ECCV 2026 Papers

We present SLAM-Former, a neural approach that integratesfull SLAM capabilities into a single transformer. Similar to traditionalSLAM systems, SLAM-Former comprises both a frontend and a back-end that operate in tandem. The frontend processes sequential monoc-ular images in real-time for incremental mapping and tracking, whilethe backend performs global refinement to ensure a geometrically con-sistent result. This alternating execution allows the frontend and back-end to mutually promote one another, enhancing overall system perfor-mance. Comprehensive experimental results demonstrate that SLAM-Former achieves superior or highly competitive performance comparedto state-of-the-art dense SLAM methods.

Implicit Neural Representations (INRs) are widely used formodeling continuous 2D images, enabling high-fidelity reconstruction,super-resolution, and compression. Architectures such as SIREN, WIRE,and FINER demonstrate their ability to capture fine image details. How-ever, conventional INRs lack explicit geometric structure, limiting localediting, and integration with physical simulation. To address these lim-itations, we propose GaINeR (Geometry-Aware Implicit Network Rep-resentation), a novel framework for 2D images that combines trainableGaussian distributions with a neural network-based INR. For a givenimage coordinate, the model retrieves the K nearest Gaussians, aggre-gates distance-weighted embeddings, and predicts the RGB value viaa neural network. This design enables continuous image representation,interpretable geometric structure, and flexible local editing, providing afoundation for physically aware and interactive image manipulation. Ourmethod supports geometry-consistent transformations, seamless super-resolution, and integration with physics-based simulations. Moreover, theGaussian representation allows lifting a single 2D image into a geometry-aware 3D representation, enabling depth-guided editing. Experimentsdemonstrate that GaINeR achieves state-of-the-art reconstruction qual-ity while maintaining flexible and physically consistent image editing.

We present FoundationGeo, a two-stage framework that ex-plicitly bridges relative and metric prediction via spatial calibration andprincipled data design. Stage 1 learns a high-fidelity, affine-invariant ge-ometry model by initializing with DINOv3 and training on a curated10.2M-sample multi-domain corpus with complementary local–detail su-pervision, yielding sharp boundaries and strong cross-domain general-ization. Stage 2 moves beyond global scaling by introducing lightweightpixel-wise calibration fields for metric estimation: a scale field for spa-tially varying metric alignment and a ray-direction correction field thatmitigates directional bias in point-map geometry, together producingmetrically consistent 3D point maps. Beyond model design, we identifycamera intrinsic coverage, especially focal length distribution mismatchbetween training and test data, as a key bottleneck for zero-shot met-ric generalization: performance drops sharply when test intrinsics falloutside the training distribution. To address this, we synthesize addi-tional training data across diverse focal lengths using a Blender-baseddata engine, repairing under-covered focal regimes and improving robust-ness under intrinsic shift. Extensive zero-shot evaluations across sevenbenchmarks show that FoundationGeo significantly strengthens cross-domain robustness, staying near the top across diverse domains whileavoiding the sharp cross-domain performance drops observed in othermethods. This consistency translates into the best overall performance,surpassing heavier baselines by over 5.2% on average. Project page:https://mx-liu6.github.io/FoundationGeo-web/

Vision-language models (VLMs) commonly formulate visualgrounding and detection as a coordinate-token generation problem, serial-izing each 2D box into multiple 1D tokens that are learned and decodedlargely independently. This token-by-token decoding mismatches the cou-pled structure of box geometry and creates a practical inference bottleneckdue to strictly sequential generation. We introduce LocateAnything,a unified generative grounding and detection framework based on Par-allel Box Decoding (PBD). By decoding geometric elements such asbounding boxes and points as atomic units in a single step, LocateAny-thing preserves intra-box geometric coherence and unlocks substantialparallelism. We show that PBD improves both decoding throughput andlocalization accuracy. We further develop a scalable data engine andcurate LocateAnything-Data, a large-scale dataset with more than138 million training samples, substantially increasing data diversity forhigh-precision localization. Extensive evaluations show that LocateAny-thing advances the speed–accuracy frontier, achieving significantly higherdecoding throughput while improving high-IoU localization quality acrossdiverse benchmarks. The results highlight the complementary benefits ofParallel Box Decoding and large-scale training data in enabling efficientand precise unified visual grounding and detection.

