arXiv subjects

Shai Avidan

Publications and source records attributed to Shai Avidan.

At least 19 recordsLinked to original sources

arXiv2026

Object-Only Is Not Context-Free: A Counterfactual Diagnostic for Context Removal in Object Retrieval

Object-centric retrieval pipelines often mask the background before pooling backbone features, aiming for a descriptor that reflects the object rather than its scene. Whether this truly yields an object-only descriptor, however, is rarely tested. We introduce the Keyhole Aperture Probe (KAP), a counterfactual diagnostic that repurposes the ground-truth masks of existing panoptic segmentation datasets (COCO, ADE20K) for retrieval: it holds the pooled object region fixed and varies only the pixels outside it, summarizing a backbone's context dependence in a single number. We find that the pooled object region has already integrated its surround. Replacing the context, with a neutral fill or even a different real scene, changes the descriptor, and when retrieval fails under a blank fill, the descriptor matches plain-background images rather than its own object. Object-only is not context-free. Across ten modern backbones, object-only descriptors retain only about a quarter to two-thirds of their full-image accuracy, and the effect is not unique to Transformer self-attention: a convolutional backbone shows it too. The dependence is also reducible: a lightweight adapter that encourages the descriptor to stay stable as context is removed lowers it and improves full-image instance retrieval on INSTRE, transferring to ILIAS with accuracy improved or preserved. A matched contrastive-fine-tuning control does not reproduce those gains. We release KAP as a simple, backbone-agnostic evaluation protocol to measure and reduce this dependence before deployment.

cs.CV↗

arXiv2026

SoftCache: Adaptive Step Size Caching for Diffusion Transformers

Diffusion and flow models generate a sample over iterative steps, each a full pass through a large network, and that pass dominates generation cost. Training-free caching accelerates them by reusing activations from recent evaluations instead of recomputing them. Dynamic caching methods decide when to refresh by accumulating a cheap probe of the network input into a staleness signal and spending a fresh evaluation once it crosses a threshold. They perform hard caching: the signal is a binary switch between refresh and reuse, and every step, fresh or cached, advances by the prescribed step size. We observe that the signal carries much more than this switch. How far it sits below the threshold, its headroom, predicts the error induced by reuse. We propose SoftCache, which performs soft caching: instead of a binary switch, the headroom sets how far each cached step goes. Where the headroom is large, the sampler jumps past several scheduled nodes, and as a refresh nears, its steps shrink back toward the original step. These adaptive jumps may land between the scheduled nodes, taking the sampler off its fixed grid. SoftCache requires no training and adds no per-backbone settings: on FLUX for images and on HunyuanVideo and Wan2.1 for video, at speedups from $3.5\times$ to $6.9\times$, it is ahead of every baseline on all four fidelity metrics.

cs.CV↗

arXiv2026

Web Scale on a Budget: Tail-Aware Retrieval Evaluation

A common practice in instance retrieval is to grow the gallery size to improve a benchmark's discrimination capability. One prominent example is ILIAS, which adds up to 100 million distractor images. The prevailing assumption is that faithful evaluation demands web scale, and that indexing all 100M images through the backbone is the unavoidable price of a comparable number. That price is steep: hundreds of GPU-hours and ~200 GB of descriptors per model, paid again for every method or ablation. We show the web-scale indexing price is largely avoidable: evaluation itself obeys a scaling law. The number of distractors that outrank a positive is Binomial in the gallery size. We define the per-query retrieval hardness, a gallery-scale agnostic quantity that yields a retrieval metric at any scale, and demonstrate it can be derived from a small random sample. The underlying sampled-rank machinery is long established in recommender-system and biometric evaluation, but to our knowledge has never been adapted to visual retrieval. We make the law per-query and estimable, and validate it at true 100M scale on ILIAS and INSTRE, and on the full 759k-distractor Google Landmarks v2 pool. Concretely, within a 5% sample (a flat 20x less indexing), every backbone's tail-aware (EVT) estimate has converged onto the large-scale mAP, to within ~0.01 (mean absolute error 0.005), across five backbones, three benchmarks, and two distractor pools (mAP@1000 at 100M on ILIAS/INSTRE; the official mAP@100 at the full 759k pool on GLDv2); the recommender-systems count baseline over-predicts everywhere (+0.037 to +0.068). This is an order of magnitude better than naively reporting the metric on the same subset (+0.08 to +0.13). We release code and the per-query hardness to facilitate faster, cheaper, and greener retrieval evaluation.

