Vision Foundation Models

Forest inventories increasingly rely on artificial intelligence (AI) models to derive forest attributes from large-scale 3D point clouds. Current models are typically specialized to a single task, sensor, and forest type, making adaptation expensive in terms of annotations, computation, and expertise. We ask whether a single pretrained model can instead learn transferable representations across diverse forest inventory settings. Inspired by recent developments in language modelling and computer vision, we take a step toward a foundation model (FM) for 3D forestry. Using LitePT as backbone, we first establish a strong supervised baseline that sets a new state of the art on forest semantic and instance segmentation, tree species classification, and age regression benchmarks. We then curate a large-scale unlabelled corpus spanning airborne, UAV, and mobile laser scanning across diverse forest ecosystems, and pretrain the same backbone using self-supervised learning. We systematically evaluate representation learning strategies by comparing training from scratch, supervised pretraining, and self-supervised pretraining across four representative forestry tasks, under varying annotation budgets. Compared with training from scratch, self-supervised pretraining accelerates model convergence and consistently improves performance when annotations are scarce. Compared with task-specific supervised pretraining, self-supervised pretraining yields more transferable representations across downstream forestry tasks. These findings identify the practical regime in which pretrained representations are most valuable and suggest that instance discrimination, rather than forest semantics, is the main remaining obstacle to a general-purpose 3D forest foundation model. Code and models are available at: https://github.com/prs-eth/ForPT.

Vision foundation models targeting Earth observation (EO) tasks are commonly evaluated on clean downstream benchmarks, but operational EO products can already contain spatial, radiometric, alignment, noise, and harmonization defects before reaching the model. Existing robustness evaluations often use generic image corruptions or broad domain shifts, which do not isolate these product-level failure modes. We introduce \textbf{RSPDBench}, a physically grounded \textbf{r}emote-\textbf{s}ensing-\textbf{p}roduct \textbf{d}egradation \textbf{b}enchmark for vision foundation models. RSPDBench evaluates five EO datasets, seven foundation-model entries, and two supervised baselines under audited primitive degradations and compound product chains. Each model is evaluated under its clean-selected native protocol, with robustness measured as the drop from its own clean baseline. Our analysis reveals that degradation sensitivity is strongly structured: resolution-conditioned and channel-grouped encoders protect different failure axes, and the same physical defect can hurt one model while helping another. Compound chains expose failures that isolated degradations do not predict, with model-dependent amplification, saturation, or component dominance, and excess drops up to $38$ percentage points beyond the strongest component. These results show that EO robustness cannot be characterized by clean accuracy or generic perturbation tests alone; it must also be measured against the structured defects that remote-sensing products carry into deployment.

Geospatial foundation models can provide strong flood-segmentation performance, but their size limits deployment on memory-constrained edge hardware. We distill a 300-million-parameter Prithvi-EO-2.0 teacher, fine-tuned on the 252 manually labeled Sen1Floods11 training scenes, into a 0.7-million-parameter EfficientViT-B0 student. The teacher supervises additional unlabeled Sentinel-2 imagery, allowing the student training set to grow without new manual annotations. At the matched budget of 252 scenes, teacher-supervised training is competitive with direct training and improves STURM-Flood performance across tested configurations; a geometry-matched control shows that label source alone does not explain the difference. Scaling the teacher-supervised pool to 2,500 scenes narrows the remaining student--teacher gap: the float student reaches 0.787 water intersection over union on the Sen1Floods11 test split against 0.822 for the teacher, matches the teacher on STURM-Flood under our evaluation protocol, and remains below it on WorldFloods-v2. After activation replacement and quantization-aware training, the student runs as a 1.5-megabyte 8-bit integer (INT8) TensorRT engine on a Jetson Xavier NX at 5.57 milliseconds of graphics processing unit (GPU) compute per 512-by-512 image, with approximately 14 megabytes of runtime device memory. A fixed modified normalized difference water index (MNDWI) threshold is competitive with both models on the two clean external benchmarks, so we interpret those benchmarks as generalization tests rather than as evidence of learned-model superiority over a spectral rule. The results support the conclusion: foundation-model supervision can amplify a fixed manual annotation budget into a substantially larger training set and yield a compact, deployable edge model.

Iris presentation attack detection (PAD) is security-critical when a subsystem that appears reliable during development encounters presentation attack instruments (PAIs) or acquisition conditions absent from validation data. We benchmark three compact scratch-trained computer-vision models, each with at most approximately 0.26 million trainable parameters, on the Notre Dame subset of LivDet-Iris 2017 under PAI-driven domain shift and environmental degradation. All models are trained without external pretraining or data augmentation and evaluated over five seeds. A validation-selected threshold is transferred unchanged to the known-attack, unknown-attack, corrupted, and pooled test partitions. From known to unknown attack presentations, Attack Presentation Classification Error Rate (APCER) increases by 17.11-30.47 percentage points and Detection Equal Error Rate (D-EER) increases by 7.38-12.73 percentage points. At the validation-selected threshold, ZACH-ViT obtains the lowest unknown-attack APCER (47.69 +/- 4.84%) and D-EER (38.87 +/- 0.93%), while Compact-TransMIL obtains the lowest Bona Fide Presentation Classification Error Rate (BPCER). ZACH-ViT also gives the lowest unknown-attack BPCER at an APCER limit of 10% (81.29 +/- 1.95%). The high absolute errors show that the comparative advantage of the best compact model does not constitute deployment readiness under unknown PAIs.

This study developed and evaluated a deep-learning-based perception framework for selective robotic cotton picking. The dataset contained 1,008 annotated field images collected using three cameras under varying natural lighting and weather conditions. Object-detection models from the YOLOv8 through YOLOv13 families were evaluated using their default configurations, while segmentation performance was assessed using YOLOv8-seg, YOLOv11-seg, YOLOv12-seg, the Segment Anything Model (SAM), SAMv2.1, FastSAM, and Grounded-SAM with the Recognize Anything Model (RAM). Among the detection models, GELAN-s achieved the most favorable balance between mean average precision (mAP) and inference speed, obtaining an mAP of 86.1%, precision of 81.6%, recall of 76.6%, and an F1-score of 79.0%, with an average inference time of 42.3 ms per image. Among the direct segmentation models, YOLOv12-m-seg provided the most favorable balance between AP@0.5 and FPS, achieving a segmentation AP@0.5 of 83.7% with an inference time of 20.4 ms per image. In the detection-prompted segmentation approach, bounding-box prompts generated by GELAN-s improved the localization of cotton bolls for SAM and SAMv2.1, while SAMv2.1 Tiny consistently outperformed FastSAM and Grounded-SAM with RAM. In the area-based evaluation against manually annotated segmentation masks, YOLOv12-m-seg achieved an $R^2$ value of 0.966, compared with 0.860 for GELAN-s + SAMv2.1 Tiny. Field experiments conducted using a UR5e robotic manipulator, a custom end-effector, and a ZED2i stereo camera further validated the effectiveness of the YOLOv12-m-seg model for real-time cotton boll detection, segmentation, and selective picking under varying confidence levels. These results demonstrate that YOLOv12-m-seg provides an efficient perception model for robotic cotton harvesting and has strong potential for field deployment.

Scaling deep learning faces critical bottlenecks: data exhaustion, exponential training costs, and resource concentration. Model merging combines pre-trained checkpoints without gradient descent, offering orders-of-magnitude savings versus retraining. Combining independently trained vision models is difficult when their architectures and parameter shapes differ. Existing weight-space merging methods generally assume aligned, shape-compatible checkpoints, whereas a Vision Transformer (ViT) and a state-space model (SSM) implement token mixing with different operators. We study a hybrid Heterogeneous merging setting that retains both architectures while aligning parameter groups by semantic role. Our proposed Riemannian--Lorentz Parameter Fusion (RLPF) method projects aligned groups to common coordinates, lifts selected coordinates to the Lorentz hyperboloid model of hyperbolic space, computes a regularized geodesic barycenter, and decodes the result into the two branches. A learned gate then combines branch logits for each input. Component groups use fixed curvature values, with normalization parameters treated as Euclidean. In the results available in this manuscript, the fine-tuned system obtains 82.37% on CIFAR-10, 75.04% on Oxford-IIIT Pet, and 78.58% top-1 accuracy on ImageNet-1K; the corresponding best-parent accuracies are 76.54%, 71.42%, and 76.42%. On ImageNet-1K, the reported pre-fine-tuning initialization reaches 77.80%. These results support further study of geometry-aware heterogeneous fusion, but not a training-free single-checkpoint merge: RLPF is a two-branch hybrid whose gate and reported final models are trained.

woma is a real-time foundation model for gastrointestinal endoscopy: a network trained without labels on about a million endoscopy frames, from which task models are fine-tuned. We contribute a systematic design for production. Requirements and pass marks were fixed before any run, eight candidates screened under pre-registered rules, self-supervised training taken to a stopping rule, then fine-tuning and deployment optimisation, all on one self-contained library, numbat. We also contribute woma itself with two fine-tuned models, every outcome reported met or missed. Our colonoscopy model finds and outlines polyps, names which colon segment is in view, suggests polyp type and grades bowel preparation. Our gastroscopy model names a station out of 22 protocol sites, flags and outlines lesions, and names one of seven findings. Every number was read on data never seen in training, and shipped weights were chosen on that record. In colonoscopy, 96% of polyps in a six-hospital PolypGen set are found at precision >=0.85, and 19 of 19 polyps across fifteen full REAL-Colon videos at 1.6 false alarms per procedure. In gastroscopy, landmark region is named correctly on 92% of frames from unseen patients, and 37 of 39 held-out neoplasia frames are flagged at specificity 0.91. On one workstation GPU every task runs over 1080p video at about 100 frames per second, faster than PyTorch, ONNX Runtime and TensorRT in all four precision regimes tested. TensorRT comes closest: one pass of our foundation model takes it 3 to 27% longer than ours, and we deliver 6 to 31% more frames per second from frame to results. A second build links no vendor library at all -- our own kernels over Vulkan -- so a site deploys two files and needs no toolkit, no cuDNN and no framework; in f32 it beats the CUDA build on the same card.

