ECCV 2026 Papers

Multi-modal image matching is essential for visual localiza-tion and multi-sensor fusion, but it is hindered by the scarcity of large-scale training data with precise geometric annotations. Existing real-world datasets suffer from prohibitive costs, limited scene diversity, anderrors in SfM-MVS pipelines, while synthetic methods struggle to main-tain 3D geometric consistency or achieve photorealistic appearance. Toaddress this, we propose AnyMatch, a novel framework that leveragesabundant, easily accessible single-view images at minimal cost to gen-erate rich multi-modal training data. AnyMatch integrates monoculardepth estimation, 3D reprojection, diffusion-based inpainting, and cross-modal image translation to synthesize multi-view, multi-modal imagepairs with 3D geometric fidelity. Crucially, our method provides annota-tions that strictly adhere to 3D geometric consistency through explicit3D reprojection, avoiding SfM-MVS error accumulation. Furthermore,AnyMatch offers strong scalability, enabling controllable scene diversityand annotation difficulty via adjustable input and camera parameters.We construct Any-syn, a large-scale synthetic multi-modal dataset us-ing AnyMatch. Experimental results show that matching networks (e.g.,LoFTR, EDM, RoMa) fine-tuned on Any-syn achieve substantial perfor-mance gains on multi-modal benchmarks, exhibiting superior generaliza-tion and robustness compared to models trained on existing data.

Maintaining spatial world consistency over long horizons remains a central challenge for camera-controllable video generation. Existing memory-based approaches often condition generation on globally reconstructed 3D scenes by rendering anchor videos from the reconstructed geometry in the history. However, reconstructing a global 3D scene from multiple views inevitably introduces cross-view misalignment, as pose and depth estimation errors cause the same surfaces to be reconstructed at slightly different 3D locations across views. When fused, these inconsistencies accumulate into noisy geometry that contaminates the conditioning signals and degrades generation quality. We introduce AnchorWeave, a memory-augmented video generation framework that replaces a single misaligned global memory with multiple clean local geometric memories and learns to reconcile their cross-view inconsistencies. To this end, AnchorWeave performs coverage-driven local memory retrieval aligned with the target trajectory and integrates the selected local memories through a multi-anchor weaving controller during generation. Extensive experiments demonstrate that AnchorWeave significantly improves long-term scene consistency while maintaining strong visual quality, with ablation and analysis studies further validating the effectiveness of local geometric conditioning, multi-anchor control, and coveragedriven retrieval. We provide more qualitative demonstration videos in the supplementary material via an anonymized local webpage.

Modern stereo-capable smartphones enable immersive XRcontent capture. However, hardware heterogeneity across camera mod-ules often causes severe asymmetric blur artifacts. Existing methods andbenchmarks largely assume homogeneous stereo setups and therefore donot explicitly address such asymmetric degradation. To bridge this gap,we present a dedicated framework for heterogeneous stereo deblurring.First, we introduce the heterogeneous stereo deblurring (HSD) dataset,constructed from real smartphone stereo captures via multi-frame inte-gration. Second, we propose physically- and epipolar-constrained crossattention (PECA), a lightweight module that restricts cross-view match-ing to an epipolar search window bounded by an optics-derived dis-parity upper bound. By enforcing physically valid disparity constraints,PECA enables efficient and reliable cross-view feature fusion. Moreover,our confidence-weighted attention with residual fusion emphasizes cross-guided deblurring when correspondences are reliable, while naturallyfalling back to self-deblurring in occluded or unreliable regions. PECAis architecture-agnostic and consistently improves CNN-, Transformer-,and NAFNet-based baselines. Extensive experiments on HSD show thatPECA-enhanced models achieve improved restoration performance withfavorable efficiency. The dataset and source code are publicly released athttps://github.com/shinhoju/PECA.

