Visual Representation Learning

Latent representations learned by neural networks often exhibit semantic structure, where concept similarity is reflected by geometric proximity in embedding space. However, comparing such spaces across models remains difficult: changes in architecture, pretraining data, objective, or random seed can yield embeddings with similar content but incompatible geometry. This latent space alignment problem is central to interpretability, transfer and multimodal learning, federated systems, and semantic communication; however, progress remains limited by the lack of large-scale, model-diverse, and metadata-rich benchmarks. To address this gap, we introduce SEMASIA, a large-scale collection of latent representations extracted from approximately 1,700 pretrained vision models across eight standard image-classification benchmarks. SEMASIA pairs embeddings with structured metadata describing architectures, training regimes, pretraining sources, and model scale. We demonstrate three applications of the resource. First, we analyze the conceptual organization of individual latent spaces, showing consistent prototype-like clustering and hierarchical semantic neighborhoods across models and datasets. Second, we benchmark supervised alignment mappings between latent spaces using reconstruction error and downstream task performance. Third, we perform a large-scale regression analysis of how pretraining-data complexity, specialization, transfer learning, augmentation, and model scale relate to geometric and probing properties of embeddings. By coupling representational scale with standardized metadata, SEMASIA provides a reproducible foundation for studying latent geometry, evaluating alignment methods, and developing next-generation heterogeneous and interoperable AI systems.

Self-supervised pretraining has become the mainstream approach for learning MRI representations from unlabeled scans. However, most existing objectives still treat each scan primarily as static aggregations of slices, patches or volumes. We ask whether there exists an intrinsic form of self-supervision signal that is different from reconstructing the masked patches, through transforming the 3D volumes into controllable 2D rendered sequences: by rendering slices at continuous positions, orientations, and scales, a 3D volume can be converted into dense video-action sequences whose controls are the action trajectories. We study this formulation with an action-conditioned pretraining objective, where a tokenizer encodes slice observations and a latent dynamics model predicts the evolution of latent features. Across representative anatomical and spatial downstream tasks, the proposed pretraining is evaluated against standard static-volume baselines, tokenizer-only pretraining, and dynamics variants without aligned actions. These results suggest that controllable MRI slice navigation provides a useful complementary pretraining interface for learning anatomical and spatial representations from large unlabeled MRI collections.

Video semantic search in densely crowded scenes remains a challenging task due to visual encoders tendency to prioritize salient foreground regions while neglecting contextually important, background areas. We propose an Inverse Attention Embedding mechanism that explicitly captures and highlights these overlooked regions. By combining inverse attention embeddings with traditional visual embeddings, our method significantly enhances semantic retrieval performance without additional training. Initial experiments and ablation studies demonstrate promising improvements over existing approaches in recall for video semantic search in crowded environments.

Domain generalization requires identifying stable representations that support reliable classification across domains. Domains may differ in low-level attributes, such as color, texture, or visual style, while preserving the same structural relationships among their underlying components. Existing methods primarily address these differences by improving the training process or aligning features across domains. However, since they leave this shared compositional structure implicit, they may overlook a more reliable source of invariance and consequently generalize less effectively to unseen domains. We propose Compositional Relational Invariance from Spatial Primitives (CRISP), an image classification framework that factors visual recognition into visual primitives and their relational composition. We represent these compositions using soft unary, binary, and ternary predicates over primitive locations and appearance, yielding differentiable measures of spatial and visual alignment that can be learned end-to-end. To learn primitives and relational structure jointly, we design an end-to-end architecture with three components: (1) a visual backbone that extracts generalized features, (2) a concept bottleneck layer that maps these features to primitive heatmaps with differentiable spatial coordinates, and (3) a structural scoring layer that evaluates candidate spatial relations among the detected primitives. Finally, we compute class probability from the joint evidence of its class-specific relational compositions and localized primitive appearance. We evaluate \method{} on five real-world image-classification datasets from the widely used DomainBed suite, covering shifts in depiction style, dataset provenance, and camera-trap location and achieving the new state-of-the-art on both benchmarks.