Vision-Language Models such as CLIP enable effective few-shot medical anomaly detection (AD) via strong image-text semantic alignment. However, their globally contrastive pretraining lacks explicit spatial supervision, limiting precise lesion localization. In contrast, Vision Foundation Models (VFMs) such as DINO learn spatially coherent patch representations via self-distillation and local-to-global consistency, better capturing fine-grained anatomical structures. Leveraging this complementarity, we propose Spatial-FAD, a spatial-aware few-shot medical AD framework that improves lesion localization by combining VFM spatial priors with CLIP semantics. Specifically, we introduce a VFM-enhanced adapter that injects a structural affinity prior derived from DINO into CLIP features. This structure-guided refinement encourages visual embeddings to better adhere to lesion boundaries while maintaining semantic alignment. To address the loss of spatial detail from patchification and the limited input resolution of CLIP, we adopt a sliding-window aggregation strategy. This generates high-resolution, spatially dense embeddings to further enhance localization granularity. Moreover, we introduce a prototype-enhanced support memory scheme to efficiently exploit the few-shot support set. This module stores compact prototypes for normal and abnormal patterns, reducing memory costs while boosting performance by fusing patch-to-prototype and image-text similarities. Extensive experiments on three benchmark datasets, including Liver CT, Retinal OCT, and Brain MRI, demonstrate that Spatial-FAD significantly outperforms state-of-the-art methods, especially in lesion segmentation. Notably, in the 4-shot scenario, our method achieves an average improvement of over 11.4% in Dice score and 1.8% in AUC. Code is available at: https://github.com/JuzhengMiao/Spatial-FAD.

Solving novel view synthesis (NVS) for millimeter-wave (mmWave) radar requires a renderer that is physically faithful, complex-valued, and multi-viewpoint-tractable. No prior method achieves these three properties simultaneously. Differentiable Monte Carlo (MC) ray tracers implement the radar forward model directly with explicit material modeling and complex outputs, but do not scale to the multi-view optimization NVS demands. Optical-NVS ports of NeRF, hash grids, and 3D Gaussians train fast but discard phase and replace explicit material modeling with opaque learned features, restricting them to power-only range-azimuth (RA) magnitudes. We propose 3D Point Splatting (3DPS), the first differentiable point renderer for radar, derived directly from the standard solid-angle form of the radar equation. Each oriented 3D point carries an ITU-R P.2040 material model, evaluated in closed form, with the resulting complex phasor splatted into range bins through a precomputed point spread function (PSF). The complex-valued output makes the renderer product-agnostic. The same optimized scene yields analog-to-digital converter (ADC), complex range profile (CRP), and RA outputs through standard fast Fourier transform (FFT) pipelines without retraining for each format. On six outdoor ColoRadar scenes, 3DPS reaches 0.587 mean Pearson correlation on held-out RA images. This is between 1.7x and 5.2x the three optical-NVS baselines (RadarSplat, Radar Fields, DART). Training takes approximately 3 minutes per scene on a single RTX 4090.

A language model normally begins training with random word embeddings: whatever 'banana' means must be learned from training corpora. I implement St. Augustine's picture of word learning, meaning by ostension, for a small masked language model (DeBERTa) trained on 10M words: before training, visually grounded tokens receive embeddings derived from the image regions they label; other tokens start random. Visual initialization leaves a measurable imprint that lasts until the end of training. At the same time, the effect remains invisible under most BabyLM benchmarks, which probe abstract grammatical knowledge: visual initialization does not affect performance there. The only zero-shot exception is object-property knowledge (COMPS), where seeding helps in every configuration. To follow up on this result, I build a corpus-tailored version of the Visual-Property Swap benchmark, which tests color, material, size, and shape knowledge, with per-item training frequency and seeded status. Here, vision-seeded models have a persistent, seed-replicated advantage. Function words and abstract vocabulary also receive strong visual seeds and retain them throughout training, and the training objective draws on them: held-out mask-prediction loss falls for these words in every seed. However, no benchmark I run registers this. What evaluation would pick this up remains an open question.

State-of-the-art vision models process images in their entirety, lacking the ability to selectively zoom in on relevant regions. This limitation is particularly acute in scenarios where processing must be conditioned on a specific task - such as instance detection, which requires localizing a specific object in a high-resolution, cluttered scene. In such settings, critical details are easily lost as images are often resized to match the model dimensions and computational constraints. We introduce Coarse-to-Fine Glimpse-based Active Perception (CF-GAP), a task-driven front-end that enhances high-resolution processing of existing instance detectors. CF-GAP selectively directs a sequence of limited view glimpses across the scene, utilizing task information to iteratively refine focus on the most relevant regions. These localized regions are then processed at high resolution by a downstream instance detector. By avoiding full-image processing and eliminating irrelevant confounding information, CF-GAP improves Average Precision (AP) by up to 20% across various state-of-the-art instance detectors on the HR-InsDet and Robotools benchmarks, while further enabling lightweight detectors to outperform their larger counterparts.

We present DXPR, a depth-based cross-modal place recognition (CMPR) framework that uses vision foundation models (VFMs) to match monocular camera queries against a LiDAR map without modality-specific encoders. This enables robots and autonomous vehicles to robustly localize using only cameras within pre-built LiDAR maps, even under severe seasonal, weather, and illumination changes. The key idea is to convert both camera images and LiDAR scans into a unified depth image representation so that a single VFM backbone with an aggregation head can learn modality-invariant global descriptors. To make pairwise metric learning faithful to scene geometry, we introduce a geometry-aware overlap miner: after cross-modal scale alignment of camera and LiDAR depth, we forward-warp measurements between views to compute a pixel-level overlap score. This score relabels ambiguous pairs and adaptively modulates the positive margin in a multi-similarity loss to avoid overfitting on weakly overlapping views. Extensive experiments on KITTI odometry and Boreas demonstrate strong performance and robustness across seasons, weather, and day/night. On KITTI, DXPR achieves near-perfect Recall@1 on most sequences and outperforms prior CMPR baselines. On Boreas, DXPR achieves intra-sequence performance on par with a strong single-modal baseline (DINOv2-SALAD), while showing clear improvements in the more challenging inter-sequence setting. Compared with RangeBEV, our method consistently performs better in both intra- and inter-sequence evaluations, demonstrating robustness under diverse seasonal and illumination changes.

Semantic and goal-oriented communication is increasingly studied for 6G, but generalization beyond seen data remains a key weakness under tight rate budgets. Many existing systems overfit their training data and degrade sharply at very low bit rates because they attempt to compress the entire signal. We introduce Foundation Model-Guided Semantic and Goal-Oriented Communication (FMSGOC), a framework that uses broad visual-linguistic Foundation Model priors to mitigate overfitting. It further improves rate efficiency by concentrating bits on sparse, goal-aligned anchors and relying on generative foundation-model priors to reconstruct the masked regions. By decoupling what to send from how to reconstruct, a vision-language foundation model selects and transmits a sparse set of semantic anchors, while a pretrained diffusion model, fine-tuned for masked completion, reconstructs the image at the receiver. In our experiments, FMSGOC reaches 0.039 bits per pixel (BPP), maintains high semantic fidelity (cosine similarity 0.87-0.90 on CIFAR-10), remains robust on previously unseen inputs (0.83-0.86 on ImageNet), and shows good perceptual similarity (0.1278/0.1558, CIFAR-10/ImageNet), outperforming strong end-to-end baselines at lower bit rates.