Vision–Language–Action (VLA) models have become a com-pelling route toward robotic foundation models by mapping multimodalperception and language instructions directly into executable actions. Be-yond the prevalent flow-matching VLAs, discrete diffusion VLA (dVLA)models further unify perception and control in a single discrete space,offering inherent advantages in unified multimodal alignment and under-standing, while better preserving the prior knowledge of vision-languagemodels. Yet this strength comes with a practical obstacle that inferenceis far too slow for real-time control, as the parallel decoding prevents theutilization of Key-Value (KV) cache. In this paper, we aim to tackle thischallenge by leveraging an intriguing observation that the dVLA withbidirectional attention still adheres to a block-wise left-to-right order,which motivates the application of block diffusion. However, directlyapplying block diffusion precludes inter-block parallelism, a key accel-eration factor. To address the limitation, we introduce Fast-dVLA, anblock-wise diffusion acceleration strategy. Fast-dVLA takes the full actiontoken sequence at each timestep as an action block and denoises themtogether. It then decodes different blocks in an autoregressive mannerto allow KV cache reusing, while allowing inter-block parallel decodingin a diffusion-forcing manner. For efficient training, we directly conductan asymmetric distillation on a finetuned dVLA. Extensive evaluationson Discrete Diffusion VLA, Dream-VLA, and UD-VLA across diversesimulated benchmarks validate that our Fast-dVLA achieves 2.8×-4.1×speedup and maintains state-of-the-art performance. Moreover, the resultson real-world high-dynamics tasks demonstrate the potential of real-timedeployment and application of our Fast-dVLA.

Online video understanding requires models to perform continuous perception and long-range reasoning within potentially infinite visual streams. Its fundamental challenge lies in the conflict between the unbounded nature of streaming media input and the limited context window of Multimodal Large Language Models (MLLMs). Current methods primarily rely on passive processing, which often face a trade-off between maintaining long-range context and capturing the fine-grained details necessary for complex tasks. To address this, we introduce EventMemAgent, an active online video agent framework based on a hierarchical memory module. Our framework employs a dual-layer strategy for online videos: short-term memory detects event boundaries and utilizes event-granular reservoir sampling to process streaming video frames within a fixed-length buffer dynamically; long-term memory structuredly archives past observations on an event-by-event basis. Furthermore, we integrate a multi-granular perception toolkit for active, iterative evidence capture and employ Agentic Reinforcement Learning (Agentic RL) to end-to-end internalize reasoning and tool-use strategies into the agent’s intrinsic capabilities. Experiments show that EventMemAgent achieves competitive results on online video benchmarks. The code will be released here: https://github.com/lingcco/EventMemAgent.

We introduce EgoPolice, a carefully curated dataset of real,egocentric police–civilian interactions, sourced from publicly availablebody-worn camera videos. We select police-civilian action labels that arecritical for police behavioral research and annotate them at a second-by-second granularity. The videos feature rapid and irregular cameramotion, dense human interactions, and rare high-stakes events, mak-ing the dataset a challenging benchmark for motion-robust and context-aware egocentric perception. We provide two different tasks, classifica-tion and multiple-choice question-answering, and benchmark both open-source and closed-source models. We find that even the best video modelslike Gemini 2.5 Pro still struggle to accurately predict high-risk actionssuch as “Weapon Out”. Beyond serving as a benchmark, EgoPolice pro-vides a foundation for developing models capable of identifying eventsof interest in large-scale body-worn camera video repositories, enablingmore efficient downstream human review.Content Warning: This paper includes real police body-worn camerafootage, including potentially distressing scenes.