Unified visual tokenization faces a fundamental trade-off between high-fidelity pixel reconstruction (spatial equivariance) and semantic abstraction (conceptual invariance). We attribute this conflict to Manifold Misalignment: naive joint optimization induces opposing gradients, creating a zero-sum game between reconstruction and perception. To address this, we propose MUSE, a framework based on Topological Orthogonality. By treating Structure as an orthogonal bridge, MUSE decouples optimization within Transformers: structural gradients refine attention topology, while semantic gradients update feature values. This turns destructive interference into Mutual Reinforcement. Experiments show that MUSE breaks the trade-off, achieving state-of-the-art generation quality (gFID 3.08) and surpassing its teacher InternViT-300M in linear probing (85.2% vs. 82.5%), demonstrating that structurally aligned reconstruction can enhance semantic perception. Code is available at https://github.com/PanqiYang1/MUSE.

Artificial neural networks trained on visual tasks develop internal representations resembling those of the primate visual system, a discovery that has guided a decade of computational neuroscience. Research on building brain-aligned models has progressively embraced finer-grained learning ob- jectives, from object classification to contrastive self-supervised objectives that maximize distinc- tions among individual images. Yet the effect of learning-signal granularity on brain alignment remains largely unexamined. Here we systematically investigate how the granularity of a learning signal shapes representational alignment with human vision. We parametrically vary the number of training classes using a data-driven approach that partitions a set of training images into differ- ent numbers of categories via PCA-based splits of pretrained embeddings. We train hundreds of neural networks across convolutional and transformer architectures on these coarse classification tasks and compare their representations with human fMRI responses, macaque electrophysiology recordings, and human behavior. We find that networks trained to distinguish as few as eight broad categories learn representations that match or exceed the neural alignment of models distinguishing 1,000 classes. Even more strikingly, these coarsely trained networks align more closely with hu- man perceptual similarity judgments than all other models evaluated, including networks trained with fine-grained supervision or self-supervision as well as leading large-scale vision models. These results demonstrate that human-like visual representations can emerge from surprisingly simple learning objectives, reframing what learning signals vision may require and opening a path toward building AI systems that are more aligned with human perception.

In this paper, we explore deep learning techniques for individual identification of animals based on their skin patterns. Individual identification is crucial in biodiversity monitoring, since it enables analysis of decline or growth of populations, or intra-species interactions within populations. Models trained for the task of individual identification often do not focus on the skin pattern of animals, but on background details or body shape details. These characteristics are not individually specific, or can change drastically through time. We focus on techniques that will make machine learning models more responsive to skin pattern structure when extracting individual visual embeddings from images. For this, we explore image inpainting of task-specific masks as an auxiliary task to enhance ML-based individual identification from animal skin patterns. We propose a comparative analysis among four models as an encoder backbone for the individual identification task. We focus on the case study of zebrafish, which is a widely recognized biological model organism, and which exhibits individually identifying skin patterns. To evaluate encoder backbone performance, we present standard metrics for classification accuracy, embedding clustering metrics, and GradCAM visualizations.

EEG-based visual neural decoding aims to align neural responses with visual stimuli for tasks such as image retrieval. However, limited paired data and a fundamental mismatch between high-fidelity digital images and biological visual perception - distorted by retinotopic mapping and subject-specific neuroanatomy - severely impede cross-modal alignment. To address this, we propose MB2L, a Multi-Level Bidirectional Biomimetic Learning framework that incorporates structured physiological inductive biases into representation learning. Specifically, we propose Adaptive Blur with Visual Priors to mitigate perceptual-structural mismatch by reweighting visual inputs according to retinotopic priors. We further propose Biomimetic Visual Feature Extraction to learn multi-level visual representations consistent with hierarchical cortical processing, enhancing subject-invariant encoding. These modules are jointly optimized via Multi-level Bidirectional Contrastive Learning, which aligns EEG and visual features in a shared semantic space through bidirectional contrastive objectives. Experiments show MB2L achieves 80.5% Top-1 and 97.6% Top-5 accuracy on zero-shot EEG-to-image retrieval, significantly outperforming prior methods and demonstrating strong generalization across subjects and experimental settings.

Vision Transformers trained only on image classification routinely transfer to tasks that demand spatial understanding, yet they receive no spatial supervision during pretraining. We ask where and how robustly such structure is encoded. Probing a frozen ViT-B/16 layerwise for two complementary properties, local patch boundaries (BSDS500) and per-patch depth (NYU Depth V2), reveals a clear hierarchy: boundary structure becomes linearly decodable at layers 5-6 (AP = 0.833), while depth, which requires integrating global cues, peaks two to three layers later at layer 8 (MAE = 0.0875). Both signals collapse at the final classification layer, and random-weight controls confirm the encodings are learned rather than architectural. Causal interventions add specificity: ablating the single direction a linear depth probe reads degrades depth decoding by up to 165%, while ablating any other direction changes it by less than 1%. Targeted activation patching along that direction shows the depth signal is partially re-derived at each layer rather than passively carried in the residual stream, with mid-layer interventions persisting most strongly downstream. The result is that a classification-trained ViT develops an actively maintained spatial hierarchy that mirrors the early-to-late progression observed in the primate visual cortex.

