ECCV 2026 Papers

Real-time monocular 3D object detection remains challeng-ing due to severe depth ambiguity, viewpoint shifts, and the high compu-tational cost of 3D reasoning. Existing approaches either rely on LiDARor geometric priors to compensate for missing depth or sacrifice efficiencyto achieve competitive accuracy. We introduce LeAD-M3D, a monocu-lar 3D detector that achieves state-of-the-art accuracy and real-time in-ference without extra modalities. Our method is enabled by three keycomponents. Asymmetric Augmentation Denoising Distillation (A2D2)transfers geometric knowledge from a clean-image teacher to a MixUp-noised student via a quality- and importance-weighted depth-feature loss,enabling stronger depth reasoning without LiDAR. 3D-aware ConsistentMatching (CM3D ) improves prediction-to-ground truth assignment by in-tegrating 3D MGIoU into the matching score, yielding stable and precisesupervision. Finally, Confidence-Gated 3D Inference (CGI3D ) acceleratesinference by restricting expensive 3D regression to confident regions. To-gether, these contributions set a new Pareto frontier for monocular 3Ddetection: LeAD-M3D achieves state-of-the-art accuracy on KITTI andWaymo, and the best reported car AP on Rope3D, while running up to3.6 × faster than prior high-accuracy models (e.g., MonoDiff). LeAD-M3D demonstrates that high fidelity and real-time monocular 3D de-tection is simultaneously attainable, without LiDAR, stereo, or stronggeometric assumptions.

Controllable image generation methods, such as ControlNet,have demonstrated a remarkable capacity to introduce visual conditions(e.g., depth maps) to guide image generation. However, these methodsoften struggle with complex multi-instance scenes, frequently leading toattribute confusion among instances. While recent approaches attemptto mitigate this via manual instance labeling, such requirements arelabor-intensive. In this paper, we propose InstanceControl, a novel multi-instance controllable generation method that eliminates the need for in-stance labeling. We identify the primary bottleneck in existing meth-ods as the inability to accurately associate instance descriptions withtheir corresponding regions within visual conditions. To address this, weleverage the Vision-Language Model (VLM) to establish instance-levelcorrespondences between text prompts and visual conditions. Specifi-cally, the VLM automatically parses instance descriptions from the textprompts and simultaneously predicts instance masks based on the visualconditions. Furthermore, since the predicted masks may contain noise,we introduce an adaptive mask refinement strategy that dynamically re-fines these instance masks during the generation process. Extensive ex-periments demonstrate that our approach outperforms state-of-the-artmethods, achieving superior fidelity and precise instance-level control.The code and dataset will be made publicly available.

Functional 3D scene graphs offer a versatile and flexible rep-resentation for 3D scene understanding and robotic manipulation, de-fined by object nodes, interactive elements, and functional relationshipedges. However, their potential remains underexplored due to the limitedcoverage of existing benchmarks and the overly straightforward designof previous pipelines, which primarily focus on large-scale furniture butlack of hierarchical structures. Therefore, in this work, we extend thebenchmark coverage by introducing dense tabletop objects and explicitmulti-level functional relationships. This expansion introduces criticalchallenges involving small-scale, dense, and similar instances, with lack ofvisual anchoring in relational reasoning, instance confusion during cross-frame fusion, and attribution uncertainty under dynamic viewpoints. Toaddress these issues, we propose an open-vocabulary pipeline based on 2Dvisual grounding and 3D graph optimization. Specifically, we anchor fine-grained functional edges from 2D visual evidence, and associate nodesacross frames in 3D using multiple cues. Furthermore, edge associationis formulated as temporal graph optimization, integrating evidence ac-cumulation, entropy regularization, and temporal smoothing to robustlydetermine the functional connections of each node. Finally, global hierar-chy shaping is performed to recover the hierarchical graph structure. Ex-tensive experiments demonstrate that the proposed method can reliablyinfer functional 3D scene graphs in challenging real-world scenes, therebyfurther unlocking their potential for practical applications. Code is avail-able at https://github.com/Hbelief1998/HHOpenFunGraph-ECCV26.