Synthetic Aperture Radar (SAR) is an important modality in a wide range of imaging applications due to its versatile, long range and near all weather operating capabilities. However, Automatic Target Recognition (ATR) remains a challenging problem due to limited labelled data, the strong speckle in SAR images and the significant domain gap between SAR and more abundant optical imagery. In contrast, electro-optical (EO) imagery benefits from massive datasets, clearer visual structure and powerful foundation models. In this work, we investigate how vision foundation models trained on optical data can provide class level supervision for SAR classification. We propose a cross-modal EO to SAR prototype alignment framework in which a frozen EO encoder, based on a DINOv3 vision foundation model, is used to construct class level optical prototypes without requiring strict EO/SAR pairs. A SAR model is then trained to classify SAR images while aligning its embeddings to the corresponding EO class prototype. At inference time, the SAR model operates independently, without access to optical imagery. We evaluate our approach on the UNICORNv2 dataset, an EO and SAR dataset of civilian vehicles with heavily speckled images and severe class imbalance. EO prototype alignment improves SAR classification accuracy over frozen DINOv3, SAR only finetuning and unpaired distribution alignment baselines, and t-SNE visualizations provide qualitative evidence of clearer separation among classes in the trained SAR embedding space. These results suggest that optical vision foundation models, despite being trained on visible spectrum imagery, provide transferable information for SAR image classification, offering a practical method for using large scale pretrained vision foundation models across challenging sensing modalities.

Training-free collaborative pipelines that integrate Vision Foundation Models such as CLIP, SAM, and DINO achieve strong open-vocabulary dense prediction and are increasingly deployed in safety-critical applications. The security of these systems is commonly assumed to follow from the robustness of their individual models. We challenge this assumption. We identify a vulnerability shared by every collaborative pipeline: each model consumes the intermediate output of another without verifying semantic consistency, an unverified premise that we term the semantic-spatial alignment dependency. Existing adversarial attacks target a single model and overlook this premise, leaving the inter-model interface entirely unguarded. We propose CrACK (Cross-model Adversarial Consistency attack), an inference-time attack that exploits this interface without modifying any input pixel, model weight, or training data. CrACK operates in two stages: Adversarial Affinity Contradiction Injection corrupts the cross-modal affinity matrix by inverting SAM encoder features under the guidance of CLIP patch-level semantics, and Semantic Interface Poisoning steers the prediction through a max-distance label permutation derived from CLIP text embeddings. Experiments on four collaborative pipelines across eight benchmarks show that CrACK causes catastrophic degradation while every individual model continues to produce its unchanged standalone output, rendering per-model defenses structurally blind. The corruption further cascades into large vision-language model reasoning, driving models such as LLaVA to produce erroneous responses from visually intact inputs. Our results show that the security of a collaborative AI system cannot be reduced to the robustness of its components, and that inter-model feature interfaces must be treated as first-class security boundaries.

We present ENEAS, a unified, text-promptable method for instance tracking and semantic discovery. Text-promptable segmentation models, including the latest foundation models such as SAM 3, still suffer from temporal hallucinations, spatial fragmentation, and semantic misclassification: they fail to report target absence when an object leaves the field of view, segment local textures instead of the complete object during extreme close-ups, and prioritize visual features over ontological reality, so that visually similar artifacts such as statues, paintings, or reflections are segmented as target entities. ENEAS works two ways from a single method: precise tracking and high-quality segmentation of a unique instance, and open-concept discovery of every instance a text query names, resolved by a semantic verification layer. For tracking, we extend the geometrically robust SeC architecture, previously limited to point interactions, with a text-prompting adapter and leverage its temporal memory, so that the target is held through disappearance without drifting to distractors and kept whole even when it fills the entire view. For discovery, the verification layer combines high-speed visual embedding matching with conditional VLM refinement, invoking semantic reasoning only for ambiguous candidates, which filters out the ontological errors that visual-only models cannot distinguish while keeping latency low. Designed with 3D reconstruction in mind, where a single misclassified distractor corrupts the asset, ENEAS unlocks high-quality semantic tracking and segmentation of video, of broad libraries, and of collections of temporally or spatially unordered data, together with the discrimination to tell true instances from their doppelgangers: things that look alike but are not the same. The code and models are available at https://github.com/speridlabs/eneas

Vision Foundation Models (VFMs) provide transferable patch representations for few-shot industrial anomaly detection, but their attention computation is typically inherited from pretraining objectives centered on semantic aggregation. This creates a potential mismatch: token relations that support semantic recognition may not adequately expose the localized texture and structural deviations required for anomaly localization. We therefore investigate the hypothesis that the attention computation of a frozen VFM can be reconfigured as a task-relevant component of anomaly detection. We instantiate this idea with Power-Law Self-Correlation Enhanced Attention (PL-SCEA), which retains the semantic context of pretrained query-key attention while constructing token-adaptive self-correlations over contextualized value features. Positive-correlation filtering and power-law reweighting then emphasize relations that are salient relative to each token's relational background, without introducing additional trainable attention projections. The resulting features are modeled by a lightweight variational autoencoder that provides a fixed-size reconstruction-based representation of category-specific normality. The two stages serve complementary roles: attention reconfiguration shapes how local relational deviations are represented, while reconstruction-based modeling converts deviations from learned normality into anomaly scores. Across MVTec AD and VisA, the complete framework achieves competitive image-level detection and consistently strong pixel-level localization across the evaluated few-shot settings. Ablations further show that PL-SCEA improves localization with either the VAE or a memory bank under the tested setting. These results support the view that task-aligned attention reconfiguration can improve the anomaly-localization capability of frozen pretrained representations.

Edge vision models are difficult to deploy on resource-constrained hardware, making low-bit post-training quantization (PTQ) attractive. In practice, standard FP32 training often produces heavy-tailed activation distributions whose outliers destabilize activation quantization: preserving the full range wastes quantization bins on rare extremes, while aggressive clipping causes information loss. Existing solutions typically rely on quantization-aware training (QAT), which adds training complexity and bit-width coupling, or advanced PTQ procedures that repair the model after training. We present SCULPT (Statistical Clipping and Uniform Loss for Post-Training), a training-time method that improves PTQ readiness during ordinary FP32 fine-tuning. SCULPT combines a topology-aware activation regularizer that suppresses quantization-hostile skewness and kurtosis with a stable percentile-based clipping mechanism that learns deployment-ready activation bounds. Unlike QAT, SCULPT does not simulate quantization during optimization; unlike post hoc outlier-repair PTQ methods, it does not require runtime activation transformations. The learned clipping bounds can be exported directly into a standard PTQ workflow for low-bit deployment, including INT8 and lower-bit settings such as W4A8.

Self-supervised video foundation models learn rich spatiotemporal representations, yet it remains unclear what visual concepts these representations encode, where they emerge across transformer layers, and how they are geometrically organized. In this work, we tackle these three questions through a systematic layer-wise analysis of V-JEPA 2 and VideoMAE-v2. We leverage lightweight probes trained to discover three temporally grounded properties: (i) camera motion understanding, (ii) intuitive physics, and (iii) anomaly detection. Both models encode camera motion, with best results ($>90$ ROC AUC) emerging at 60-70% of network depth, and achieve moderate anomaly detection performance ($>60$ ROC AUC), but remain near chance on intuitive-physics tasks, suggesting a limited encoding of deeper physical reasoning. Beyond classification, we find that temporal features from individual videos form smooth low-dimensional trajectories in representation space, suggesting that camera motion is not only linearly decodable but also geometrically organized. Based on these results, we apply geometry-aware spline-based steering in the model's latent representations to interpolate camera motion, yielding steered videos with smoother trajectories and more coherent temporal progression than linear interpolation.

Vision models deployed on microcontrollers (MCUs) are quantized to integer-only arithmetic and run in inference-only runtimes that do not carry the machinery backpropagation needs: the standard tool for adapting a model to the distribution shift (sensor noise, blur, lighting) it meets in the field. Existing forward-only test-time adaptation (TTA) methods either run only on server- or edge-GPU-class models (not true microcontroller integer execution), or require the batch-normalization (BN) layers that integer deployment fuses away. We present a forward-only TTA method that operates on deployed, BN-folded, integer-only convolutional networks. The key observation is that fusing BN into the preceding convolution, a mandatory step for integer inference, destroys the statistics that normalization-based adaptation relies on. We restore adaptation by re-normalizing each folded convolution's per-channel output to its clean training statistics, using only forward-pass estimates. The method (i) recovers most of gradient-based TENT's accuracy gain (+20.9 vs. +24.9 points) and matches forward-only BN adaptation, while being the only method that runs on a folded integer-only model; (ii) needs to adapt only 3 of 21 layers (selected without seeing the test corruptions) to recover 93% of the benefit; (iii) survives single-sample streaming with a batch-size-scaled momentum; and (iv) generalizes across three datasets (up to 200 classes) and two architectures. We validate bit-exact int8 convolution execution and deploy on an ESP32-S3, where, measured with a Nordic PPK2 power profiler, the forward-only adaptation (a lightweight fp32 recalibration around the int8 convolutions) costs only 8.3 mJ (6.8% of inference energy) and 21.9 ms on the deployed SIMD-optimized model: forward-only adaptation is cheap on a real microcontroller.