Do Video-LLMs have consistent temporal understandingwhen videos capture the same event from different viewpoints? To studythis question, we introduce EgoExo-Con(sistency), a benchmark of syn-chronized egocentric and exocentric video pairs with human-refinedqueries that ensure all concepts are visible in both viewpoints. EgoExo-Con emphasizes two temporal understanding tasks: Temporal Verificationand Temporal Grounding. It evaluates not only correctness but consis-tency across viewpoints. Our analysis reveals two critical limitations ofexisting Video-LLMs: (1) models often fail to maintain consistency, withresults far worse than their single-view performances. (2) When naivelyfinetuned with synchronized videos of both viewpoints, the models showimproved consistency but often underperform those trained on a singleview. For improvements, we propose View-GRPO, a novel reinforcementlearning framework that effectively strengthens view-specific temporalreasoning while encouraging consistent comprehension across viewpoints.Our method demonstrates its superior temporal understanding capabili-ties, especially for improving cross-view consistency. All resources havebeen made available at EgoExo-Con.

Adapting 2D foundation models for 3D spatial understanding faces a critical dilemma. Fine-tuning 2D models to learn 3D geometry causes the catastrophic forgetting of native 2D knowledge. Conversely, freezing the 2D model restricts 3D spatial perception. To resolve this issue, we propose DINOv3D, a joint optimization framework built upon DINOv3 that employs a homologous teacher-student architecture to establish a regularized integration between the 2D and 3D understanding. To preserve original 2D priors, DINOv3D distills 2D knowledge into a 3D Gaussian regularization field, which aligns the rendered features with the reference visual features from a frozen teacher model. Meanwhile, a student model processes long-context multi-view inputs through parameter-efficient fine-tuning. This step injects 3D spatial consistency priors to 2D model while using the regularization field to mitigate the forgetting of 2D knowledge. To enrich the hierarchical representation of 3D spatial understanding, the student model predicts additional semantic and instance Gaussian features. We then apply a ray-depth-semantic alignment mechanism, which uses 3D depth-ray priors to enforce multiview consistency across 2D semantic rendering. Extensive experiments demonstrate that, despite updating only 10% of the parameters of a 1B-parameter foundation model VGGT, DINOv3D achieves state-of-theart results in comprehensive 3D scene understanding across challenging novel view synthesis, depth estimation, and open-vocabulary semantic segmentation. Furthermore, DINOv3D consistently enhances the generalizability of the DINOv3 backbone on 2D linear probing benchmarks.

As Vision-Language Models (VLMs) become increasingly so-phisticated and widely used, it becomes more and more crucial to un-derstand their decision-making process. Traditional explainability meth-ods, designed for classification tasks, struggle with modern autoregressiveVLMs due to their complex token-by-token generation process and in-tricate interactions between visual and textual modalities. We presentDEX-AR (Dynamic Explainability for AutoRegressive models), a novelexplainability method designed to address these challenges by generat-ing both per-token and sequence-level 2D heatmaps highlighting imageregions crucial for the model’s textual responses. The proposed methodoffers to interpret autoregressive VLMs, accounting for the varying im-portance of layers and generated tokens—by computing layer-wise gradi-ents with respect to attention maps during the token-by-token generationprocess. DEX-AR introduces two key innovations: a dynamic head filter-ing mechanism that identifies attention heads focused on visual informa-tion, and a sequence-level filtering approach that aggregates per-tokenexplanations while distinguishing between visually-grounded and purelylinguistic tokens. Our evaluation on ImageNet, VQAv2, and PascalVOC,shows a consistent improvement in both perturbation-based metrics, us-ing a novel normalized perplexity measure, as well as segmentation-basedmetrics. Code is available at https://walidbousselham.com/DEX-AR

Fine-tuning Multimodal Large Language Models (MLLMs)on specialized tasks often leads to catastrophic forgetting of their gen-eral capabilities. Existing model merging methods to combat this areoften heuristic or use sub-optimal objectives. We propose Curvature-Guided Mixing (CGM), a theoretically grounded framework that mergespre-trained and fine-tuned models. CGM formulates a joint optimiza-tion objective and uses a second-order (Hessian) approximation of theloss landscapes to analytically derive an optimal, closed-form “soft mix-ing” ratio. This ratio intelligently blends parameters based on their rel-ative task-specific curvatures. We also introduce CGM† , a robust “hardmixing” variant that performs sparse parameter selection guided by anovel, curvature-aware score. Experiments on LLaVA-1.5 and Qwen-2.5VL across multiple downstream tasks show that CGM and CGM†consistently improve the trade-off between task specialization and gen-eral knowledge retention over existing methods. Code is available atgithub.com/zzsyjl/CGM-ECCV-2026.