Object-Goal Navigation (ObjectNav) is a key capability fordeploying mobile robots in everyday environments such as homes, schools,and workplaces. In this task, an agent must locate an instance of a targetobject category in previously unseen environments using only onboardperception, requiring the integration of semantic understanding, spatialreasoning, and long-horizon planning. Reinforcement learning (RL) hasbecome a dominant paradigm for ObjectNav, yet modern systems involvenumerous design choices across perception modules, policy architectures,and inference-time strategies. The relative impact of these components,however, remains poorly understood. In this work, we present a large-scale empirical study of modular RL-based ObjectNav systems. We de-compose the navigation pipeline into three key components: perception,policy, and test-time enhancement, and conduct extensive controlled ex-periments to analyze their individual contributions. Our results suggestthat improvements in perception quality and test-time strategies oftenyield larger performance gains than policy improvements alone, high-lighting the importance of understanding how different components inter-act within modular navigation systems. Motivated by these findings, weintroduce a unified framework for systematically studying modular Ob-jectNav systems. Guided by our analysis, we build an enhanced systemthat achieves state-of-the-art performance on the Gibson benchmark, im-proving SPL by 6.6% and success rate by 2.7% over prior methods. Wealso introduce a human expert baseline, achieving 98% success, high-lighting the significant gap between current RL agents and human-levelnavigation. Finally, we provide practical insights and design recommen-dations for each module to help guide future research. Project page:https://honwang0054.github.io/What-matters-in-RL-ObjNav-web/.

Vision Language Models (VLMs) face signi_x001C_cant challengeswith ultra-long, interleaved image-text sequences due to the quadraticcomplexity of self-attention. Current solutions either resort to aggres-sive token pruning, risking irreversible information loss, or adopt e_x001E_-cient but less precise architectures, while largely ignoring the equallyvital textual component. We introduce VLZip, a framework that uni-_x001C_es visual and textual compression for high-_x001C_delity reasoning within apure Transformer. At its core, VLZip hierarchically distills visual andtextual segments into compact, layer-speci_x001C_c "soft pre_x001C_xes" and injectsthem into each decoder layer's hidden states, drastically shortening theattention sequence while preserving _x001C_ne-grained global context. To ad-dress de_x001C_cient evaluations in the _x001C_eld, we also introduce LongVLBench,a new benchmark derived from video narratives that demands holistic,narrative-level reasoning. Extensive experiments show VLZip achievesleading performance on long-context multimodal reasoning, enabling train-ing up to 120K tokens_x0016_a 6× increase over the baseline_x0016_and inferencebeyond 280K tokens with signi_x001C_cantly reduced memory, while demon-strating the memory scalability to handle up to 2M tokens. By excellingat extreme context lengths where existing methods collapse, VLZip es-tablishes an e_x001E_cient and powerful new standard for long-context mul-timodal AI. Code is available at https://github.com/ShareLab-SII/VLZip.

Large Vision Language Models (VLMs) excel at general vi-sual reasoning tasks, but their performance degrades sharply when de-ployed in novel domains with substantial distribution shifts compared towhat was seen during pretraining. Existing approaches to adapt VLMsto novel target domains rely on finetuning standard VLM components.Depending on which components are finetuned, these approaches eitherlimit the VLMs ability to learn domain-specific features, or lead to catas-trophic forgetting of pre-existing capabilities. To address this, we intro-duce Vision Contextualized Probing (VisCoP), which augments theVLM’s vision encoder with a compact set of learnable visual probes,enabling domain-specific features to be learned with only minimal up-dates to the pretrained VLM components. We evaluate VisCoP acrossthree challenging domain adaptation scenarios: cross-view (exocentric →egocentric), cross-modal (RGB → depth), and cross-task (human under-standing → robot control). Our experiments demonstrate that VisCoPconsistently outperforms existing domain adaptation strategies, achiev-ing superior performance on the target domain, while better retainingcapabilities from the source domain. We will release all code, models,and evaluation protocols to facilitate future research in this direction.