Global welfare often depends on the correct interpretation of aerial and satellite imagery. Acting on such imagery (mapping flooded ground, crop extent, or damaged infrastructure) demands pixel-level segmentation to ensure perfect class localization. Pretrained general foundation models, when applied directly, often miss important features and cannot always find all the classes belonging to a given scene, overlooking smaller objects that matter most. We use a single consumer-grade GPU running a vision-language model (VLM) to supply this missing guidance, improving segmentation while producing structured, auditable evidence that drives the result and can be inspected on its own. We fuse three approaches: the frozen foundation model that labels every pixel, and two queries to a VLM, one to choose the classes that matter, and one to locate the small objects the base model misses. Evaluating across four aerial datasets, we see consistent gains at each stage where the base model is competent.

Promptable segmentation foundation models such as SAM3 accept an open-vocabulary text concept and return every instance matching it, but adapting them to a specialized domain by full fine-tuning is computationally prohibitive for the organizations that would benefit most. This study applies Low-Rank Adaptation (LoRA) to SAM3 for multi-class structural defect segmentation and examines both how such a model can be supervised from conventional annotation and whether the resulting efficiency gain transfers across datasets. Two contributions are methodological. First, we describe a supervision procedure that trains a concept-promptable model directly from COCO-style class-labeled instance segmentation by using the category name itself as the prompt, requiring no prompt templates, no synonym expansion, and no learned class embeddings. Second, we identify and mitigate a failure mode specific to this setting: because a conventional annotation file yields positive prompts exclusively, the model's presence prediction decouples from the text condition and degenerates into responding to any prompt, a collapse that is invisible to every metric computed on positive prompts alone. Exhaustive hard-negative prompting, in which every dataset category absent from an image is issued as a zero-detection query, addresses this at no annotation cost. Two adapter placements were compared under an identical protocol, updating 0.121% and 1.341% of model parameters. On a purpose-built tunnel lining dataset, pixel intersection-over-union improved from 0.017 to 0.338 and instance-level recall from 0.375 to 0.672; on the independent public Structural Defects Dataset, from 0.017 to 0.855 and from 0.574 to 1.000. Improvements were directionally consistent across ten metrics on both datasets, and the largest per-category gains occurred precisely where zero-shot competence was absent.

Pretrained vision embeddings are increasingly used as general-purpose representations for modelling how people appraise urban scenes, and are validated almost entirely by how well they predict human ratings. High predictive accuracy does not establish that these embeddings organise scenes as human perception does. We test the two properties separately against brain data. Using openly released EEG from 63 adults who viewed and rated 56 Berlin street scenes, we estimate the representational geometry of the scenes over time, the proportion of that geometry that is explainable at all, and its correspondence with seventeen feature spaces spanning language-supervised, self-supervised, category-supervised and dense-prediction training, two orders of magnitude of scale, and interpretable controls. Correspondence is low throughout: the best representation, DINOv2 ViT-B, reaches 29.6% of the lower bound of the noise ceiling, the panel spans 11.0% to 29.6%, and a Gabor energy descriptor is indistinguishable from the best model while outperforming every language-supervised model tested. Within a model, deeper layers still match later neural responses, so the hierarchical correspondence found for object recognition survives even at this low overall level. The same embeddings predict held-out appraisal ratings well, up to r = 0.87, and the two measures do not track each other across models; reweighting features towards the neural geometry lowers appraisal prediction for every model tested, against a control of matched dimensionality. Predicting how a street is appraised is therefore weak evidence that a model represents the street as the brain does. The benchmark uses only public data and requires no training, so evaluating a new representation needs only its embeddings for 55 images.

Geospatial foundation models have emerged as state-of-the-art methods for downstream Earth observation tasks. However, existing pretraining methodologies process imagery through a single-concept lens, failing to capture the highly compositional nature of complex satellite scenes. We propose a composition-aware pretraining framework that explicitly encodes fractional land-cover mixtures. Each satellite image cell is mapped to a histogram representing its fractional land-cover distribution, which we term the "composition target". These targets serve as the primary prediction objective and are distilled into the backbone using Earth Mover's Distance. Experimental evaluation shows that composition-aware pretraining yields substantial gains on region-level understanding tasks requiring semantic similarity judgment, including zero-shot image retrieval and scene classification, while remaining competitive on tasks requiring fine-grained spatial precision, such as segmentation and object detection. With a 36.8M-parameter backbone, our framework outperforms SatMAE and Prithvi-EO-2.0, which contain 303M and 600M parameters, respectively, in most retrieval and scene classification settings. On the fine-grained ForestNet-12 dataset, a rigorous testbed for compositional discrimination, our method boosts baseline mAP@10 from 0.279 to 0.434, a 55.6% relative improvement, providing direct evidence for the effectiveness of explicit composition modeling. The code implementation can be found at https://github.com/05kashyap/GFM_Composition_Pretraining

Global food security and sustainable climate action increasingly rely on robust, scalable agricultural monitoring. Earth observation foundation models have emerged as powerful, label-efficient tools across general remote sensing domains, yet early attempts to deploy them for agricultural applications have yielded surprisingly poor results. We hypothesize that this performance gap stems from the extreme heterogeneity of agricultural landscapes and the inherent inability of current earth observation foundation models to adapt to task-specific nuances. In this work, we systematically evaluate two critical bottlenecks hindering the deployment of foundation models in agricultural tasks, benchmarking two earth observation foundation models, a foundation model designed for tabular data, and conventional supervised baselines across seven real-world agricultural datasets spanning yield prediction, phenology estimation, and crop classification. First, we identify a pretraining-deployment modality gap: agricultural downstream tasks frequently require diverse, non-imagery data modalities that earth observation foundation models are architecturally unequipped to ingest, while a foundation model built for tabular data handles this heterogeneity more naturally. Second, we formalize the agricultural task space across five structural axes to demonstrate why current models fail to generalize reliably, resulting in highly unstable model rankings across evaluation settings. By characterizing these structural and modal gaps, our insights highlight the friction between general-purpose architectures and specialized agricultural downstream data, providing a strategic roadmap for developing the next generation of domain-aware foundation models.

Accurate uncertainty estimation is essential for machine learning systems de- ployed in high-stakes domains such as medicine. Traditional approaches primarily rely on probability outputs from trained models (point predictions), which provide no formal guarantees on prediction coverage and often require additional calibra- tion techniques to improve reliability. In contrast, conformal prediction (region prediction) offers a principled alternative by generating prediction sets with finite- sample validity guarantees, ensuring that the ground truth is contained within the set at a specified confidence level. In this study, we explore the impact of pre-training approach, dataset scale and domain on both point and region-level uncertainty quantification, by studying domain-specific vision medical foundation models vs. general domain vision foundation models. We conduct a comprehensive evaluation across foundation models trained on retinal, histopathological, and Chest X-Rays data, applying various calibration techniques. Our results demonstrate that (1) pre-training on higher-quality domain-specific datasets along with self-supervised learning leads to better-calibrated point predictions than general domain pre-training, (2) stan- dard re-calibration methods alone cannot fully mitigate uncertainty discrepancies across models trained on different data sources, (3) domain-specific foundation model can lead to more efficient conformal prediction. These findings highlight the importance of careful model selection and the inte- gration of both point and region prediction to enhance the reliability and trust- worthiness of medical AI systems. Our work underscores the need for a holistic approach to uncertainty quantification in recent development of medical vision foundation model, ensuring robust and interpretable AI-driven decision-making.

Vehicle attribute recognition is an important task in intelligent transportation systems, particularly when Automatic License Plate Recognition (ALPR) is unavailable or unreliable. Although vision foundation models have shown strong transferability across domains, their effectiveness for fine-grained vehicle classification remains underexplored. Moreover, given the inherently three-dimensional structure of vehicles, it is unclear whether emerging 3D-aware foundation models offer advantages over standard 2D architectures. This paper presents an empirical benchmark of 14 state-of-the-art 2D and 3D-aware vision foundation models. Using the challenging real-world UFPR-VeSV dataset, we evaluate these models as frozen feature extractors via linear probing for vehicle type, make, and model recognition. We further stress-test the best-performing models under few-shot learning and Out-of-Distribution (OOD) domain shifts. Our results show that standard 2D self-supervised models, particularly DINOv3, substantially outperform 3D-aware models in fine-grained tasks, achieving over 93% Macro-Accuracy for make and model recognition. However, the 3D-aware Depth Anything v2 exhibits stronger invariance to viewing angles in vehicle type classification. These findings motivate hybrid approaches that combine 2D and 3D priors for robust vehicle recognition. Our code is publicly available at https://github.com/UFPR-IPASPPR/3D-Vision-Benchmark/.