Technical traders have long relied on visual analysis of candlestick charts to identify market patterns and predict price movements. While deep learning has achieved remarkable success in image classification, its application to financial chart images remains underexplored. This paper presents a systematic study comparing different visual representations for cryptocurrency regime prediction. We evaluate three image encoding methods (raw candlestick charts, Gramian Angular Fields, and multi-channel GAF), five chart component configurations, four neural network architectures (CNN, ResNet18, EfficientNet-B0, and Vision Transformer), and the impact of ImageNet transfer learning. Through eight controlled experiments on Bitcoin, Ethereum, and S&P 500 data spanning 2018-2024, we identify optimal configurations for visual regime classification. Our results show that a simple 4-layer CNN on raw candlestick charts achieves 0.892 AUC-ROC, outperforming larger pretrained models. Surprisingly, simpler representations (price-only charts, 128x128 resolution) consistently outperform more complex alternatives. We provide interpretability analysis using GradCAM and demonstrate that transfer learning improves performance by 4-16% despite the domain gap between natural images and financial charts.

Self-supervised learning has achieved remarkable success in learning visual representations from clean data, yet remains challenging when clean observations are sparse or not available at all. In this paper, we demonstrate that pretrained vision models can be leveraged to learn distortion-robust representations, which can then be effectively applied to downstream tasks operating on distorted observations. In particular, we propose an asymmetric knowledge distillation framework in which both teacher and student are initialized from the same pretrained Vision Transformer but receive different views of each image: the teacher processes clean images, while the student sees their distorted versions. We introduce multi-level distillation that aligns global embeddings, patch-level features, and attention maps and show that the student is able to approximate clean-image representations despite never directly accessing clean data. We evaluate our approach on image classification tasks across several datasets and under various distortions, consistently outperforming existing alternatives for the same amount of human supervision.

Visual Place Recognition (VPR) determines a query image's geographic location by matching it against geotagged databases. However, existing methods struggle with perceptual aliasing caused by irrelevant regions and inefficient re-ranking due to rigid candidate scheduling. To address these issues, we introduce FoL++, a method combining robust discriminative region modeling with adaptive re-ranking. Specifically, we propose a Reliability Estimation Branch to generate spatial reliability maps that explicitly model occlusion resistance. This representation is further optimized by two spatial alignment losses (SAL and SCEL) to effectively align features and highlight salient regions. For weakly supervised learning without manual annotations, a pseudo-correspondence strategy generates dense local feature supervision directly from aggregation clusters. Our Adaptive Candidate Scheduler dynamically resizes candidate pools based on global similarity. By weighting local matches by reliability and adaptively fusing global and local evidence, FoL++ surpasses traditional independent matching systems. Extensive experiments across seven benchmarks demonstrate that FoL++ achieves state-of-the-art performance with a lightweight memory footprint, improving inference speed by 40% over FoL. Code and models will be released (and merged with FoL) at https://github.com/chenshunpeng/FoL.

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

On-device continual learning (CL) is critical for edge AI systems operating on non-stationary data streams, but most existing methods rely on backpropagation or exemplar-heavy classifiers, incurring substantial compute, memory, and latency overheads. Hyperdimensional computing (HDC) offers a lightweight alternative through fast, non-iterative online updates. Combined with a compact convolutional neural network (CNN) feature extractor, HDC enables efficient on-device adaptation with strong visual representations. However, prior HDC-based CL systems often depend on multi-tier memory hierarchies and complex cluster management, limiting deployability on resource-constrained hardware. We present ImageHD, an FPGA accelerator for on-device continual learning of visual data based on HDC. ImageHD targets streaming CL under strict latency and on-chip memory constraints, avoiding costly iterative optimization. At the algorithmic level, we introduce a hardware-aware CL method that bounds class exemplars through a unified exemplar memory and a hardware-efficient cluster merging strategy, while incorporating a quantized CNN front-end to reduce deployment overhead without sacrificing accuracy. At the system level, ImageHD is implemented as a streaming dataflow architecture on the AMD Zynq ZCU104 FPGA, integrating HDC encoding, similarity search, and bounded cluster management using word-packed binary hypervectors for massively parallel bitwise computation within tight on-chip resource budgets. On CORe50, ImageHD achieves up to 40.4x (4.84x) speedup and 383x (105.1x) energy efficiency over optimized CPU (GPU) baselines, demonstrating the practicality of HDC-enabled continual learning for real-time edge AI.