Recent advancements in vision-language models (VLMs) haveleveraged large language models (LLMs) to achieve performance on parwith closed-source systems like GPT-4V. However, deploying these mod-els in real-world scenarios, particularly on resource-constrained devices,remains challenging due to their substantial computational demands.This has spurred interest in distilling knowledge from large VLMs intosmaller, more efficient counterparts. A key challenge arises here from thediversity of VLM architectures, which are built on different LLMs andemploy varying token types—differing in vocabulary size, token splits,Changing Teacher VLMs for Distillation MMMU70GenRecalMMB GPT-4oInternVL2.5-78B → InternVL2.5-8B60 Qwen2-VLAccuracy (%)MM-Vet InternVL2.550 MolmoMMMU40NVLM-72B → InternVL2.5-8B LLaVA-OneVisionInternVL2-76B → InternVL2.5-8BMMMU-Pro Qwen2-VL-72B → InternVL2.5-8BInternVL2.5-78B → InternVL2.5-8B300 20 40 60 80 1 2 4 7 8 … 72 78Accuracy (%) Model Size (B)Fig. 2: (Left) Comparison of the challenging benchmark performances, MMB [36], MM-Vet [63], MMMU [65], and MMMU-Pro [66] by changing large VLMs. The more pow-erful large VLMs we select, the greater the performance improvement we can achieve.(Right) Comparing the performance of the challenging benchmark: MMMU [65], withGenRecal and various VLMs across model sizes. Note that all the experiments in Fig. 1and Fig. 2 are conducted on the equal training dataset.and token index ordering. To address this challenge of limitation to aspecific VLM type, we present Generation after Recalibration (Gen-Recal), a general-purpose distillation framework for VLMs. GenRecalincorporates a Recalibrator that aligns and adapts feature representa-tions between heterogeneous VLMs, enabling effective knowledge trans-fer across different types of VLMs. Through extensive experiments onmultiple challenging benchmarks, we demonstrate that GenRecal signif-icantly improves baseline performances, eventually outperforming large-scale open- and closed-source VLMs.

This paper presents DINO-SLAM, a DINO-informed designstrategy to enhance implicit (Neural Radiance Field – NeRF) and explicitrepresentations (Gaussian Splatting – GS) in SLAM systems through themore comprehensive semantic understanding enabled by DINO. This lat-ter alone, however, lacks proper 3D geometry understanding, allowingonly for marginal improvements. Therefore, we rely on a Scene Geom-etry Encoder (SGE) to lift DINO features into geometry-aware DINOfeatures (geoDINO), to better understand those geometric relationshipsthat vanilla DINO features fail to capture. Building upon it, we pro-pose two foundational paradigms for NeRF and GS SLAM systems in-tegrating geoDINO features. Compared to state-of-the-art methods, ourDINO-informed pipelines achieve superior performance on the Replica,ScanNet, and TUM datasets.

Many challenges in scientific imaging involve solving ill-posedinverse problems, where the goal is to recover spatio-temporal fields fromindirect, noisy, and highly sparse measurements — often without accessto ground truth data or reliable simulators. To address this challengingscenario, we present NeuralDMD, an interpretable, untrained (per-instance) reconstruction framework that combines neural implicit rep-resentations with Dynamic Mode Decomposition (DMD) to reconstructcontinuous spatio-temporal dynamics directly from measurements. Neu-ralDMD parameterizes DMD modes as continuous neural fields, andimposes a low-rank linear dynamics prior with spectral time evolutionto enforce temporal continuity. This formulation enables both forecast-ing under sparsity, and yields interpretable modes and spectra. We findthat NeuralDMD outperforms baselines on a wide variety of tasks: fromweather data assimilation from sparse station observations to interfero-metric (Fourier domain) observations of Sagittarius A*, the black hole atthe center of our galaxy. Moreover, NeuralDMD remains stable whenextrapolating into the future. While this framework is most naturallysuited to linear dynamics, we show that it can be applied to nonlinearregimes, though with extrapolation performance that degrades with in-creasing nonlinearity. Together, these results show that NeuralDMDenables interpretable reconstruction and forecasting of spatio-temporaldynamics directly from sparse and indirect measurements without rely-ing on numerical simulators or training data.

Adapting video foundation models to volumetric medicalsegmentation by treating axial depth as a temporal dimension is attrac-tive, but the resulting slice propagation can be unstable when anatom-ical structures branch, disappear, or become poorly contrasted acrossslices. Static geometric prompts provide little guidance for newly ap-pearing or disconnected components, and SAM2-style streaming mem-ory may accumulate errors in these regions. We propose Di_x001B_uPrompt,a backbone-frozen, single-prompt framework that augments SAM2 withlatent trajectory priors for volumetric propagation. Given a user prompton a representative slice, an LLM-conditioned Neural ODE predicts acontinuous, patient-conditioned latent trajectory and converts it into apersistent prior memory. A dual-stream spatial gate then fuses this priormemory with the standard observation-driven working memory, allowingthe model to use image evidence when it is reliable and trajectory pri-ors when local propagation becomes ambiguous. Experiments on MSD,BTCV, and BraTS show improved promptable volumetric segmentationover SAM/SAM2-based baselines, especially on boundary-sensitive met-rics.