Cross-scene hyperspectral image (HSI) classification facespersistent challenges due to domain shifts in spectral signatures, spatialstructures, and semantic compositions. Existing domain generalizationmethods typically align either features or predictions, but they overlookhow local context (e.g., boundary mixing and class co-occurrence pat-terns) changes across scenes. As a result, one-sided alignment often yieldspartial adaptation: feature-invariant models may still produce unstableoutputs, while output-aligned models can rely on domain-specific short-cuts. We propose Context-Aware Joint Alignment (CAJA), a pixel-faithful framework that preserves the single-label objective while inject-ing explicit context into training. CAJA combines Context-Aware Su-pervision (CAS) to improve boundary-aware supervision using neigh-borhood context and homogeneity, and Context-Conditioned JointAlignment (CCJA) to enforce context-conditioned alignment in bothfeature and prediction spaces across domains. By coupling supervisionand joint alignment under shared context, CAJA mitigates the partial-alignment limitation of one-sided methods. Extensive experiments onthree cross-scene HSI benchmarks show that CAJA consistently improvesgeneralization, especially at class boundaries where context shift is mostsevere, establishing a simple, principled, and deployment-friendly solu-tion for cross-scene HSI domain generalization.

Appearance-based gaze estimation typically relies on facenormalization to reduce appearance variability, but this requires costlyand error-prone landmark detection and head pose estimation. Whilecrop-based alternatives have been explored, their geometric propertiesand performance trade-offs relative to normalization remain underex-plored. In this work, we provide the first systematic comparison of nor-malization versus crop-based gaze estimation. To enable fair compari-son, we formalize the crop-based approach through Constrained Rota-tion Optimization (CROp), making its geometric transformation explicitand comparable to normalization. We further adopt multi-task learningto recover head pose information lost in cropping. Through extensiveexperiments across various datasets, head pose distributions, and pre-processing conditions, we identify the conditions under which each ap-proach excels. CROp shows advantages under extreme poses and noisydetection, while normalization benefits from landmark-based refinementin moderate conditions. Our analysis provides practical guidelines forchoosing preprocessing strategies in real-world gaze estimation systems.

Streaming feed-forward 3D reconstruction enables real-timejoint estimation of scene geometry and camera poses from RGB images.However, without explicit dynamic reasoning, streaming models can beaffected by moving objects, causing artifacts and drift. In this work, wepropose RayMap3R, a training-free streaming framework for dynamicscene reconstruction. We observe that RayMap-based predictions exhibita static-scene bias, providing an internal cue for dynamic identification.Based on this observation, we construct a dual-branch inference schemethat identifies dynamic regions by contrasting RayMap and image pre-dictions, suppressing their interference during memory updates. We fur-ther introduce reset metric alignment and state-aware smoothing to pre-serve metric consistency and stabilize predicted trajectories. Our methodachieves state-of-the-art performance among streaming approaches ondynamic scene reconstruction across multiple benchmarks. The projectpage and code are available at https://raymap3r.github.io/.

Current instruction-based image retrieval systems are powerful but limited to single-turn interactions, failing to capture the iterative nature of complex, real-world visual searches. To overcome this limitation, we introduce Contextual Composed Image Retrieval (CoCo-IR), a novel task that enables users to progressively refine search results through interactions. We address this new task by proposing a new model based on a Large Multimodal Model (LMM) that functions as a context-aware reasoner for CoCo-IR. Our model interprets the entire interaction history to generate Transformable Image Embeddings (TIE) that evolve across turns. To fuel the model training without expensive human annotations, we develop a fully autonomous, scalable data engine that leverages LMMs to generate high-quality contextual retrieval data, and uses model-guided verification to mine challenging hard negatives. Extensive experiments demonstrate that our approach establishes new state-of-the-art performance: We achieve 39.4 mAP@5 on the challenging single-turn benchmark CIRCO; furthermore, on our new CoCo-IR benchmark, our model maintains robust performance with 44.1 R@1 on 4-turn dialogues, dramatically outperforming existing methods (28.2 4-turn R@1) that fail to handle multi-turn context. Project page: https://CoCo-IR.github.io.