Visual foundation models are a cornerstone of image and video understanding but typically require large amounts of data and computation. The current scale required for pretraining visual foundation models may be unsustainable or unnecessary, and significant benefits arise when effective models can be obtained with fewer resources. To better understand how self-supervised learning (SSL) objectives behave under resource constraints, we conduct a controlled study of image and video SSL objectives under matched data, architecture, and compute budgets. We compare contrastive, reconstruction, feature-prediction, and diffusion objectives and evaluate both standalone and jointly trained image-video SSL formulations across a diverse set of image and video understanding tasks. Our results show that DINOv2-style pretraining consistently provides the strongest overall performance under limited resources. Furthermore, combining DINOv2 with video SSL objectives such as VideoMAE substantially improves image classification and segmentation performance, but degrades video tracking and camera-pose estimation performance, revealing an important tradeoff between semantic and geometric representation learning. These findings suggest that combining image and video SSL objectives can be beneficial in resource-limited settings, while highlighting the need for improved methods that better balance semantic, temporal, and geometric supervision.

Vision foundation models have recently emerged as powerful feature extractors for detecting AI-generated images, achieving strong generalization across generators and robustness to common image degradations. However, the reason behind their effectiveness is poorly understood. In this work, we investigate what cues are exploited by foundation-model-based detectors to distinguish real images from diffusion-generated ones. To this end, we design an ad hoc analysis protocol based on DDIM inversion. Given a real image we generate a sequence of synthetic copies by changing the depth of DDIM inversion. Even though most copies are semantically identical to the real reference, the detector score varies significantly across them due to subtle traces introduced by the diffusion synthesis, showing that its decision is not primarily driven by semantic failures. Through a frequency-swapping analysis, we further reveal that the discriminative cues exploited by the detectors are mainly localized in the low-to-mid frequency range, rather than only in the high-frequency range, as is the case for artifacts commonly associated with generative models. Finally, a latent-space analysis shows that regenerated images exhibit reduced variance and effective dimensionality, indicating that diffusion models do not fully reproduce the variability of real data. Overall, our results suggest that foundation-model-based detectors succeed by capturing non-semantic low-to-mid frequency distributional discrepancies between real and diffusion-generated images. These findings provide new insight into the robustness and generalization of such detectors and suggest directions for more interpretable forensic methods.

Air traffic controllers perceive traffic complexity through the radar display, suggesting that a computer vision model operating on the same imagery may provide a natural architecture for modeling controller-perceived complexity; however, whether radar imagery is a viable input format for deep learning vision models remains unclear. Unlike natural images, radar images are extremely sparse and self-similar, consisting primarily of a black background and a few visually identical aircraft blobs, while small changes in aircraft positions can substantially alter sector-level complexity. To test whether a vision model can capture these operationally important differences, we encode each traffic situation as a position image supplemented by five channels representing aircraft state variables, including heading, speed, and altitude, and train a Vision Transformer (ViT) to regress four intrinsic complexity components derived from pairwise geometric relations among aircraft. The model achieves $R^2 > 0.96$ for all four components, and a one-aircraft-removal perturbation study shows that its response changes proportionally to how much the removed aircraft contributed to sector complexity rather than treating every removal as equivalent. These results demonstrate that, despite its atypical visual characteristics, radar imagery is a viable input format for air traffic complexity modeling.

Cross-view feature matching aims to establish reliable correspondences across images with large viewpoint variations. Over the past decade, the field has evolved from task-specific models toward increasingly unified and generalizable correspondence models, with recent progress further driven by the emergence of vision foundation models (VFMs). Despite these advances, existing studies remain highly diverse in their problem formulations, model architectures, training paradigms, and evaluation protocols, making it difficult to obtain a unified understanding of the field. In this survey, we present a unified review of cross-view feature matching. We first introduce a structured taxonomy covering feature extraction, single-type feature matcher, multi-type feature matcher, VFMs based methods, training strategy and robust estimation, providing a coherent framework for analysis and comparison. We further examine recent advances, distilling key design principles and highlighting the shift toward unified and generalizable correspondence models. We also provide a unified experimental benchmarking of representative state-of-the-art methods under consistent protocols, enabling fair and comprehensive performance comparisons. In addition, we discuss open challenges and future directions, including efficiency, robustness under extreme conditions, and cross-domain generalization. This survey aims to provide a comprehensive and structured reference for understanding the evolution, current landscape, and future development of cross-view feature matching in the era of vision foundation models.

Spatio-temporal PV data are essential for understanding adoption processes in off-grid regions, yet such data remain largely unavailable. Automated segmentation of remote sensing (RS) imagery offers a promising solution; yet, residential PV systems remain challenging targets because of their small size and sparse distribution, resulting in severe target-background imbalance. Vision-language foundation models (FMs) provide a data-efficient paradigm through prompt-based semantic and spatial guidance, but the relative contribution of different prompt types remains unclear. We systematically evaluate SAM3 for small-scale PV segmentation in RS imagery by comparing textual, geometric, and hybrid prompting, under varying supervision levels, training strategies, spatial resolutions, and imaging conditions. Multi-temporal aerial imagery from a large off-grid rural region serves as a study site, with findings validated across three additional datasets. Prompting strategy emerged as the dominant factor governing model behavior. Textual prompting consistently produced the lowest performance and showed the greatest sensitivity to supervision and imaging conditions. In contrast, spatial guidance substantially improved both segmentation accuracy and robustness. Hybrid prompting achieved the highest accuracy and stability, indicating that semantic and spatial guidance provide complementary information. Most performance gains were achieved with only a few hundred annotated samples, demonstrating strong data efficiency. Transfer learning had limited overall impact, with only modest improvements observed for textual prompting under limited supervision. Overall, our findings establish prompting strategy as a key determinant of SAM3 adaptation, robustness, and generalization, highlighting the potential of promptable FMs for scalable PV mapping in data-constrained off-grid regions.

Recent Vision Foundation Models (VFMs) predict depth, camera pose, and pointmap in a single forward pass without per-scene optimization, achieving strong generalization. However, enforcing explicit multi-view geometric consistency, e.g., through bundle adjustment, is computationally costly and is thus not imposed during VFM pretraining, so such inconsistency can arise. To address this, implicit self-consistency derived from model outputs (e.g., pointmaps, features), though enforced at test-time in prior work, delivers inherently limited performance gain, especially on scenes where the pretrained VFM is highly inaccurate. In contrast to this implicit signal, we propose Self-Geometry, a plug-and-play test-time adaptation pipeline that directly imposes explicit multi-view geometric constraints using 2D pixel correspondences as pseudo ground-truth. Our proposed Self-Geometry consists of Geometric Disentanglement Optimization, which combines Multi-View Consistency and Epipolar Consistency losses with Gradient Disentanglement to prevent gradient conflict; Frame Angular-Neighbor, a view sampler based on SO(3) geodesic distances for lightly imposing these constraints; and Lightweight TTA, which adapts VFMs via LoRA. Our method achieves consistent improvements in both pose and geometry estimation across six VFMs (VGGT, $π^3$, DA3-Giant/Large/Base/Small) and four benchmarks (7Scenes, ETH3D, ScanNet++, HiRoom).

Leukemia cell image classification is challenged by real-world domain shifts from acquisition, staining, illumination, and site protocols, causing single-dataset models to generalize poorly in real clinical scenarios. This work presents a robust framework for leukemia classification across multiple heterogeneous datasets using a two-stage pipeline with a pretrained vision foundation model. Stage 1 performs binary classification (leukemia vs. non-leukemia) and is trained using 122,167 single-cell images. Stage 2 is conditionally applied to Stage 1 positives to perform subtype classification into Acute Lymphoblastic Leukemia (ALL) and Acute Myeloid Leukemia (AML), trained using 69,400 single-cell images. Labels are harmonized across five heterogeneous datasets to enable cross-dataset training, and performance is evaluated on a held-out dataset protocol to assess domain-shift generalization. Within this pipeline, three encoders are benchmarked (DinoBloom, pretrained on single-cell images; BiomedCLIP, pretrained on biomedical data; and CLIP as a general-purpose model) under linear probing, Low-Rank Adaptation (LoRA), and a Retrieval-Augmented Classification (RAC) module that retrieves the top-k most similar cell images to provide cytomorphological grounding. The objective is to quantify how much domain-specific pretraining contributes to performance under domain shift, and whether cost-effective adaptation and retrieval can be a viable alternative to expensive domain-specialized pretraining. The held-out protocol additionally serves as a diagnostic tool, revealing when classification performance is attributable to dataset-specific artifacts rather than to cytomorphological features.