Relative position embedding has become a standard mech-anism for encoding positional information in Transformers. However,existing formulations are typically limited to a fixed geometric space,namely 1D sequences or regular 2D/3D grids, which restricts their ap-plicability to many computer vision tasks that require geometric rea-soning across camera views or between 2D and 3D spaces. To addressthis limitation, we propose URoPE, a universal extension of RotaryPosition Embedding (RoPE) to cross-view or cross-dimensional geomet-ric spaces. For each key/value image patch, URoPE samples 3D pointsalong the corresponding camera ray at predefined depth anchors andprojects them into the query image plane. Standard 2D RoPE can thenbe applied using the projected pixel coordinates. URoPE is a parameter-free and intrinsics-aware relative position embedding that is invariant tothe choice of global coordinate systems, while remaining fully compatiblewith existing RoPE-optimized attention kernels. We evaluate URoPE asa plug-in positional encoding for transformer architectures across a di-verse set of tasks, including novel view synthesis, 3D object detection,object tracking, and depth estimation, covering 2D–2D, 2D–3D, and tem-poral scenarios. Experiments show that URoPE consistently improvesthe performance of transformer-based models across all tasks, demon-strating its effectiveness and generality for geometric reasoning. Our codeis available on our project website: https://urope-pe.github.io/.

In this paper, we propose GTA-VLA (Guide, Think,Act), an interactive Vision-Language-Action (VLA) framework thatenables spatially steerable embodied reasoning by allowing users to guiderobot policies with explicit visual cues. Existing VLA models learn adirect "Sense-to-Act" mapping from multimodal observations to robotactions. While effective within the training distribution, such tightly cou-pled policies are brittle under out-of-domain (OOD) shifts and difficultto correct when failures occur. Although recent embodied Chain-of-Thought (CoT) approaches expose intermediate reasoning, they stilllack a mechanism for incorporating human spatial guidance, limitingtheir ability to resolve visual ambiguities or recover from mistakes. Toaddress this gap, our framework allows users to optionally guide thepolicy with spatial priors, such as affordance points, boxes, and traces,which the subsequent reasoning process can directly condition on. Basedon these inputs, the model generates a unified spatial-visual Chain-of-Thought that integrates external guidance with internal task planning,aligning human visual intent with autonomous decision-making. Forpractical deployment, we further couple the reasoning module with alightweight reactive action head for efficient action execution. Extensiveexperiments demonstrate the effectiveness of our approach. On thein-domain SimplerEnv WidowX benchmark, our framework achievesa state-of-the-art 81.2% success rate. Under OOD visual shifts andspatial ambiguities, a single visual interaction substantially improvestask success over existing methods, highlighting the value of interactivereasoning for failure recovery in embodied control.

Vision and vision–language models rely on high-level visualrepresentations that are increasingly used across recognition, retrieval,and multimodal reasoning pipelines. However, recent advances in gen-erative modeling have shown that such features can often be inverted,enabling realistic reconstructions of the underlying image and raisingsignificant privacy risks. We revisit this problem through the lens ofreconstruction and propose TrustCLIP, a reconstruction-driven frame-work that treats a feature-conditioned generator as an explicit privacyadversary. TrustCLIP learns a projection between encoder features anddownstream modules that is explicitly optimized to degrade the recon-structions produced by generative attackers while retaining the necessarysignals for downstream tasks. Unlike prior defenses that rely on discrimi-native privacy metrics, TrustCLIP directly optimizes against a generativereconstruction attacker, targeting a threat not captured by standard eval-uation protocols. We demonstrate its effectiveness in both conventionalclassification and multimodal large language model pipelines. Acrossthese settings, TrustCLIP consistently reduces the fidelity of genera-tive inversions while maintaining downstream task performance. Projectpage: atnikos.github.io/trustclip.