Image Manipulation Localization (IML) aims to achieve pixel-level localization of locally manipulated images. Its core challenge is howto distinguish the inconsistencies between authentic and tampered ar-eas. To this end, existing IML methods typically exploit patch-level andedge-level forensic cues for accurate localization. However, these methodsextract forged information at both levels with fixed parameters, leadingto interference from real information that is detrimental to localizationas more realistic manipulated textures are enhanced by the upgrading oftampering techniques. To tackle the above issue, this paper proposes aDisentangling and Gathering Forensic Cues method, called DG-Force,which explicitly disentangles and adaptively aggregates region and edgecues to preserve the most informative evidence. Specifically, we proposea Patch-based Forensic Disentangling (PFD) module and an Edge-basedForensic Disentangling (EFD) module to decompose forensic cues of dif-ferent granularities to explicitly suppress the imbalance between forensictraces in both patch-level and edge-level areas. In addition, a Multi-scale Forensic Transfer (MFT) module is further designed to aggregateand balance multi-granularity information through the interaction acrossdifferent scales for robust key inconsistencies in forensic traces discovery.Extensive experiments on multiple benchmarks demonstrate the superi-ority of our method in the image manipulation localization task.

Feed-forward 3D reconstruction models such as DUSt3R, VGGT, and Depth Anything 3 (DA3) are transformer–based foundation models that infer camera geometry and dense scene structure in a single forward pass. Trained at scale in a supervised fashion, they raise a central question: do these models build upon geometric principles akin to traditional multi–view pipelines, or do they primarily rely on learned priors arising from the large–scale training setup? We find that epipolar geometry emerges within the intermediate layers of all three models and is causally linked to correspondence patterns in attention heads. To study this, we perform a systematic analysis of their internal representations across three real–world datasets and a controlled synthetic dataset. We quantify geometric understanding by probing intermediate features, analyzing attention patterns to identify correspondence matching patterns, and performing targeted interventions at the attention level. Further, we assess the role of learned priors by applying challenging input–level perturbations, such as occlusions, scene ambiguities, and varying camera configurations, and compare them against classical multi–stage reconstruction pipelines.

Infrared small target detection (IRSTD) in high-resolutionimages is crucial for many practical applications, such as surveillanceof unmanned aerial vehicles (UAVs) and UAV-based ground monitoring.However, IRSTD remains challenging due to the small size and weakfeatures of targets, as well as significant interference from complex dy-namic backgrounds. Existing detection methods often suffer from redun-dant computations on non-target background regions and insufficient ex-ploitation of target context information, which limits their performancein complex backgrounds. To address these issues, we propose an efficientcoarse-to-fine infrared small target detection framework with attentionprior-guided knowledge distillation, termed ECFNet. In the coarse stage,we design a region binary classification network (RBCN) on grid-basedmulti-scale feature maps to efficiently recognize target-containing con-text region proposals. Moreover, we introduce a novel denoising-assistedtraining strategy that incorporates noisy ground-truth (GT) masks intoRBCN feature maps. By training the network to reconstruct the GTmasks via a denoising task, this strategy encourages the model to ex-plicitly capture target-background context, thereby enhancing its abil-ity to distinguish target proposals from background regions. In the finestage, we customize a lightweight target detector to the coarse stage’sregion proposals for balancing accuracy and efficiency. Furthermore, wepropose a knowledge distillation strategy guided by the teacher-studentcross-attention prior. This mechanism directs the student to focus oncritical target regions, thereby enhancing the discriminative feature rep-resentation for infrared small targets. Extensive experiments on threereal infrared datasets demonstrate that our method outperforms bothexisting single-stage and two-stage approaches while maintaining highreal-time processing efficiency.