cs.CV↗

arXiv2026

Texture Interpolation as Texton Matching

Texture interpolation asks what should lie between two textures. Standard methods typically blend pixels or features at corresponding spatial locations, implicitly assuming that co-located elements should be mixed. For unrelated textures, however, this correspondence is arbitrary: a crack at one location has no particular relationship to a fiber at the same location in another image. We study whether this hidden pairing choice is itself a major determinant of interpolation quality. We represent textures as collections of local appearance vectors, or textons, and replace positional pairing with appearance-aware matching while keeping the encoder, representation, and decoder fixed. This isolates the effect of correspondence from changes in model capacity or training. Across texture pairs, appearance-aware matching consistently improves the distributional quality of interpolated midpoints over positional feature blending. Soft matching produces the strongest midpoint statistics, whereas one-to-one assignments yield more balanced transitions and closer agreement with the target endpoint. The same intervention also improves a second pretrained texture interpolator without retraining. These results identify correspondence as a first-class design choice in texture interpolation: meaningful transitions depend not only on how strongly two textures are mixed, but also on which local appearances are mixed with one another.

cs.CV↗

arXiv2026

Beyond Resolution: Object-to-Image Ratio Mismatch in Instance Retrieval

Visual instance retrieval often fails when the same object appears at different apparent sizes in the query and gallery. We show that the dominant cause is usually not resolution loss but object-to-image (O2I) ratio mismatch: the object occupies different fractions of the two images. On a controlled benchmark of 3,021 Objaverse objects rendered at five camera distances, more than 80% of the cross-distance degradation is attributable to O2I mismatch rather than resolution for 9 of 12 pretrained backbones; multi-scale architectures cut the resolution-only effect to single digits yet remain equally susceptible. The failure is also asymmetric: tight queries retrieve more reliably against wide gallery images than the reverse. Guided by this analysis, query-side scale augmentation and an OWLv2 crop reranker reach state of the art on ILIAS 100M (29.2 mAP@1000 before reranking, 42.0 after) without training or modifying the precomputed gallery index, and a LoRA fine-tune matches the query-side gains at a single forward pass, showing that O2I robustness is learnable.

cs.CV↗

arXiv2026

Scalable Black-Box Model Attribution for Images

The rapid proliferation of generative models raises the model attribution problem: given only an image, can we determine which model produced it? We propose a lightweight CNN to solve this problem in a strict black box setting. The CNN operates on multiple image patches to handle varying image size and improve accuracy. It attributes more models at higher accuracy than prior work, reaching 98.9% on 25-class DRAGON and 95.0% on 27-class OpenFake; runs in a few milliseconds at a cost nearly independent of candidate-set size; and remains robust to transformations encountered in the wild. Beyond closed-set attribution, its learned representation serves as a reusable fingerprint backbone: it supports open-set rejection and few-shot enrollment, recovers model-family structure without lineage supervision, and groups images from unseen generators by source. Controlled ablations and causal perturbations show that its attribution decisions are driven by a spatially local, low-level signal that behaves as a generator-specific fingerprint.

cs.CV↗

arXiv2026

Multi-View Foundation Models

Foundation models are vital tools in various Computer Vision applications. They take as input a single RGB image and output a deep feature representation that is useful for various applications. However, in case we have multiple views of the same 3D scene, they operate on each image independently and do not always produce consistent features for the same 3D point. We propose a way to convert a Foundation Model into a Multi-View Foundation Model. Such a model takes as input a set of images and outputs a feature map for each image such that the features of corresponding points are as consistent as possible. This approach bypasses the need to build a consistent 3D model of the features and allows direct manipulation in the image space. Specifically, we show how to augment Transformers-based foundation models (i.e., DINO, SAM, CLIP) with intermediate 3D-aware attention layers that help match features across different views. As leading examples, we show surface normal estimation and multi-view segmentation tasks. Quantitative experiments show that our method improves feature matching considerably compared to current foundation models.