Recent works show that image and video generators exhibit zero-shot visual understanding behaviors, in a way reminiscent of how LLMs develop emergent capabilities of language understanding and reasoning from generative pretraining. While it has long been conjectured that the ability to create visual content implies an ability to understand it, there has been limited evidence that generative vision models have developed strong understanding capabilities. In this work, we demonstrate that image generation training serves a role similar to LLM pretraining, and lets models learn powerful and general visual representations that enable SOTA performance on various vision tasks. We introduce Vision Banana, a generalist model built by instruction-tuning Nano Banana Pro (NBP) on a mixture of its original training data alongside a small amount of vision task data. By parameterizing the output space of vision tasks as RGB images, we seamlessly reframe perception as image generation. Our generalist model, Vision Banana, achieves SOTA results on a variety of vision tasks involving both 2D and 3D understanding, beating or rivaling zero-shot domain-specialists, including Segment Anything Model 3 on segmentation tasks, and the Depth Anything series on metric depth estimation. We show that these results can be achieved with lightweight instruction-tuning without sacrificing the base model's image generation capabilities. The superior results suggest that image generation pretraining is a generalist vision learner. It also shows that image generation serves as a unified and universal interface for vision tasks, similar to text generation's role in language understanding and reasoning. We could be witnessing a major paradigm shift for computer vision, where generative vision pretraining takes a central role in building Foundational Vision Models for both generation and understanding.

Ultra-fine-grained visual categorization (Ultra-FGVC) aims to classify highly similar subcategories within fine-grained objects using limited training samples. However, holistic yet discriminative cues, such as leaf contours in extremely similar cultivars, remain under-explored in current studies, thereby limiting recognition performance. Though crucial, modeling holistic cues with complex morphological structures typically requires massive training samples, posing significant challenges in data-limited scenarios. To address this challenge, we propose a novel Divide-and-Conquer Holistic Cognition Network (DHCNet) that implements a divide-and-conquer strategy by decomposing holistic cues into spatially-associated subtle discrepancies and progressively establishing the holistic cognition process, significantly simplifying holistic cognition while reducing dependency on training data. Technically, DHCNet begins by progressively analyzing subtle discrepancies, transitioning from smaller local patches to larger ones using a self-shuffling operation on local regions. Simultaneously, it leverages the unaffected local regions to potentially guide the perception of the original topological structure among the shuffled patches, thereby aiding in the establishment of spatial associations for these discrepancies. Additionally, DHCNet incorporates the online refinement of these holistic cues discovered from local regions into the training process to iteratively improve their quality. As a result, DHCNet uses these holistic cues as supervisory signals to fine-tune the parameters of the recognition model, thus improving its sensitivity to holistic cues across the entire objects. Extensive evaluations demonstrate that DHCNet achieves remarkable performance on five widely-used Ultra-FGVC datasets.

Identity-document fraud detection is not a stationary binary classification problem. Adaptive attackers modify templates and fabrication pipelines, making historical fraud labels stale, and successful forgeries recur at scale as coherent campaigns. We therefore study layout-aware representation learning for open-set fraud discovery rather than only closed-set classification. We adapt DINOv3 to the document domain via context-aware SimMIM fine-tuning and supervised metric learning with composite loss that encourages inter-class separability and intra-class compactness. The model is trained with U.S. IDs only. With a lightweight MLP and softmax classifier, the embedding achieves 99.83% layout classification accuracy on Canadian layouts. Moreover, on a dataset of 20,448 Canadian IDs, embedding-space analysis surfaces 276 adaptive physical-fraud cases, including 222 not surfaced by incumbent detectors. The embedding supports similarity-based expansion from a single confirmed seed to additional related cases not linked by conventional metadata graphs. The layout-aware document embeddings provide a production-aligned basis for discovering novel and campaign-scale fraud under distribution shift.