Open-vocabulary remote sensing segmentation has recently emerged as a promising paradigm that enables pixel-level recognition of arbitrary categories specified by natural language, including classes unseen during training. However, geospatial domain shifts caused by heterogeneous regions, spatial resolutions, and acquisition platforms weaken visual-text matching and limit cross-dataset generalization. Recent attempts have begun to incorporate auxiliary vision foundation models (VFMs), typically coupling their features with text embeddings as additional matching evidence. However, this strategy may introduce inconsistent matching signals while leaving the structure-sensitive representations of VFMs insufficiently exploited. We therefore propose GeoSeg-OV, which decouples auxiliary VFM features from visual-text matching and repurposes them as structural guidance for cost aggregation and decoding. GeoSeg-OV constructs an orientation-robust cost volume from multi-rotation CLIP features, while a frozen VFM extracts multi-scale structure-sensitive features in parallel. We propose Structure-Guided Aggregation (SGA), which integrates cost tokens and CLIP semantic guidance with VFM-derived pairwise structural biases for coherent spatial propagation, followed by text-conditioned class-wise reasoning. We further introduce Cost-Aware Decoding (CAD) to adaptively refine and fuse multi-scale semantic and structural guidance based on the current decoder context. On the global High-Resolution Land Cover (HRLC) benchmark spanning seven datasets across six continents, GeoSeg-OV outperforms the state-of-the-art by +2.5 and +2.7 average mIoU under two training settings. A large-scale zero-shot case study further demonstrates its generalization across geographic domains and category systems without target-domain annotations or retraining.

Deep neural networks are increasingly deployed in safety-critical domains as perception modules, where failures are often caused due to rare and under-represented scenarios. This necessitates the need to evaluate the semantic robustness of perception models; conformance of behavior to high-level requirements over real-world perceptual variability. To address this, we propose SeFaR, a framework for systematic semantic-feature-centric testing of vision models. Given a natural-language requirement and a set of satisfying inputs, SeFaR evaluates robustness with respect to diverse realistic semantic variations that preserve requirement satisfaction. The approach employs a novel hierarchical concept model enabling structured exploration of the feature space and incorporation of domain knowledge via user-defined concepts. State-of-the-art diffusion and vision-language models are leveraged to generate photorealistic semantics-preserving perturbations and identification of previously unknown features impacting behavior. A feedback-driven adaptive process is adopted to generate interpretable failure-inducing semantic concepts along with corresponding test inputs. Evaluation on case studies demonstrates that the proposed framework effectively satisfies requirement preconditions while identifying requirement-independent features that influence model decisions, enabling it to both uncover faults and relate them to such features.

Human vision organizes what it sees into wholes: same-colored points group into series, similar marks cohere into categories, and shapes complete into recognizable objects. These are the Gestalt operations that visualization design builds on. Whether vision models organize visual content this way has not been systematically tested. We introduce a behavioral battery that scores models against human data from prior perception studies on four grouping tasks: mark-color odd-one-out, color-series counting, silhouette recognition, and object odd-one-out. We apply it to 45 models across five training families: supervised, self-supervised, and contrastive vision-language encoders, open-weight VLMs, and closed foundation models. The battery reveals that agreement with human responses captures aspects of perceptual organization that conventional performance metrics fail to distinguish, with several closed models exhibiting substantially lower alignment than their benchmark accuracy would suggest. Scoring against published perception data therefore gives visualization research a reusable yardstick, requiring no new user study, for auditing whether the models now entering visualization pipelines organize what they see the way their human audience does.

Face presentation attack detection (PAD) aims to reliably detect a wide range of presentation attacks. While PAD methods achieve strong performance within individual datasets, their performance degrades under cross-dataset evaluation. Variations in sensors or lighting conditions can reduce the effectiveness of detectors from near-perfect to nearly random. Foundation models (FMs) have emerged as a promising alternative because typical PAD datasets, such as the MCIO benchmarks (MSU-MFSD, CASIA-FASD, Replay-Attack, and OULU-NPU), are small relative to the scale used for web-based pretraining. However, existing PAD systems primarily focus on CLIP-based foundation models, while overlooking other FMs with different architectures and training procedures. This study addresses this question by systematically evaluating 32 FMs. Zero-shot prompting achieves performance near chance across model families and scales. The vision encoders, when low-rankadapted (LoRA) with fewer than 1% trainable weights, achieve below 2% intra-dataset ACER in most cases, while cross-dataset ACER is substantially higher. LoRA primarily refines the decision boundary within a dataset, suggesting that pretrained representations and the adaptation dataset play a larger role in cross-dataset generalization than the evaluated lightweight adaptation strategy.

Pretrained self-supervised representations have emerged as a core component of current deepfake detection methods, yet it remains unclear which of their properties make real and fake media distinguishable. In this work, we uncover a surprisingly consistent phenomenon: across multiple pretrained models, datasets, and both image and video domains, fake samples systematically produce lower-magnitude representations than their real counterparts. Motivated by this finding, we formulate deepfake detection as an anomaly detection problem and show that simple statistics of feature magnitude achieve competitive performance with far more sophisticated deepfake detection methods. We further investigate the origin of this effect and demonstrate that reduced feature magnitude is primarily associated with semantic shifts introduced by fake content, while low-level generative fingerprints play a comparatively smaller role. Finally, we show that this discriminative signal strengthens as the size of the underlying foundation model grows, suggesting that advances in representation learning naturally translate into stronger zero-shot deepfake detectors.

Newly triggered landslides rarely carry immediate annotations, so cross-domain transferability determines the value of landslide mapping for emergency response and regional risk assessment. Vision foundation models have strengthened representational transfer, yet on unseen regions, events, and data sources they still generate high-confidence false alarms. Terrain, material, and rainfall triggering can constrain such errors, but their supports are local, regional, and event-scale, so that resampling onto a 10~m grid misaligns them with the segmentation decision unit and compounds the uncertain geographic context problem (UGCoP). We propose GeoPhysAdapter, which anchors on a frozen vision foundation model, restricts terrain, material, and triggering to dense spatial guidance, regional modulation, and event-timing forcing, and applies bounded adaptation at two decision units, the pixel and the candidate landslide body, reverting exactly to the visual prediction where support is insufficient. On an event-isolated PILD dataset of four public sources, 55 global landslide events, and 7,890 test samples, 70.3% of cross-domain false-positive mass lies in near-pure spurious bodies of median equivalent diameter 207m, matching coarse-prior support rather than the pixel. Pixel-level adaptation removes a net 507,817 erroneous pixels and reduces error by 7.76%, whereas raising the decision unit to the candidate body, under identical samples, anchor, and baseline, increases error reduction to 23.99%, approximately 3.1 times the pixel-level effect, improves IoU by 0.031 (14.2% relative), and corrects 9.92 pixels per pixel harmed. The data and code are publicly available at: https://github.com/Liu-Zhihang/geophysadapter.

Semi-supervised adaptation of vision foundation models (VFMs) commonly freezes the pretrained backbone and updates lightweight modules such as LoRA. However, pseudo-labels have mixed reliability, and a single LoRA adapter must absorb reliable, ambiguous, and noisy gradients in the same low-rank space. This can make VFM adaptation sensitive to pseudo-label noise. We propose \textbf{TriNoL}, a \textbf{Tri}ple-expert learning framework from \textbf{No}isy \textbf{L}abels for semi-supervised VFM adaptation. TriNoL routes unlabeled samples into three confidence regions and assigns them to three LoRA experts: a Positive Expert for high-confidence pseudo-labels, an Alignment Expert for medium-confidence ambiguous samples, and a Negative Expert for low-confidence noisy samples. The VFM backbone remains frozen, and only the LoRA experts and classifier head are updated. By separating different pseudo-label reliability regions into specialized adaptation paths, TriNoL improves robustness to noisy supervision while keeping the training cost low.

Fine-grained damage classification of 3D point cloud data (PCD) remains a persistent challenge, constrained by high computational demands and limited labeled data. This study examines two methods: 3D PCD-based damage assessment (3PDA) algorithm and 2D projection damage assessment (2PDA) In our 3PDA analysis algorithm, TDA is used to derive compact representations of 3D PCD segmented by pointNet, which are then integrated with anomaly detection algorithms to quantify structural degradation. We show that TDA effectively compresses geometric structure from VFM-segmented components into discriminative feature vectors and that anomaly detection models can reliably distinguish components with varying damage severity using only 3D PCD inputs. In the 2D projection analysis algorithm, we leverage large VFMs for granular damage detection by projecting 3D PCD into 2D views. These projections allow VFM based models to achieve competitive classification performance while requiring only a fraction of the computational cost associated with full 3D data processing. Our results demonstrate that 2D VFM pipelines in 2PDA can perform strongly on fine-grained damage classification tasks, highlighting their viability as lightweight, resource-efficient alternatives to traditional 3PDA architectures. Comparative evaluation shows that the 3PDA attains higher accuracy but only for a narrow subset of object geometries and at substantially higher computational cost due to its reliance on TDA and the scarcity of high-fidelity 3D datasets. In contrast, the 2PDA algorithm yields slightly lower accuracy but offers an order of magnitude reduction in time complexity and generalizes across a far broader range of object categories.