Predicting object dynamics (i.e., world modeling) is a fun-damental challenge for robotic manipulation, and modeling deformableobjects presents a particularly difficult case due to their high-dimensionalstate spaces and complex material properties. While current world mod-els approach this through two distinct paradigms: learning the dynamicsover the 2D pixel space or more explicit 3D geometric space. A systematicunderstanding of their relative strengths and limitations remains elusivedue to the lack of diverse, large-scale real-world data. To address this,we present Deform360, a large-scale visuotactile dataset featuring 198daily-life objects, 1,980 interaction sequences, and over 215 hours of ob-servations from 41 surround-view cameras and bimanual tactile grippersto capture both global motion and contact-induced local deformations.Leveraging a novel markerless visuotactile 3D tracking pipeline to ex-tract dense geometry and motion, we systematically evaluate currentstate-of-the-art world models, comparing 2D video models against 3Dparticle models. Finally, we provide a preliminary demonstration indi-cating the real-world applicability of our dataset by performing robotplanning tasks on deformable objects. Our analysis reveals key insightsinto the trade-offs between structural priors and scalability, providing asolid benchmark for future research in generalizable deformable object-centric world modeling. Project website: https://deform360.lhy.xyz

Understanding 3D objects from images is fundamental to robotics and AR/VR applications. While recent work has made progress in category-level pose estimation, current representations fail to capture the fine-grained semantics needed for reasoning about object parts, functions, and interactions. In this work, we study category-level 3D correspondence in camera space—predicting, from a single image, 3D locations that remain consistent across instances within a category—and show that it can emerge without explicit correspondence supervision by learning a shared morphable object prior. To enable research in this direction, we introduce HouseCorr3D, the first large-scale benchmark for single-view category-level 3D correspondence with 178k images across 50 household object categories, 280 unique instances, and 3D keypoint annotations directly on CAD models. Crucially, HouseCorr3D provides amodal correspondence labels for occluded regions and explicit symmetry annotations, addressing key limitations of existing datasets. We further propose Morpheus, a method that learns morphable category-level shape priors by disentangling canonical shape, deformation, and object pose. Through this shared canonical grounding, semantically meaningful 3D correspondences in camera space emerge implicitly. These emerging 3D correspondences set a new state of the art on HouseCorr3D, demonstrating that semantic 3D object understanding can arise without direct correspondence supervision. Data and code: §/GenIntel/HouseCorr3D.

Shadows encode rich information about scene geometry andillumination, yet existing methods either predict a unified shadow maskor overlook attached shadows entirely. We address this gap by propos-ing a framework for jointly detecting cast and attached shadows throughexplicit physical modeling of light direction and surface geometry un-der a dominant directional-light setting. Our approach is grounded in asimple observation: surfaces facing away from the light source tend tofall into shadow. We exploit the reciprocal relationship between shadowformation and light estimation to construct a closed feedback loop, adual-module architecture in which a shadow detection module and alight estimation module iteratively refine each other. At each pass, up-dated light estimates, together with surface normals, produce partial at-tached shadow maps that guide detection, while improved shadow pre-dictions sharpen light estimation. To support training and evaluation,we introduce a dataset of 1,458 images with manually annotated castand attached shadow masks sourced from three existing benchmarks.Experiments demonstrate that our proposed method outperforms priormethods, with at least a 33% reduction in attached-shadow BER, whilemaintaining strong full-shadow and cast-shadow performance. Projectpage: https://shilin21.github.io/attached_detection/

The next frontier for video generation lies in developing models capable of zero-shot reasoning, where understanding real-world scientific laws is crucial for accurate modeling of physical outcomes under diverse conditions. However, existing video benchmarks are physical commonsense-based, offering limited insight into video models’ scientific reasoning capability. We introduce VideoScience-Bench1, a benchmark for evaluating undergraduate-level scientific understanding in video models. Each prompt encodes a composite scientific scenario that requires understanding and reasoning across multiple scientific concepts to generate the correct phenomenon. The benchmark comprises 200 evaluation items built from 160 expert-curated test cases (160 text-to-video prompts; 40 also include image conditioning) spanning 14 topics and 103 concepts. We conduct expert-annotated evaluations across seven state-of-the-art video models in T2V and I2V settings, along five dimensions: Prompt Consistency, Phenomenon Congruency, Dynamism, Immutability, and Spatio-Temporal Coherence. Using a VLM-as-a-Judge to assess video generations, we observe strong correlation with human assessments. To the best of our knowledge, VideoScience-Bench is the first benchmark to evaluate video models not only as generators but also as reasoners, requiring their generations to demonstrate scientific understanding consistent with expected physical and chemical phenomena.