We present lightweight and efficient architectures to detect weather conditions from RGB images, predicting the weather type (sunny, rain, snow, fog) and 11 complementary attributes such as intensity, visibility, and ground condition, for a total of 53 classes across the tasks. This work examines to what extent weather conditions manifest as variations in visual style. We investigate style-inspired techniques, including Gram matrices, a truncated ResNet-50 targeting lower and intermediate layers, and PatchGAN-style architectures, within a multi-task framework with attention mechanisms. Two families are introduced: RTM (ResNet50-Truncated-MultiTasks) and PMG (PatchGAN-MultiTasks-Gram), together with their variants. Our contributions include automation of Gram-matrix computation, integration of PatchGAN into supervised multi-task learning, and local style capture through local Gram for improved spatial coherence. We also release a dataset of 503,875 images annotated with 12 weather attributes under a Creative Commons Attribution (CC-BY) license. The models achieve F1 scores above 96 percent on our internal test set and above 78 percent in zero-shot evaluation on several external datasets, confirming their generalization ability. The PMG architecture, with fewer than 5 million parameters, runs in real time with a small memory footprint, making it suitable for embedded systems. The modular design of the models also allows style-related or weather-related tasks to be added or removed as needed.

Modern machine learning models typically represent inputs as fixed points in a high-dimensional embedding space. While this approach has been proven powerful for a wide range of downstream tasks, it fundamentally differs from the way humans process information. Because humans are constantly adapting to their environment, they represent objects and their relationships in a highly context-sensitive manner. To address this gap, we propose a method for context-sensitive similarity computation from neural network embeddings, applied to modeling a triplet odd-one-out task with an anchor image serving as simultaneous context. Modeling context enables us to achieve up to a 15% improvement in odd-one-out accuracy over a context-insensitive model. We find that this improvement is consistent across both original and "human-aligned" vision foundation models.

Self-supervised learning has emerged as a major technique for the task of learning from unlabeled data, where the current methods mostly revolve around alignment of representations and input recon struction. Although such approaches have demonstrated excellent performance in practice, their scope remains mostly confined to learning from observed data and does not provide much help in terms of a learning structure that is predictive of the data distribution. In this paper, we study some of the recent developments in the realm of self-supervised learning. We define a new category called Predictive Representation Learning (PRL), which revolves around the latent prediction of unobserved components of data based on the observation. We propose a common taxonomy that classifies PRL along with alignment and reconstruction-based learning approaches. Furthermore, we argue that Joint-Embedding Predictive Architecture(JEPA) can be considered as an exemplary member of this new paradigm. We further discuss theoretical perspectives and open challenges, highlighting predictive representation learning as a promising direction for future self-supervised learning research. In this study, we implemented Bootstrap Your Own Latent (BYOL), Masked Autoencoders (MAE), and Image-JEPA (I-JEPA) for comparative analysis. The results indicate that MAE achieves perfect similarity of 1.00, but exhibits relatively weak robustness of 0.55. In contrast, BYOL and I-JEPA attain accuracies of 0.98 and 0.95, with robustness scores of 0.75 and 0.78, respectively.

Object binding is a foundational process in visual cognition, during which low-level perceptual features are joined into object representations. Binding has been considered a fundamental challenge for neural networks, and a major milestone on the way to artificial models with flexible visual intelligence. Recently, several investigations have demonstrated evidence that binding mechanisms emerge in pretrained vision models, enabling them to associate portions of an image that contain an object. The question remains: how are these models binding objects together? In this work, we investigate whether vision models rely on the principle of Gestalt continuity to perform object binding, over and above other principles like similarity and proximity. Using synthetic datasets, we demonstrate that binding probes are sensitive to continuity across a wide range of pretrained vision transformers. Next, we uncover particular attention heads that track continuity, and show that these heads generalize across datasets. Finally, we ablate these attention heads, and show that they often contribute to producing representations that encode object binding.