We present BWAFDA, a dual-branch no-reference image quality assessment model that replaces uniform averaging with detail-aware, block-wise fusion. The context encoder branch produces multi-scale block tokens, and a lightweight detail branch extracts high-frequency cues. These streams are combined late. First, block tokens are aggregated via block-wise attention across scales, guided by a soft multi-scale prior that stabilizes cross-scale interactions and preserves localized artifacts. The resulting descriptors are then fused per scale using learned weights. Subsequently, the detail cues are injected to refine distortion-sensitive evidence. Finally, a class-token-guided head performs “where-to-trust-what” selection over blocks to produce the quality score, without the need for region labels. Comprehensive evaluations on seven public NR-IQA benchmarks show state-of-the-art (SOTA) performance. Across multiple controlled ablations, we observe consistent drops in correlation compared to the full model, indicating that region-aware masking, contextual features, and multi-scale aggregation contribute jointly rather than redundantly. Overall, detail-aware, block-wise fusion delivers SOTA accuracy while preserving data efficiency and transferability by aligning localized artifact cues with global context.

In human visual perception, uppercase lettering serves as anatural salience cue that captures attention within lowercase text. Inthis paper, we present a systematic empirical characterization study re-vealing that Large Language Models (LLMs) exhibit an analogous prop-erty: letter casing modulates internal attention allocation. Through anal-ysis across 13 models—nine LLMs and four Vision-Language Models(VLMs)—with diverse tokenization schemes, we show that formattingtarget information in alternating or uppercase against a lowercase con-text concentrates attention on those textual spans. In text this e!ectis universal, holding across every evaluated non-reasoning model. Weframe it as a previously under-explored latent property of pretrainedtransformers rather than a prescriptive method. Our investigation re-veals a central attention–performance divergence: while this “casing ef-fect” robustly shifts attention, its impact on downstream accuracy isnon-trivial—increased concentration does not inherently improve taskaccuracy and, in high-entropy contexts like alternating case, can degradeit. We further identify a boundary condition: the deliberative “thinking”phase in reasoning models acts as a semantic bu!er that mitigates ty-pographic sensitivity in text. Extending the study to VLMs, we find thee!ect transfers partially: the same prompt-side casing reorganizes cross-modal attention along two coupled axes—predominantly a macroscopicdisengagement from the image toward the text prompt, and secondar-ily a concentration of the residual visual attention on the target region.By isolating casing as a zero-shot mechanism for attention steering thatrequires no model access or fine-tuning, we provide a new foundationalunderstanding of how pretraining internalizes typographic emphasis.

Recent advancements have explored agentic zero-shot 3D un-derstanding by reformulating it as video keyframe understanding withMultimodal Large Language Models (MLLMs). However, existing meth-ods face an intrinsic bottleneck due to the finite observation perspectivesinherent in videos and the implicit perception of 3D scenes. In this pa-per, we propose a collaborative multi-agent framework that assigns aPlanning Agent to handle high-level viewpoint planning and supplementnovel perspectives, and a Perception Agent to explicitly summarize the3D scene into a structured holistic cognitive map. Specifically, PlanningAgent first analyzes this cognitive map to determine query-relevant view-points and supplements missing critical perspectives to ensure compre-hensive observation. Subsequently, Perception Agent documents object-level attributes from these views by assigning consistent instance identi-fiers across viewpoints, thereby integrating fragmented observations intothe holistic cognitive map. In parallel, it provides feedback to filter outmismatched candidate objects and guide subsequent viewpoint planning.Through this closed-loop iterative process, two agents collaboratively fig-ure out candidates until Perception Agent determines that sufficient in-formation has been captured to complete the task. Extensive experimentsdemonstrate that our method achieves state-of-the-art performance on6 benchmarks, with improvements of 11.1% Acc@0.5 on ScanRefer, 14.6BLEU-1 on 3D-assisted dialog, and 2.1 EM on SQA3D. Project Page:https://zhangbo135.github.io/agentic-collaborative-cognition/

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