. To improve the reliability of real-world pseudo supervision, we further introduce pseudo-label filtering and an error-clamping operation, enabling smoother synthetic-to-real transfer. We instantiate LAS2 as a family of models, including feed-forward variants for different efficiency budgets and an iterative variant for higher accuracy. Extensive experiments show that LAS2 achieves state-of-the-art accuracy among efficient stereo methods while maintaining significantly lower latency. Specifically, LAS2-H achieves stronger overall zero-shot performance than the iterative method Fast-FoundationStereo, with 1.8x and 2.7x faster inference on H200 and Orin, respectively. The project page, demos, and code are available at https://tomtomtommi.github.io/LiteAnyStereoV2/.

Vision Foundation Models (VFMs) have significantly advanced dense feature matching, yet severe in-plane rotation remains a critical challenge. Existing solutions face a fundamental dilemma: data-driven methods require inefficient parameter scaling to implicitly learn rotations, whereas strictly equivariant networks lack the semantic capacity of modern VFMs. Consequently, current frameworks typically freeze VFMs and shift the entire burden of rotation generalization to the downstream decoder. To break this architectural bottleneck, we propose REDI-Match, an efficient framework driven by a novel Rotation-Equivariant Distillation (REDI) paradigm. Instead of relying on rotation data augmentation to establish rotational correspondences, REDI distills the non-equivariant semantic representations of a VFM into a lightweight, strictly rotation-equivariant encoder, leveraging an equivariant geometric architecture to constrain robust high-dimensional semantics. To fully exploit these features, we equip the decoder with an entropy-driven spatial alignment module. By evaluating discrete rotation hypotheses, this mechanism explicitly locks onto the canonical coordinate system, eliminating global ambiguity before continuous refinement. Extensive experiments demonstrate that REDI-Match establishes a new state-of-the-art (SOTA) across multiple benchmarks. Notably, it achieves a 13.89% absolute pose accuracy improvement on the highly challenging SatAst dataset while operating 1.9x faster than the current SOTA (RoMa v2), enabling real-time inference (~41 FPS) on a single RTX 4090 GPU. Code: https://github.com/YinjiGe/REDI-Match.

Cross-domain Few-shot Segmentation (CD-FSS) aims to transfer knowledge learned from source domain to distinct target domains, segmenting unseen target classes with only a few annotated samples. Although existing methods have made significant progress, they still rely on training or fine-tuning processes, which incur high computational costs and risk overfitting. We observe that when powerful and general-purpose vision foundation models are incorporated into these methods, their performance shows only marginal improvement or even degrades due to overfitting. To address this, we eliminate trainable parameters and propose a training-free framework to avoid both training overhead and overfitting. Built upon the self-supervised vision encoder DINOv3, our framework addresses cross-domain challenges through three core modules. First, the Semantic-aware Feature Re-fusion (SAFR) module identifies and re-fuses features that emphasize semantic patterns, generating representations with enhanced semantic discriminability. Additionally, the Adaptive Support Enhancement (ASE) module narrows semantic gaps between support and query through robust query information aggregation. Finally, the Hybrid Prototype Matching (HPM) module integrates matching results from diverse prototypes to adapt to varying semantic complexity across domains. Extensive experiments on four target domain datasets demonstrate that our method achieves state-of-the-art performance in CD-FSS without any training.

We introduce FiCA, a Feed-forward, instant Gaussian Codec Avatar generation pipeline that creates lifelike avatars from a single portrait image. Generating a photorealistic and drivable avatar from just a single image is significantly challenging due to the limited visual information available to accurately infer the 3D appearance and geometry of human heads. To address this, we develop a novel system that combines human-centric vision foundation models with a diffusion model. This system is designed to fully exploit partial visual observations to generate lifelike human avatars. Our proposed diffusion model learns a generative mapping from these partial observations to complete and authentic 3D mesh reconstruction. Additionally, we introduce a feed-forward mesh refinement network that enhances the fidelity and identity preservation of the generated avatars, eliminating the need for person-specific test-time optimization. By leveraging a universal prior model that decodes a generated mesh into a set of 3D Gaussians, we generate a photorealistic 3D Gaussian avatar, capable of being driven with novel expressions in real-time. Our experiments demonstrate that the avatars generated by our feed-forward approach faithfully represent diverse identities and surpass the visual quality of avatars produced by recent competing methods.