cs.CV↗

arXiv2026

Ill-Posed by Design: Probing Evidence Use in VLMs

Counterfactual analysis is widely used to study evidence use in vision-language models, but its diagnostic value is limited on well-posed tasks: when several cues independently support the same answer, removing one may not change the prediction. We propose monocular metric object-size estimation as an ill-posed diagnostic setting for evidence selection: because physical size cannot be determined from a single uncalibrated image, models must rely on imperfect cues category priors, target appearance, local context, apparent image size, and scene geometry. We assemble Metric VQA ($10{,}813$ dimension queries from Objectron and $331$ tape-measured in-the-wild scenes) and evaluate $12$ open-weight VLMs ($3$--$397$\,B parameters) with counterfactual analysis decomposing six visual and language evidence channels. Even the largest VLMs tested (Qwen3-VL-235B, Qwen3.5-397B, InternVL3.5-241B) trail a text-only frontier LLM on the in-the-wild split. The diagnostic analysis shows: target identity is the most load-bearing cue, target pixels and local context help only some models, apparent size shifts predictions without a directional readout, and global scene geometry is largely unused. We analyze LoRA fine-tuning as an actionable intervention specific to metric estimation: while the task is learnable, the models do not learn to leverage scene geometry.

cs.CV↗

arXiv2026

Sub-Semantic Image Segmentation

Images can be segmented based on visual cues (i.e., texture segmentation) or into objects (i.e., semantic segmentation). We propose a new category of sub-semantic image segmentation that blurs the line between the two. In sub-semantic image segmentation, language is not used to name whole objects. Instead, it is used to partition an image into stable appearance patterns that can be described by language. To do that, we couple a general-purpose vision-language model to SAM 3, a promptable segmentation backbone whose native text pathway can ground rich descriptions into masks. Simple coupling fails for a number of reasons that we identify in the paper, and we overcome them by introducing DETECTURE that resolves three concrete failure modes -- language leakage between texture regions, prompt competition inside the segmentation backbone, and semantic distortion at the language-to-mask interface. Since there is no dataset of sub-semantic image segmentation, we introduce one, termed TextureADE. The new dataset is derived from the ADE20K dataset using a system we designed. We compare DETECTURE to a number of baselines and find that it achieves the strongest performance on several datasets using different metrics. Code is available at https://github.com/Scientific-Computing-Lab/TextureDetecture.

cs.CV↗

arXiv2026

Where Does Texture Evidence Live in SAM? Features, Proposal Masks, and Texture Segmentation

Texture segmentation stresses foundation segmentation because meaningful regions are defined by material or repeated appearance rather than object identity. Segment Anything Models (SAMs) often fail by default on such texture-defined partitions, but this failure is ambiguous: the texture evidence may be absent, missing from the proposal bank, or present but selected or assembled incorrectly by an object-centric readout. We ask what texture-relevant evidence is already preserved in frozen SAM before adaptation. We study two frozen evidence spaces: multiscale features, probed with a minimal clustering readout, and the automatic proposal bank, treated as evidence for a supervised consolidation readout. SAM is frozen throughout; we do not fine-tune the backbone or retrain the proposal generator. Across RWTD, STLD, an ADE20K-selected refined-crop complement, and a ControlNet-stitched PTD bridge archive, frozen SAM is not a texture segmenter by default, but its failures are not simple texture blindness. Coarse frozen features preserve texture organization, and proposal banks often contain texture-aligned masks or fragments. Natural scenes more often require assembly and commitment over fragments, while cleaner synthetic cases more often reduce to selecting an already coherent proposal. Default mask failure should therefore be decomposed into representation evidence, proposal-bank support, readout mismatch, and commitment failure.