We introduce an unsupervised visual representation learning system based entirely on local plasticity rules, without labels, backpropagation, or global error signals. The model is a VisNet-inspired hierarchical architecture combining opponent color inputs, multi-frequency Gabor and wavelet feature streams, competitive normalization with lateral inhibition, saliency modulation, associative memory, and a feedback loop. All representation learning occurs through continuous local plasticity applied to unlabeled image streams over 300 epochs. Performance is evaluated using a fixed linear probe trained only at readout time. The system achieves 80.1 percent accuracy on CIFAR-10 and 47.6 percent on CIFAR-100, improving over a Hebbian-only baseline. Ablation studies show that anti-Hebbian decorrelation, free-energy inspired plasticity, and associative memory are the main contributors, with strong synergistic effects. Even without learning, the fixed architecture alone reaches 61.4 percent on CIFAR-10, indicating that plasticity, not only inductive bias, drives most of the performance. Control analyses show that independently trained probes match co-trained ones within 0.3 percentage points, and a nearest-class-mean classifier achieves 78.3 percent without gradient-based training, confirming the intrinsic structure of the learned features. Overall, the system narrows but does not eliminate the performance gap to backpropagation-trained CNNs (5.7 percentage points on CIFAR-10, 7.5 percentage points on CIFAR-100), demonstrating that structured local plasticity alone can learn strong visual representations from raw unlabeled data.

Frozen visual embeddings (e.g., CLIP, DINOv2/v3, SSCD) power retrieval and integrity systems, yet their use on face-containing data is constrained by unmeasured identity leakage and a lack of deployable mitigations. We take an attacker-aware view and contribute: (i) a benchmark of visual embeddings that reports open-set verification at low false-accept rates, a calibrated diffusion-based template inversion check, and face-context attribution with equal-area perturbations; and (ii) propose a one-shot linear projector that removes an estimated identity subspace while preserving the complementary space needed for utility, which for brevity we denote as the identity sanitization projection ISP. Across CelebA-20 and VGGFace2, we show that these encoders are robust under open-set linear probes, with CLIP exhibiting relatively higher leakage than DINOv2/v3 and SSCD, robust to template inversion, and are context-dominant. In addition, we show that ISP drives linear access to near-chance while retaining high non-biometric utility, and transfers across datasets with minor degradation. Our results establish the first attacker-calibrated facial privacy audit of non-FR encoders and demonstrate that linear subspace removal achieves strong privacy guarantees while preserving utility for visual search and retrieval.

Pretrained Vision Transformers (ViTs) such as DINOv2 and MAE provide generic image features that can be applied to a variety of downstream tasks such as retrieval, classification, and segmentation. However, such representations tend to focus on the most salient visual cues in the image, with no way to direct them toward less prominent concepts of interest. In contrast, Multimodal LLMs can be guided with textual prompts, but the resulting representations tend to be language-centric and lose their effectiveness for generic visual tasks. To address this, we introduce Steerable Visual Representations, a new class of visual representations, whose global and local features can be steered with natural language. While most vision-language models (e.g., CLIP) fuse text with visual features after encoding (late fusion), we inject text directly into the layers of the visual encoder (early fusion) via lightweight cross-attention. We introduce benchmarks for measuring representational steerability, and demonstrate that our steerable visual features can focus on any desired objects in an image while preserving the underlying representation quality. Our method also matches or outperforms dedicated approaches on anomaly detection and personalized object discrimination, exhibiting zero-shot generalization to out-of-distribution tasks.

When evaluating identity-focused tasks such as personalized generation and image editing, existing vision encoders entangle object identity with background context, leading to unreliable representations and metrics. We introduce the first principled framework to address this vulnerability using Near-identity (NearID) distractors, where semantically similar but distinct instances are placed on the exact same background as a reference image, eliminating contextual shortcuts and isolating identity as the sole discriminative signal. Based on this principle, we present the NearID dataset (19K identities, 316K matched-context distractors) together with a strict margin-based evaluation protocol. Under this setting, pre-trained encoders perform poorly, achieving Sample Success Rates (SSR), a strict margin-based identity discrimination metric, as low as 30.7% and often ranking distractors above true cross-view matches. We address this by learning identity-aware representations on a frozen backbone using a two-tier contrastive objective enforcing the hierarchy: same identity > NearID distractor > random negative. This improves SSR to 99.2%, enhances part-level discrimination by 28.0%, and yields stronger alignment with human judgments on DreamBench++, a human-aligned benchmark for personalization. Project page: https://gorluxor.github.io/NearID/

Modern vision pipelines increasingly rely on pretrained image encoders whose representations are reused across tasks and models, yet these representations are often overcomplete and model-specific. We propose a simple, training-free method to improve the efficiency of image representations via a post-hoc canonical correlation analysis (CCA) operator. By leveraging the shared structure between representations produced by two pre-trained image encoders, our method finds linear projections that serve as a principled form of representation selection and dimensionality reduction, retaining shared semantic content while discarding redundant dimensions. Unlike standard dimensionality reduction techniques such as PCA, which operate on a single embedding space, our approach leverages cross-model agreement to guide representation distillation and refinement. The technique allows representations to be reduced by more than 75% in dimensionality with improved downstream performance, or enhanced at fixed dimensionality via post-hoc representation transfer from larger or fine-tuned models. Empirical results on ImageNet-1k, CIFAR-100, MNIST, and additional benchmarks show consistent improvements over both baseline and PCA-projected representations, with accuracy gains of up to 12.6%.