Timely, high-resolution maps of flood extent around settlements are essential for emergency response and damage assessment. We consider airborne RGB imagery for flood mapping as it can be collected rapidly at low cost. To produce flood maps, deep learning models for water segmentation are often used. CNN based and small vision transformer models are used. However, they need much data for adaptation to a change of scenery, i.e., another flooding event. Vision foundation models or large vision transformers are known to generalize across domains. Recently, foundation models for Earth observation became available. They are pretrained on satellite data, whose spatial resolution, viewing geometry, and radiometry differ from nadir RGB imagery. Thus, adaptation is required. We investigate how a satellite-pretrained Earth observation foundation model can be adapted to centimeter-scale floodwater mapping from RGB imagery. Specifically, we fine-tune a model we call Prithvi-2.0-UPN consisting of the Prithvi-EO-2.0-600M Vision Transformer combined with a UPerNet decoder for binary water segmentation on two RGB datasets (BlessemFlood21, NeuenahrFlood). In a first experiment we observe that Prithvi-2.0-UPN reaches state-of-the-art results on BlessemFlood21 and NeuenahrFlood, when trained on their datasets. In a second experiment we show that Prithvi-2.0-UPN performs better than state-of-the-art baseline models for transfer to a new flood event (trained on BlessemFlood21, tested on NeuenahrFlood) in a zero-shot setting. However, the performance indicates room for improvement. In this respect, we investigate in a third experiment how performance improves when further fine-tuning the models with small shares of NeuenahrFlood training data: Prithvi-2.0-UPN improves the fastest and reaches almost the performance level when fully trained on NeuenahrFlood, indicating transfer capabilities.

Pre-trained Vision Foundation Models (VFMs) have become central to modern computer vision due to their powerful semantic representations and strong generalization ability. However, their patchified or pooled outputs are inherently low-resolution, limiting their effectiveness in tasks requiring fine-grained, pixel-level reasoning. Existing feature upsampling approaches either degrade semantic fidelity or rely on VFM-specific retraining and heavy architectures, hindering efficiency and scalability. To address these challenges, we propose RaysUp, an ultra-lightweight, task-agnostic, and VFM-agnostic feature upsampling framework that reconstructs high-resolution feature maps at arbitrary resolutions. Unlike conventional 2D interpolation or attention-based schemes, RaysUp lifts feature reconstruction into a geometry-aware ray domain. Specifically, we introduce a Spatially Decoupled Guidance Encoder for direction-aware guidance encoding, an Any-Resolution Cross-Attention mechanism for resolution-flexible reconstruction, and a novel Ray Positional Encoding (RayPE) that injects implicit 3D geometric priors via 6D Plucker ray coordinates. Finally, a Geometry-Aware Neighborhood Attention module further ensures content-adaptive bilateral aggregation while preserving geometric consistency. Extensive experiments across diverse dense prediction tasks demonstrate that RaysUp achieves state-of-the-art performance while using only 16% of the parameters of AnyUp and delivering approximately 7x faster inference. These results highlight a substantially improved accuracy-efficiency trade-off and establish RaysUp as a practical and scalable solution for universal feature upsampling. Code is available at https://github.com/MAP-RaysUp/RaysUp.

Cell segmentation is critical for computational pathology and biomedical discovery. While recent Vision Foundation Models (VFMs) have demonstrated remarkable universal feature representations, unlocking their full potential for cellular imaging is currently bottlenecked by resource-intensive adaptation paradigms. Existing methods typically rely on fine-tuning heavy visual encoders, leading to extensive computational overhead and a dependency on large-scale annotations. To address this, we propose the EffiCell-Seg framework for highly efficient cell segmentation without re-training the visual encoder. Our core insight is that pretrained VFMs intrinsically encode complementary structural priors: global saliency for localizing potential cells, and local morphological patterns for delineating cellular structures. To harness these priors, we devise a Cell Structure Prompt Encoder (CSP-Encoder) that synthesizes semantic-aware saliency and principal morphological features from frozen VFM representations into explicit structural prior maps. Moreover, we propose a Synergistic Mask Decoder (SM-Decoder) that enforces contextual consistency by jointly predicting geometric distance fields and semantic maps via mutual cross-guidance. Extensive experiments demonstrate that EffiCell-Seg outperforms state-of-the-art methods across diverse cell imaging modalities while requiring only ~5M trainable parameters, over 130x fewer than fully fine-tuned VFM counterparts. The code is available at https://github.com/xq141839/EffiCell-Seg.