Visual autoregressive (VAR) models synthesize images in acoarse-to-fine manner by predicting multi-scale token maps, achievingcompetitive text-to-image quality with fewer decoding iterations thanraster-scan autoregression. However, late-scale refinement dominates in-ference time because computation grows quadratically with resolutionand attention becomes increasingly costly. Prior accelerations either fil-ter tokens heuristically, risking severe artifacts, or rely on extra deci-sion models that require per-model fine-tuning with reduced flexibility.We present the first in-depth study of attention behavior across VARgeneration steps and uncover two key properties: (1) attention patternscorresponding layers remain consistent across steps; (2) conditional andunconditional branches progressively converge at later scales. Leveragingthese insights, we introduce a novel training-free, flashattention-friendlyacceleration recipe ARVAR with two primary components: (1) Atten-tion Retrospect (AR) trims tokens at high resolution phase by to-ken saliency from early stage; (2) Branch Halving (BH) drops oneof the conditional/unconditional branches in the late steps. Combined,AR and BH deliver faster and higher-quality autoregressive image gen-eration, achieving 1.56× and 3.01× speedups on HART and Infinityrespectively with negligible performance drop. Compared with existingacceleration methods, ARVAR substantially improves image fidelity.

In real-world deployment, vision-language models often en-counter disturbances such as weather, occlusion, and camera motion.Under such conditions, their understanding and reasoning degrade sub-stantially, revealing a gap between clean, controlled (i.e., unperturbed)evaluation settings and real-world robustness. To address this limitation,we propose ROVA, a novel training framework that improves robustnessby modeling a robustness-aware consistency reward under spatio-temporalcorruptions. ROVA introduces a difficulty-aware online training strategythat prioritizes informative samples based on the model’s evolving capa-bility. Specifically, it continuously re-estimates sample difficulty via self-reflective evaluation, enabling adaptive training with a robustness-awareconsistency reward. We also introduce PVRBench, a new benchmark thatinjects real-world perturbations into embodied video datasets to assessboth accuracy and reasoning quality under realistic disturbances. Weevaluate ROVA and baselines on PVRBench, UrbanVideo, and VisBench,where open-source and proprietary models suffer up to 35% and 28%drops in accuracy and reasoning under realistic perturbations. ROVAeffectively mitigates performance degradation, boosting relative accuracyby at least 24% and reasoning by over 9% compared with baseline models(QWen2.5/3-VL, InternVL2.5, Embodied-R). These gains transfer to cleanstandard benchmarks, yielding consistent improvements.

Affordance prediction, which identifies interaction regionson objects based on language instructions, is critical for embodied AI.Prevailing end-to-end models couple high-level reasoning and low-levelgrounding into a single monolithic pipeline and rely on training over an-notated datasets, which leads to poor generalization on novel objects andunseen environments. In this paper, we move beyond this paradigm byproposing A4-Agent, a training-free agentic framework that decouplesaffordance prediction into a three-stage pipeline. Our framework coordi-nates specialized foundation models at test time: (1) a Dreamer thatemploys generative models to visualize how an interaction would look;(2) a Thinker that utilizes large vision-language models to decide whatobject part to interact with; and (3) a Spotter that orchestrates vi-sion foundation models to precisely locate where the interaction area is.By leveraging the complementary strengths of pre-trained models with-out any task-specific fine-tuning, our zero-shot framework significantlyoutperforms state-of-the-art supervised methods across multiple bench-marks and demonstrates robust generalization to real-world settings.

VLM-Parser object features purpose •••3-View VLM Parse ManualImages Stool AssemblyGeneralizingAssembly VLM-PlannersequenceInferVLMUpdateAssembly LampAttempts sequence Assembly(a) Digital Twins System (b) Parser and Planner (c) Zero-shot AssemblyFig. 1: Illustrating the core concept of ZAP: a VLM-guided agent transforms multi-view RGB-D observations into an assembly plan through internal reasoning, while adigital twin is used for geometry-based feasibility verification.

Dynamic view synthesis has seen significant advances, yet re-constructing scenes from uncalibrated, casual videos remains challengingdue to ambiguity between camera and object motion, inadequate viewcoverage and slow optimization. In this work, we present Minute4D,a novel and efficient approach for high-fidelity 4D scene reconstructionfrom monocular videos using Gaussian Splatting. Our Gaussian initial-ization begins with an efficient geometric recovery leveraging pre-trainedvisual foundation models. To improve reconstruction fidelity, we intro-duce a Segmentation-Tracking Enhancement module that leverages 2Dsemantic priors and 3D point tracking to jointly enhance the geometricaccuracy and completeness of the initial Gaussians. To reduce redun-dancy and accelerate Gaussian optimization, we propose a Loss-GuidedDensity Control strategy that adaptively densifies and prunes Gaussiansbased on multi-view photometric loss. Our method reduces the optimiza-tion time to approximately 40 seconds for a typical 400-frame video,yielding a speed-up of at least 5×. Extensive experiments show thatour method not only enables substantially faster optimization but alsoachieves superior performance across several benchmarks.*: Equal contribution.†: Co-corresponding authors.