cs.CV↗

arXiv2026

Optimal Transport Flow Matching by Design

Flow matching models learn to transport samples from a simple prior distribution to a complex data distribution. When prior-data pairs are coupled via optimal transport (OT), the learned trajectories are straight and non-crossing, enabling fast, even single-step, generation. However, computing the OT coupling in high dimensions is intractable, and existing methods attempt to solve the OT problem, at the cost of persistent bias or significant overhead. Rather than solving for the OT coupling, we reformulate the problem. Once the prior is treated as a design choice rather than a fixed input, the OT coupling between prior and data is no longer unique. Many priors admit an OT-optimal identity coupling to the data, leaving us free to choose one that is also tractable to sample. We identify low-frequency projection of natural images as such a choice. The identity coupling between data and its low-frequency representation is empirically OT-optimal, the prior is structured enough to be sampled by a lightweight model at inference, and the remaining flow-matching task reduces to synthesizing high-frequency detail. Interpolating the prior with Gaussian noise further improves generation quality while preserving the OT coupling. The approach requires no modifications to the flow model itself, and integrates naturally with latent-space models, classifier-free guidance, and one-step generation frameworks. Across all benchmarks, our method reduces trajectory curvature by more than $2\times$ compared to existing flow matching methods, yielding better generation quality in the few-step regime.

cs.CV↗

arXiv2026

Scene Grounding In the Wild

Reconstructing accurate 3D models of large-scale real-world scenes from unstructured, in-the-wild imagery remains a core challenge in computer vision, especially when the input views have little or no overlap. In such cases, existing reconstruction pipelines often produce multiple disconnected partial reconstructions or erroneously merge non-overlapping regions into overlapping geometry. In this work, we propose a framework that grounds each partial reconstruction to a complete reference model of the scene, enabling globally consistent alignment even in the absence of visual overlap. We obtain reference models from dense, geospatially accurate pseudo-synthetic renderings derived from Google Earth Studio. These renderings provide full scene coverage but differ substantially in appearance from real-world photographs. Our key insight is that, despite this significant domain gap, both domains share the same underlying scene semantics. We represent the reference model using 3D Gaussian Splatting, augmenting each Gaussian with semantic features, and formulate alignment as an inverse feature-based optimization scheme that estimates a global 6DoF pose and scale while keeping the reference model fixed. Furthermore, we introduce the WikiEarth dataset, which registers existing partial 3D reconstructions with pseudo-synthetic reference models. We demonstrate that our approach consistently improves global alignment when initialized with various classical and learning-based pipelines, while mitigating failure modes of state-of-the-art end-to-end models.

cs.CV↗

arXiv2026

Edge Weight Prediction For Category-Agnostic Pose Estimation

Category-Agnostic Pose Estimation (CAPE) localizes keypoints across diverse object categories with a single model, using one or a few annotated support images. Recent works have shown that using a pose graph (i.e., treating keypoints as nodes in a graph rather than isolated points) helps handle occlusions and break symmetry. However, these methods assume a static pose graph with equal-weight edges, leading to suboptimal results. We introduce EdgeCape, a novel framework that overcomes these limitations by predicting the graph's edge weights which optimizes localization. To further leverage structural priors, we propose integrating Markovian Structural Bias, which modulates the self-attention interaction between nodes based on the number of hops between them. We show that this improves the model's ability to capture global spatial dependencies. Evaluated on the MP-100 benchmark, which includes 100 categories and over 20K images, EdgeCape achieves state-of-the-art results in the 1-shot setting and leads among similar-sized methods in the 5-shot setting, significantly improving keypoint localization accuracy. Our code is publicly available.

cs.CV↗

arXiv2026

Frequency-Aware Gaussian Splatting Decomposition

3D Gaussian Splatting (3D-GS) enables efficient novel view synthesis, but treats all frequencies uniformly, making it difficult to separate coarse structure from fine detail. Recent works have started to exploit frequency signals, but lack explicit frequency decomposition of the 3D representation itself. We propose a frequency-aware decomposition that organizes 3D Gaussians into groups corresponding to Laplacian-pyramid subbands of the input images. Each group is trained with spatial frequency regularization to confine it to its target frequency, while higher-frequency bands use signed residual colors to capture fine details that may be missed by lower-frequency reconstructions. A progressive coarse-to-fine training schedule stabilizes the decomposition. Our method achieves state-of-the-art reconstruction quality and rendering speed among all LOD-capable methods. In addition to improved interpretability, our method enables dynamic level-of-detail rendering, progressive streaming, foveated rendering, promptable 3D focus, and artistic filtering. Our code will be made publicly available.

cs.CV↗

arXiv2025

DeepShare: Sharing ReLU Across Channels and Layers for Efficient Private Inference