Open-vocabulary semantic segmentation (OVSS) repurposes a pretrained CLIP encoder for dense prediction without additional labeled supervision. Existing methods improve CLIP's spatial behavior either by redesigning its internal attention or by injecting features from auxiliary vision foundation models; both require access to the host's internal computation and are tailored to its specific forward pass. In this work, we propose Test-time Prototype Adaptation (TPA), a training-free plug-in that operates at the output level, leaving the host's forward pass and weights unmodified. By leveraging a lightweight transductive adaptation phase, TPA identifies confident anchor patches from the host's own output predictions on a small pool of unlabeled deployment-domain images, and aggregates their frozen DINO features into per-class prototypes; at inference, a single cosine similarity lookup against this frozen bank provides an auxiliary score fused linearly with the host's logits. TPA composes with five representative OVSS hosts spanning attention-redesign and VFM-injection designs, across three CLIP backbones, eight benchmarks, and multiple internal VFM choices. Under a single set of hyper-parameters and without per-host tuning or parameter updates, TPA consistently improves segmentation accuracy, with as few as approximately 10% of unlabeled deployment-domain images sufficing for effective bank construction on most benchmarks.

Field-based agricultural computer vision is important for precision agriculture, yet it largely depends on expensive annotations and costly adaptation of large pretrained models. We introduce AgriField-40K, a field-centric dataset curated from 17 public resources and covering diverse crops, weeds, and field conditions. Building on this, we present AgriMAE, a parameter-efficient continual pretraining baseline that adapts a masked autoencoder pretrained on natural images by training only lightweight adapters. We further explore semantic feature reconstruction as an alternative pretraining objective and evaluate transfer across multiple tasks. AgriMAE consistently improves downstream performance and can match or even outperform full fine-tuning while using up to $9\times$ fewer trainable parameters, showing that AgriField-40K is a practical resource for continual pretraining in agricultural vision. Project page: https://dtu-pas.github.io/agrifield40k/

Parameter-efficient fine-tuning enables the adaptation of vision foundation models to biomedical tasks under limited computational resources, but a single low-rank update can constrain all task-specific changes to one narrow parameter subspace. This restriction may prevent the model from simultaneously representing globally shared task structure and localized residual directions required for generalization to unseen imaging domains. We introduce LoRSA, a global--residual adaptation framework that jointly learns a dense low-rank component and a dynamically structured-sparse low-rank component. The dense component captures globally coordinated task adaptation, while the structured component provides complementary residual corrections whose support evolves during training. We characterize the representational capacity, approximation properties, rank structure, and singular-subspace complementarity of this decomposition. We evaluate LoRSA for four-class breast-density classification using DINOv3-Base, with VinDr-Mammo as the source domain and MammosighTR and RSNA as unseen external domains. LoRSA remains competitive on the internal validation set and achieves the best external macro-F1 on both target datasets, improving upon the strongest competing method by 2.15 percentage points on MammosighTR and 3.09 percentage points on RSNA. Weight-matrix analysis further shows that approximately $92\%$ of the energy of each adaptation component lies outside the bilateral singular subspace of the other, indicating that the two components learn largely complementary update directions. These results suggest that organizing adaptation capacity into distinct global and residual paths can improve the external-domain generalization of parameter-efficiently adapted biomedical vision models.

Deep vision models exploit shortcuts, relying on cues that correlate with supervision signals. Prior work has focused on visible biases, such as object-background or texture correlations. We identify a different source of shortcut learning: invisible metadata traces embedded at the pixel level, for metadata such as image processing and photo acquisition. We hypothesize that large-scale semantic supervision, whether through categorical labels (ImageNet) or billion-scale captions (LAION), naturally induces metadata-semantics correlations during pretraining, leading models to convert low-level signals into predictive features. By introducing controlled metadata-semantics correlations, we show that stronger ones produce systematically higher sensitivity to metadata traces and larger performance degradation under metadata distribution shifts. We further explore mitigation strategies applied during and after pretraining that reduce sensitivity not only to targeted metadata but also to unseen ones, without sacrificing performance on downstream tasks. Metadata sensitivity also has a positive side: it partly explains the strong generated-image detection ability of some encoders, while its mitigation can improve out-of-distribution generalization. Code: https://github.com/ryan-caesar-ramos/visual-encoder-traces

Prompt tuning, a leading parameter-efficient adaptation paradigm in NLP, has recently been extended to computer vision. Visual prompt tuning (VPT) adapts pre-trained vision transformers (ViTs) by updating a small set of additional prompt parameters. However, existing visual prompts are randomly initialized and do not exploit prior knowledge, such as instructions in NLP. We address this gap by injecting two complementary semantic priors into VPT. Fundamental image priors, including color, texture, and shape, are extracted with classical hand-crafted operators and injected into the input space, while self-attention maps provide instance-aware semantics in the feature space. We further propose a cascaded scheme that integrates both priors throughout ViT adaptation. Experiments on 34 challenging image classification datasets demonstrate superior downstream adaptation while tuning only 0.74% of ViT parameters. Project page: https://xixiaouab.github.io/Cascaded-Semantics/.

Self-supervised Vision Foundation Models (VFMs) have become essential backbones for downstream tasks due to their strong and transferable visual representations. However, their patch-token-level features are often too coarse for dense prediction tasks such as semantic segmentation and depth estimation when accurate fine-grained predictions are required. Feature upsampling methods have been developed to recover pixel-level detail but still face limitations. Learnable upsamplers are often designed for a specific encoders and must be retrained for different encoders. Image-guided methods that use shallow pixel encoders often introduce textural artifacts and lack the semantic guidance needed for accurate downstream predictions. We introduce PixelUp, a zero-shot VFM-agnostic upsampler achieving semantic awareness through a coarse-to-fine chain of windowed cross-attention architecture guided by multi-scale semantic features. We demonstrate that PixelUp outperforms both VFM-specific and VFM-agnostic upsamplers, achieving state-of-the-art performance on dense prediction tasks with an average improvement of +1.2 mIoU on semantic segmentation and +0.25 $δ_1$, on NYUv2 depth estimation across VFMs. PixelUp further improves training-free open-vocabulary and unsupervised semantic segmentation by an average of +1.3 mIoU and +0.5 mIoU, respectively. Code available at https://pixelup-project.vercel.app/

Promptable segmentation foundation models (FMs) such as SAM3 and Medical SAM3 promise few-shot, interactively-specified segmentation for medical imaging through a natural language interface, yet their performance on clinical tasks falls well short of this promise. We posit that this shortfall is not an artefact of insufficient medical pretraining or imperfect prompt phrasing, but a structural limitation that will persist in any domain where paired image-text supervision is scarce, as it is across most clinical modalities. We further hypothesize that the limitation is specific to natural language as a control signal: a visually grounded prompt, learned directly from the target distribution, should recover the lost performance without additional image-text data or backbone retraining. We propose Few-Shot Concept Prompt Learning (FS-CPL), which learns a continuous concept prompt embedding $\mathbf{p}^* \in \mathbb{R}^{T \times d}$ from a small support set of $K$ image--mask pairs via mask supervision, with the encoder-decoder backbone frozen. Across four public benchmarks spanning ultrasound and endoscopy (BUSI, HC18, TN3K, CVC-Clinic), FS-CPL delivers absolute Dice improvements of up to $+0.62$ over canonical text prompts and is \emph{backbone-agnostic}: it lifts both vanilla SAM3 and the domain-specifically pretrained Medical SAM3, showing that visual concept prompting is complementary to in-domain pretraining.

Fine-grained visual representations are essential for medical image analysis, particularly when diagnostically relevant evidence is subtle and spatially localized. Modern transformer-based medical vision encoders must therefore learn patch-level representations that are both clinically meaningful and spatially consistent. Without these properties, large vision-language models (LVLMs) operate on an ambiguous visual foundation, limiting their ability to generate clinically reliable and spatially grounded responses. However, existing training strategies for medical vision encoders rarely achieve both objectives. Image-text alignment provides clinically meaningful supervision primarily at the image level, leaving the spatial localization of diagnostic evidence weakly constrained. In contrast, self-supervised learning promotes spatial consistency but lacks the semantic supervision needed to distinguish visually similar yet clinically distinct regions. To address this gap, we present LoFi, a medical vision foundation model built on location-aware fine-grained representation learning. LoFi trains a vision encoder with a lightweight large language model under grounding and grounded captioning objectives. Because these objectives require predicting location from clinical text and vice versa, spatial consistency emerges without any explicit patch-level regularization. To enable training at scale, we construct MedG, a large-scale medical grounding dataset of 4.48M image-text-box triplets curated from 84 datasets spanning 7 modalities. Across phrase grounding, visual question answering, and region-based organ classification under perturbations, LoFi consistently outperforms general-purpose and medical vision foundation models as well as state-of-the-art LVLMs. Code is available at https://github.com/myeongkyunkang/lofi-medg.