Superpixels provide a compact region-based representation that preserves object boundaries and local structures, and have therefore been widely used in a variety of vision tasks to reduce computational cost. However, most existing superpixel algorithms produce irregularly shaped regions, which are not well aligned with regular operators such as convolutions. Consequently, superpixels are often treated as an offline preprocessing step, limiting parallel implementation and hindering end-to-end optimization within deep learning pipelines. Motivated by the adaptive representation and coverage property of granular-ball computing, we develop a square superpixel generation approach. Specifically, we approximate superpixels using multi-scale square blocks to avoid the computational and implementation difficulties induced by irregular shapes, enabling efficient parallel processing and learnable feature extraction. For each block, a purity score is computed based on pixel-intensity similarity, and high-quality blocks are selected accordingly. The resulting square superpixels can be readily integrated as graph nodes in graph neural networks (GNNs) or as tokens in Vision Transformers (ViTs), facilitating multi-scale information aggregation and structured visual representation. Experimental results on downstream tasks demonstrate consistent performance improvements, validating the effectiveness of the proposed method.

Stereo needs features that preserve fine cross view correspondence rather than only semantic similarity. Recent self supervised vision models transfer well, but they are not built for this goal, and geometry focused methods often rely on a binocular decoder or another explicit linkage module during pretraining. BINO asks whether strong binocular structure can instead be learned inside a compact encoder. It does this by fusing the rectified pair at the input stage, forming stereo micro cell tokens, and using a row aware patch phase positional encoding. Training uses one view masked token only distillation together with occlusion and view specific appearance mismatch. In a strict low resource setting with pretraining only on KITTI object, BINO gives the best frozen descriptor results under a no linkage probe among all compared baselines on proxy dense stereo, hard negative retrieval, and KITTI Stereo2012 disparity. With the same lightweight stereo head for every encoder, it stays near CroCov2 while using a much smaller encoder. Supplementary transfer experiments on KITTI Stereo~2015 show the same qualitative trend. These results suggest that much of the cross view reasoning often assigned to a separate linkage module can be learned inside a compact and reusable encoder.

Computational pathology relies on effective representation learning to support cancer research and precision medicine. Although self-supervised learning has driven major progress at the patch and whole-slide image levels, representation learning at the single-cell level remains comparatively underexplored, despite its importance for characterizing cell types and cellular phenotypes. We introduce LEMON (Learning Embeddings from Morphology Of Nuclei), a self-supervised foundation model for scalable single-cell image representation learning. Trained on millions of cell images from diverse tissues and cancer types, LEMON learns robust and versatile morphological representations that support large-scale single-cell analyses in pathology. We evaluate LEMON on five benchmark datasets across a range of prediction tasks and show that it provides strong performance, highlighting its potential as a new paradigm for cell-level computational pathology. Model weights are available at https://huggingface.co/aliceblondel/LEMON.

Vision Foundation Models (VFMs) have become the cornerstone of modern computer vision, offering robust representations across a wide array of tasks. While recent advances allow these models to handle varying input sizes during training, inference typically remains restricted to a single, fixed scale. This prevalent single-scale paradigm overlooks a fundamental property of visual perception: varying resolutions offer complementary inductive biases, where low-resolution views excel at global semantic recognition and high-resolution views are essential for fine-grained refinement. In this work, we propose Multi-Resolution Fusion (MuRF), a simple yet universally effective strategy to harness this synergy at inference time. Instead of relying on a single view, MuRF constructs a unified representation by processing an image at multiple resolutions through a frozen VFM and fusing the resulting features. The universality of MuRF is its most compelling attribute. It is not tied to a specific architecture, serving instead as a fundamental, training-free enhancement to visual representation. We empirically validate this by applying MuRF to a broad spectrum of critical computer vision tasks across multiple distinct VFM families - primarily DINOv2, but also demonstrating successful generalization to contrastive models like SigLIP2.