Vision foundation models have substantially advanced computer vision, enabling state-of-the-art performance in zero- and few-shot settings. They have been successfully applied to biomedical imaging tasks ranging from organ segmentation in computed tomography to cell segmentation in light microscopy. Electron microscopy (EM) is a central modality for analyzing cellular ultrastructure due to its nanometer-scale resolution. However, the application of foundation models in EM has so far been limited to specific organelles, such as mitochondria, largely due to the diversity of segmentation tasks and the scarcity of comprehensively annotated data. As a result, EM segmentation still predominantly relies on supervised learning, requiring extensive manual annotation and limiting ultrastructural analysis. To address this gap, we propose $μ$Match, a framework for semi-supervised learning and domain adaptation that leverages foundation models. We implement state-of-the-art student-teacher-based methods and evaluate multiple foundation models (SAM, SAM2, $μ$SAM, DINOv2/v3) on challenging EM tasks, including mitochondrion, nucleus, and neurite segmentation. Our results demonstrate consistent improvements over strong baselines and highlight a path toward substantially reducing the annotation effort in EM.

Leveraging Vision Foundation Models (VFMs) for camera-to-LiDAR knowledge distillation offers a promising solution to the scarcity of annotated data needed to represent the immense geometric and kinematic diversity of real-world autonomous driving (AD). However, current approaches typically treat VFMs as black-box teachers, relying exclusively on frame-wise feature similarity. Consequently, they do not fully exploit the teacher's layer-wise semantic structure and global context, as well as the rich spatiotemporal information inherent in LiDAR sequences. We propose HilDA, a self-supervised pretraining framework for LiDAR backbones that better captures the semantic what and geometric where needed for driving tasks. HilDA combines hierarchical distillation comprising multi-layer distillation for progressive semantic alignment and global context distillation for scene-level semantics, with a temporal occupancy diffusion objective promoting spatiotemporal consistency. Models pre-trained with HilDA achieve state-of-the-art results on cross-modal distillation benchmarks and outperform models trained via prior distillation approaches on 3D object detection, scene flow, and semantic occupancy prediction. Code available at: https://maxiuw.github.io/hilda.

Sparse Mixture-of-Experts (SMoE) architectures are now widely deployed in state-of-the-art language and vision models, where conditional routing allows scaling to very large networks. However, this very Top-$k$ expert selection that enables conditional routing also renders the SMoE map inherently discontinuous. In the vicinity of these discontinuity surfaces, even inputs that are arbitrarily close may activate substantially different sets of experts resulting in significantly different outputs. In this work we give a rigorous geometric and stochastic analysis of these discontinuities. We first classify them by order, determined by the number of tied experts at a switching event. Using measure-theoretic slicing arguments, we establish asymptotic volume estimates for the thickened discontinuity surfaces, showing that lower-order discontinuity sets dominate, whereas higher-order ones occupy a vanishingly small relative volume. Next, modeling random perturbations in the input space via a diffusion process, we prove that the path eventually encounter a discontinuity, and moreover that the first hit almost surely occurs on an order-1 discontinuity with explicit finite-time probability bounds. We further derive occupation-time bounds that quantify the duration the random path spend in the neighborhoods of each discontinuity order. These theoretical results imply that inputs are more likely to lie near lower order discontinuities. Motivated by this insight, we propose a simple smoothing mechanism that can be directly applied to existing SMoEs, softly incorporating experts near discontinuities; our analysis guarantees that the added computational overhead remains small while providing localized smoothing near discontinuities, and experiments across language and vision tasks show that smoothing not only enforces continuity of the SMoE map but also enhances empirical performance.