Estimating the 3D trajectory of every pixel from a monocularvideo is crucial and promising for a comprehensive understanding of the3D dynamics of videos. Recent monocular 3D tracking works demon-strate impressive performance, but are limited to either tracking sparsepoints on the first frame or a slow optimization-based framework fordense tracking. In this paper, we propose a feedforward model, calledTrack4World, enabling an efficient holistic 3D tracking of every pixel inthe world-centric coordinate system. Built on the global 3D scene repre-sentation encoded by a VGGT-style ViT, Track4World applies a novel3D correlation scheme to simultaneously estimate the pixel-wise 2D and3D dense flow between arbitrary frame pairs. The estimated scene flow,along with the reconstructed 3D geometry, enables subsequent efficient3D tracking of every pixel of this video. Extensive experiments on multiplebenchmarks demonstrate that our approach consistently outperforms ex-isting methods in 2D/3D flow estimation and 3D tracking, highlighting itsrobustness and scalability for real-world 4D reconstruction tasks. Projectpage: https://jiah-cloud.github.io/Track4World.github.io/.

Tone style transfer for photo retouching aims to adapt thestylistic tone of the reference image to a given content image. However,the lack of high-quality large-scale triplet datasets with stylized groundtruth forces existing methods to rely on self-supervised or proxy objec-tives, which limits model capability. To mitigate this gap, we design adata construction pipeline to build TST100K, a large-scale dataset of100,000 content-reference-stylized triplets. At the core of this pipeline,we train a tone style scorer to ensure strict stylistic consistency for eachtriplet. In addition, existing methods typically extract content and ref-erence features independently and then fuse them in a decoder, whichmay cause semantic loss and lead to inappropriate color transfer and de-graded visual aesthetics. Instead, we propose ICTone, a diffusion-basedframework that performs tone transfer in an in-context manner by jointlyconditioning on both images, leveraging the semantic priors of generativemodels for semantic-aware transfer. Reward feedback learning using thetone style scorer is further incorporated to improve stylistic fidelity andvisual quality. Experiments demonstrate the effectiveness of TST100K,and ICTone achieves state-of-the-art performance on both quantitativemetrics and human evaluations. The project page is available online.

Effective Human-Robot Collaboration (HRC) requires robotsto accurately infer human intents under partial observations. However,existing benchmarks mostly rely on idealized views, overlooking the spa-tiotemporal incompleteness such as low-angle truncation and dynamicocclusions inherent in real-world quadruped robot perception. Further-more, current evaluations lack diagnostic transparency, failing to distin-guish whether Vision-Language Models (VLMs) perform reliable com-pensatory reasoning or merely hallucinate based on statistical biaseswhen visual features are degraded. To bridge these dual gaps, we in-troduce CoT-HRC, a large-scale benchmark built in Habitat 3.0 thatcouples simulated robot-centric views with explicit hierarchical Chain-of-Thought (CoT) annotations. We propose a diagnostic protocol featur-ing the Step-wise Consistency Assessment (SCAR ) to penalize spuriousaccuracy by enforcing intermediate logical consistency, and the Condi-tional Reasoning Accuracy (CRA) to explicitly decouple models’ un-derlying reasoning capabilities from visual perception failures. Exten-sive experiments across state-of-the-art VLMs reveal that while currentmodels inherently possess strong physical commonsense, their reason-ing chains are severely disrupted by the dynamic geometric constraintsof egocentric views. Our fine-grained ablations further highlight criti-cal bottlenecks in temporal aggregation and the compensatory role ofsemantic priors, establishing CoT-HRC as a vital stepping stone for ro-bust embodied intent inference.The full benchmark is publicly availableat https://github.com/Thus-cx/CoT-HRC.