Frozen vision foundation models are commonly evaluated through a single global image embedding, but this interface can conflate missing information with information lost at readout time. We study this distinction by keeping a pretrained vision encoder frozen and varying only the readout applied to its final patch tokens. We compare standard global readouts against a lightweight foveated readout, which attention-pools patch tokens using a learned or question-conditioned query, and against an oracle readout with access to the annotated target region. We evaluate these interfaces on three localized binding problems: a controlled synthetic color--shape binding task under clutter, a color-free crowded shape-detection variant, and a GQA-derived natural-image task where paired questions ask for the colors of different same-category objects in the same image. Global readouts perform near perfectly when the synthetic target appears alone, but collapse under clutter and counterfactual target edits, whereas the foveated readout recovers most of the oracle-accessible signal. On the GQA-derived task, question-independent global image vectors improve only modestly over question-only priors, while question-conditioned foveation substantially improves paired localized color accuracy. A counterfactual nuisance-to-signal ratio explains the synthetic failures: global pooling dilutes localized label-changing evidence while exposing the probe to nuisance variation from irrelevant objects. These results indicate that apparent spatial blindness in frozen vision models can arise from the global embedding interface rather than from an absence of spatial information in the frozen patch tokens.

We present MeshFM, an efficient feedforward framework for extracting rich features from 3D inputs. Our method distills 2D features from visual foundation models into 3D. We train a feedforward network to directly predict 3D features without requiring optimization during inference. The approach utilizes a two-stage training strategy. First, we optimize a feature field in 3D using only 2D feature supervision. Second, we train a network to regress this feature field. The entire procedure requires no 3D annotation, instead relying on the powerful information in 2D foundation models. We demonstrate that our learned features can be immediately applied to downstream tasks, including part segmentation, dense correspondence, and mesh deformation. Extensive experiments show that MeshFM, trained solely with 2D supervision, performs on par with methods trained explicitly with 3D supervision, even without task-specific fine-tuning. Moreover, our model is trained to be robust to extreme rotations of the input objects. Project page: https://threedle.github.io/MeshFM/

Face presentation attack detection (PAD) remains challenging under cross-dataset evaluation, where domain shift degrades models trained on a single dataset. The scarcity of large-scale labeled data motivates adapting pretrained vision models rather than training task-specific architectures from scratch, raising a fundamental question: do general-purpose vision foundation models encode PAD-relevant information accessible with minimal task-specific training? To investigate, we systematically evaluate 24 frozen encoders, including self-supervised vision transformers, vision-language encoders, and supervised CNNs, using a unified linear-probing protocol on the MCIO benchmark (MSU-MFSD, CASIA-FASD, Replay-Attack, OULU-NPU). The backbone remains fixed, and only a lightweight linear head is trained to isolate the PAD information already present in the pretrained representation. Results show that frozen foundation-model representations can support strong intra-dataset PAD performance with only a linear classifier, but this performance does not reliably transfer across datasets. Model scale is beneficial within several families, although the effect is not monotonic and is strongly mediated by architecture and pretraining. InternViT-6B achieves the lowest mean intra-dataset error, whereas CLIP ViT-B/32 offers the most favorable cross-dataset transfer-compute trade-off among the evaluated probes. These findings suggest that while pretrained representations contain PAD-relevant information, explicit adaptation remains necessary to address domain shift.

Vision models do not form a representation at once; each block revises it. We ask whether the resulting computation path contains evidence that the final representation discards, and whether that evidence improves OOD detection and image classification on clean and shifted data. Unlike approaches that treat intermediate layers as separate snapshots, we retain sample identity across depth and study the transformations connecting successive states. We separate class-coherent transport from input-specific innovation, and coordinate movement from relational reorganization. Across supervised, self-supervised, vision--language, hierarchical, and convolutional encoders, these paths show strong sample-specific continuity and architecture-specific depth profiles that recur across datasets. They are also practically useful. An ID-only transition-surprise score complements strong final-state detectors, reducing FPR95 in 131/152 non-saturated comparisons on a balanced OpenOOD grid; gains are largest for visually disruptive and semantically far shifts, and remain positive on near-OOD for most detectors. Frozen update probes improve 71/72 clean model--dataset cases, while shifted-data gains vary with architecture and corruption type. Computation paths therefore provide a broadly useful reliability signal whose value is determined jointly by model organization and the shift encountered.

Contemporary machine learning struggles to learn continually, reuse prior knowledge, and expose a comprehensible internal structure. A recently proposed developmental, gradient-free learning framework addresses these limitations by learning a discrete, topological model of its inputs through local variation and selection, yielding an inherent continual-learning guarantee: new observations refine existing structure without overwriting past knowledge, and without replay buffers or predefined task boundaries. Its extension to visual inputs demonstrated this principle on shape recognition, but relied on a feature representation of limited expressivity that capped recognition accuracy. We introduce a new visual feature representation that encodes shape structure across multiple scales, capturing edge and contour features together with their spatial relations, and integrate it with the network-refinement learning process; we further improve the learning dynamics and the read-out used to predict from the learned model. The study targets two-dimensional shape, with class-incremental MNIST as a controlled, interpretable benchmark in which continual-learning behavior can be measured directly. Our approach substantially increases accuracy over the prior representation, matching or exceeding replay- and regularisation-based baselines at comparable storage while storing no past data, and preserves the framework's defining behavior: earlier-learned classes are retained as new ones are introduced, with no destructive adaptation, and the learned representations remain human-interpretable. What separates the methods is retention: the baselines surrender most of a just-trained class within its own cycle and relearn it afterwards, which ours does not. The significance lies in the manner of learning. The system integrates information one sample at a time while provably preserving its responses to...

Neurological disorders are a leading cause of global disability and are increasingly linked to environmental chemical exposures. Yet neurotoxicity assessment still relies on hand-scored morphological readouts that are subjective and poorly predictive of behavioral outcomes. Caenorhabditis elegans provides a genetically tractable, 3R-compliant alternative, but quantifying neuronal phenotypes from confocal microscopy at scale remains computationally challenging: existing vision foundation models, trained on natural or radiological images, cannot resolve the sparse signals and multi-scale lesions of neuronal imaging. Here, we introduce a dedicated self-supervised vision model for C. elegans dopaminergic neurons, together with CeNeuMorph, a multi-grained confocal benchmark of 27,117 annotated images. Specifically, moving beyond standard Masked Autoencoders, we propose a scale-adaptive masked image modeling strategy that jointly learns representations across resolutions and patch sizes under a fixed token budget. By decoupling structural semantic learning from rigid grid constraints, the model effectively resolves the full spectrum of neurodegenerative lesions - ranging from fine dendritic beading to gross soma shrinkage - within a tractable computational framework. Finally, our model surpasses both generalist and biomedical foundation models across classification, segmentation and detection tasks. Fusing visual features with morphological descriptors enables prediction of dopamine-dependent behavioral deficits ($R^2=0.498$). Screening 180 agrochemicals, we identify the benzimidazole moiety as a previously unrecognized determinant of dopaminergic neurotoxicity. Together, the work demonstrates how scale-adaptive self-supervised learning can connect morphology to function for a scalable alternative to mammalian in vivo models for neurotoxicity assessment and drug discovery.

Vision foundation models are increasingly reused as frozen backbones for downstream visual recognition, making parameter-efficient adaptation a central problem. Prompt-based adaptation, including Visual Prompt Tuning (VPT), provides a lightweight way to specialize these models, but its layer-wise behavior remains poorly understood: performance is sensitive to prompt depth, placement, and task distribution, and gains on standard in-domain benchmarks do not always translate into robust generalization. We argue that this limitation is not solely an optimization issue, but a layer-wise information allocation issue: existing prompt-based methods lack principled control over what prompt-conditioned representations should preserve, suppress, and propagate across depth. Inspired by the Information Bottleneck principle, we introduce Prompted Information Bottlenecks (PIB), a framework that regularizes layer-wise compression-sufficiency trade-offs and promotes a more coherent cross-layer information path. The key idea is that effective adaptation should be minimal yet sufficient, retaining task-relevant local evidence in earlier layers while progressively discarding nuisance factors and redundant details in deeper layers. Extensive experiments show that PIB achieves strong performance across 34 datasets, reaching 92.1% on FGVC, 93.01% on HTA, and 77.33% on VTAB-1k, while tuning only 0.35% parameters on average across the main settings. Beyond benchmark accuracy, PIB helps explain the non-monotonic behavior of prompt capacity scaling, reduces shortcut reliance, and improves robustness under distribution shift and fine-grained recognition settings. These results position PIB as both a practical method and an information-allocation perspective for adapting frozen vision foundation models. Our code is available at https://github.com/itsnotacie/MM-26-PIB

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.