Private Inference (PI) uses cryptographic primitives to perform privacy preserving machine learning. In this setting, the owner of the network runs inference on the data of the client without learning anything about the data and without revealing any information about the model. It has been observed that a major computational bottleneck of PI is the calculation of the gate (i.e., ReLU), so a considerable amount of effort have been devoted to reducing the number of ReLUs in a given network. We focus on the DReLU, which is the non-linear step function of the ReLU and show that one DReLU can serve many ReLU operations. We suggest a new activation module where the DReLU operation is only performed on a subset of the channels (Prototype channels), while the rest of the channels (replicate channels) replicates the DReLU of each of their neurons from the corresponding neurons in one of the prototype channels. We then extend this idea to work across different layers. We show that this formulation can drastically reduce the number of DReLU operations in resnet type network. Furthermore, our theoretical analysis shows that this new formulation can solve an extended version of the XOR problem, using just one non-linearity and two neurons, something that traditional formulations and some PI specific methods cannot achieve. We achieve new SOTA results on several classification setups, and achieve SOTA results on image segmentation.

cs.LG↗

arXiv2025

Coordinate Descent for Network Linearization

ReLU activations are the main bottleneck in Private Inference that is based on ResNet networks. This is because they incur significant inference latency. Reducing ReLU count is a discrete optimization problem, and there are two common ways to approach it. Most current state-of-the-art methods are based on a smooth approximation that jointly optimizes network accuracy and ReLU budget at once. However, the last hard thresholding step of the optimization usually introduces a large performance loss. We take an alternative approach that works directly in the discrete domain by leveraging Coordinate Descent as our optimization framework. In contrast to previous methods, this yields a sparse solution by design. We demonstrate, through extensive experiments, that our method is State of the Art on common benchmarks.

cs.LG↗

arXiv2025

Optimize the Unseen -- Fast NeRF Cleanup with Free Space Prior

Neural Radiance Fields (NeRF) have advanced photorealistic novel view synthesis, but their reliance on photometric reconstruction introduces artifacts, commonly known as "floaters". These artifacts degrade novel view quality, especially in areas unseen by the training cameras. We present a fast, post-hoc NeRF cleanup method that eliminates such artifacts by enforcing our Free Space Prior, effectively minimizing floaters without disrupting the NeRF's representation of observed regions. Unlike existing approaches that rely on either Maximum Likelihood (ML) estimation to fit the data or a complex, local data-driven prior, our method adopts a Maximum-a-Posteriori (MAP) approach, selecting the optimal model parameters under a simple global prior assumption that unseen regions should remain empty. This enables our method to clean artifacts in both seen and unseen areas, enhancing novel view quality even in challenging scene regions. Our method is comparable with existing NeRF cleanup models while being 2.5x faster in inference time, requires no additional memory beyond the original NeRF, and achieves cleanup training in less than 30 seconds. Our code will be made publically available.

cs.CV↗

arXiv2025

Talking Points: Describing and Localizing Pixels

Vision-language models have achieved remarkable success in cross-modal understanding. Yet, these models remain limited to object-level or region-level grounding, lacking the capability for pixel-precise keypoint comprehension through natural language. We introduce a novel framework for pixel level grounding. The framework consists of two complementary components: a Point Descriptor that generates rich, contextual descriptions of individual keypoints, and a Point Localizer that regresses precise pixel coordinates from these descriptions. Unlike prior work that relies on templated prompts or keypoint names, our approach produces free-form, coarse-to-fine descriptions that situate keypoints within their visual context. Since there is no available dataset to train such a system, we introduce LlamaPointInPart, a carefully curated dataset of 20K+ image-keypoint-description triplets synthesized from multiple vision-language models, capturing multi-scale information from scene-level context to visual features around the keypoint. For cross-category generalization, we optimize the Point Descriptor on AP-10K via GRPO, using the frozen Point Localizer as a reward model to produce descriptions that maximize localization accuracy. To evaluate our results we establish a new evaluation protocol. Instead of comparing the text description produced by our method to the ground truth, we use the localizer to determine how close is the predicted point generated to the ground truth point. Experiments demonstrate superior performance compared to baseline models on LlamaPointInPart.The bidirectional nature of our framework should enable future applications in both keypoint-guided image understanding and language-guided precise localization. Our code and dataset are publicly available at https://github.com/matanr/Talking_Points.

cs.CV↗