The visual aesthetics of photographs are deeply influencedby lens characteristics such as aperture shape, optical vignetting and op-tical diffraction, which together define a camera’s unique optical style.Existing lens effect rendering methods primarily focus on accurately sim-ulating the blur transition from small to large apertures but overlook thestylistic aspects of lens effects. As a result, they fail to produce diversebokeh effects under large apertures or capture distinctive photographicphenomena such as starbursts that emerge under small apertures. In thiswork, we introduce LensStyle, a unified framework for controllable styl-ized lens effect rendering that explicitly models lens aesthetics throughjoint continuous–discrete control. Our model incorporates a Dual-PathController that disentangles continuous optical parameter modulation(e.g., focus distance and blur strength) from discrete lens-style condi-tioning (e.g., circular, polygonal, donut, cat-eye, and starburst effects),enabling fine-grained, interpretable, and physically grounded lens manip-ulation within a single unified framework. To support model training, wecurate a comprehensive MultiLens dataset containing multi-lens imagepairs synthesized under real optical constraints. Extensive experimentsdemonstrate that LensStyle achieves superior realism, controllability, andaesthetic quality compared with existing lens effect rendering approachesand diffusion-based image editing models, advancing computational pho-tography toward multiple-lens-style simulation.

Existing video temporal grounding methods typically oper-ate under a closed-world assumption, relying on the implicit premise thatevery natural language query must have a corresponding event some-where in the video. This always-answer paradigm severely compressesthe decision boundary during training, as models are seldom exposed tothe alternative hypothesis that the queried event may be absent. Con-sequently, when users ask about specific events that cannot be matchedin the video within an open-world environment, these models exhibitsystematic temporal hallucinations. This may stem from incorrect at-tribution of textual, visual, or multimodal information, leading them toconfidently generate plausible time frames for events that do not actuallyexist. To address this limitation, we propose HRVTG, a test-time adapta-tion framework that dynamically calibrates the decision boundary of themodel during inference. Built upon a frozen grounding backbone, the pro-posed framework constructs three types of counterfactual hallucinationprobes in a self-supervised manner, targeting the aforementioned halluci-nations. We dynamically reshape the decision boundary through absten-tion signals from counterfactual probes and consistency rewards fromgenuine queries. This is executed by introducing the GRPO algorithm,effectively reformulating the task into an online policy optimization prob-lem in reinforcement learning. Furthermore, we introduce decoupled met-rics to independently evaluate hallucination resistance and real-eventgrounding. Experiments indicate that our method suppresses halluci-nated responses to fabricated queries. It achieves accurate selective pre-diction while maintaining strong grounding performance on real events.The code will be released at https://github.com/CVL-hub/HRVTG.

VAR has gained widespread popularity due to its next-scaleprediction paradigm. However, it faces substantial performance bottle-necks when handling complex scenes with multiple objects and attributes.Existing diffusion-based enhancement methods fail to adequately addressthe unique challenge of cross-scale error propagation and accumulationin VAR. To this end, we propose SynVAR, the first training-free enhance-ment framework specifically tailored for the VAR paradigm, which intro-duces a spatial-semantic collaborative control strategy to effectively sup-press propagation error and improve generation quality. SynVAR com-prises three key components: (1) Global guidance to ensure reasonablespatial structure in the early stages, (2) Receptive field constraints tomitigate early-stage semantic confusion, (3) High-frequency compensa-tion to recover fine-grained details. Extensive quantitative and qualita-tive experiments demonstrate the significant improvements in the abilityof SynVAR to enhance the VAR’s capability for complex scene modeling.

While synthetic data has proven effective for improving sci-entific reasoning in the text domain, multimodal reasoning remains con-strained by the difficulty of synthesizing scientifically rigorous images.Existing Text-to-Image (T2I) models often produce outputs that arevisually plausible yet scientifically incorrect, resulting in a persistentvisual–logic divergence that limits their value for downstream reason-ing. Motivated by recent advances in next-generation T2I models, weconduct a systematic study of scientific image synthesis across genera-tion paradigms, evaluation, and downstream use. We examine both di-rect pixel-based generation and programmatic synthesis, and instantiateImgCoder as a structured implementation of the code-driven workflow,following an explicit “understand → plan → code” prompting strategyto encourage clearer structural specification. To rigorously assess scien-tific correctness, we introduce SciGenBench, which evaluates generatedimages based on information utility and logical validity. Our evaluationreveals systematic failure modes in pixel-based models and highlightsa fundamental expressiveness–precision trade-off. Finally, we show thatfine-tuning Large Multimodal Models (LMMs) on rigorously verified syn-thetic scientific images yields consistent reasoning gains, with potentialscaling trends analogous to the text domain, validating high-fidelity sci-entific synthesis as a viable path to unlocking massive multimodal rea-soning capabilities.

Visual generation is increasingly ubiquitous in diverse do-mains, from text-to-image/video synthesis to multimodal interactive cre-ation. Yet prevailing monolithic models remain fundamentally constrainedby their inability to learn cumulatively and evolve autonomously, whichis a limitation we term the “perpetual novice” problem. They lack mech-anisms for structuring experience into reusable knowledge and there-fore rely on brittle, “from-scratch” reasoning for each task, resultingin poor compositional generalization and inefficient knowledge reten-tion. Motivated by these limitations, we propose SymbOmni, an agenticomni-model designed for cumulative evolution through Symbolic Con-cept Learning. At its core is the Symbolic Concept Box, an optimizablememory module that abstracts low-level operations into reusable Sym-bolic Workflow Instructions. SymbOmni operates through an induction-transduction cycle: experiences are abstracted into symbolic concepts(induction), which are then adaptively composed to solve novel tasks(transduction). The training is done by verbalized backpropagation withlanguage-based feedback to enable continuous self-improvement withoutgradient-based model fine-tuning. Comprehensive experiments validatethat (I) SymbOmni significantly outperforms existing agent-based sys-tems for iterative creation and also surpasses closed-source models (e.g.,Nano Banana, GPT-Image-1) in both image quality and task successrates; (II) SymbOmni effectively reduces token consumption by over 40%while maintaining competitive generation quality; and (III) SymbOmnienables effective continual learning by achieving cumulative gains acrossmultiple online-learning benchmarks and setting a new state of the art.

We present SharpGS, a differentiable blur-driven density con-trol, which enhances the quality of 3D Gaussian Splatting (3DGS). Stan-dard 3DGS often struggles to capture intricate details, particularly intextured patterns and object boundaries, due to the limited sensitivityof its density control. Simple finer densification can result in excessiveprimitive counts with marginal quality gain. To address this, we intro-duce blur as an effective perceptual cue, leveraging CUDA-based differ-entiable blur. Our density control estimates per-primitive blur levels bycomparing reconstructions to blurred ground-truth images, and identifieshigh-frequency regions where additional primitives are required. Whilethis naturally produces more primitives, we counterbalance this by sup-pressing redundant primitives. Specifically, we penalize the blur levelsof inherently smooth regions such as sky, and regularize the opacitiesof potential split/clone candidates. We experimentally demonstrate thatSharpGS greatly improves the state-of-the-art 3DGS methods in termsof quality, while keeping learned primitives compact.

Establishing accurate point-to-point correspondences betweennon-rigid 3D shapes remains a critical challenge, particularly under non-isometric deformations and topological noise. Existing functional mappipelines suffer from ambiguities that geometric descriptors alone can-not resolve, and spatial inconsistencies inherent in the projection of trun-cated spectral bases to dense pointwise correspondences. In this paper,we introduce SGMatch, a learning-based framework that couples 3D-lifted semantic cues with trajectory-level feature transport regulariza-tion. Specifically, we design a Semantic-Guided Local Cross-Attentionmodule that integrates semantic features from vision foundation modelsinto geometric descriptors while preserving local structural continuity.Furthermore, we adapt conditional flow matching as a time-conditionedfeature transport regularizer that promotes spatially coherent point-wiserecovery. Experimental results on multiple benchmarks demonstrate thatSGMatch achieves competitive performance across near-isometric set-tings and consistent improvements under non-isometric deformations andtopological noise.