ECCV 2026 Papers

Generative models for volumetric medical images have foundmany applications in medical imaging, ranging from data augmentationto serving as priors for inverse problems. For these applications, generat-ing high-resolution 3D images with strong controllability is essential butremains highly challenging. Existing approaches typically control gener-ation either through radiology reports used as text prompts or throughfull image segmentation. While text-based prompting is flexible, it pro-vides limited spatial control over the location, shape, and boundary ofabnormalities. In contrast, segmentation-based methods receive precisespatial guidance but are restrictive in requiring full-organ annotations.In this work, we propose MedSynV2, a flexible multimodal framework forcontrollable volumetric image generation that supports input from radi-ology reports and segmentation prompts (both optional). Our approachallows users to provide segmentation of a specific anatomy or abnormalitywithout requiring full-organ annotations. The semantic meaning of thesegmentation mask is specified through an accompanying text descrip-tion, resulting in a highly flexible and scalable conditioning mechanism.We develop a memory-e!cient architecture based on a modified di"usiontransformer that jointly processes image and segmentation tokens. Themodel further incorporates gated attention to e"ectively attend to longradiology reports. Experiments demonstrate that our method achievesstate-of-the-art perceptual and semantic scores (e.g., → 24% relative im-provement in mean FID), generates high-resolution anatomically consis-tent CT volumes, and improves data e!ciency when used for data aug-mentation. Radiologists’ evaluation further confirms strong alignmentbetween generated and real medical images. We also explore the abilityof the model to generalize towards concepts beyond training data. Codewill be released at https: // github. com/ batmanlab/ MedSynV2 .

The structural vulnerabilities of point cloud–based 3D objectdetectors remain poorly understood. Prior work has studied adversarialrobustness primarily on isolated 3D object models, while recent LiDARspoofing attacks target richer and more realistic driving scenes but focusmainly on physical realizability rather than understanding detector be-havior or attack efficiency. In this work, we investigate how LiDAR-baseddetectors rely on spatial evidence in complex scenes and whether these re-liance patterns can be exploited to induce failures more efficiently. To thisend, we propose an explainability-guided adversarial analysis methodology.We introduce the Saliency-LiDAR (SALL) method, which aggregates In-tegrated Gradient attributions across scenes to produce universal saliencymaps for LiDAR-based 3D object detectors. Guided by these maps, wedesign the Explainability-aware Frustum Attack (EFA), which selectivelyperturbs only the most influential frustums rather than uniformly attack-ing entire object regions. Experiments on KITTI and nuScenes, acrossdetectors such as PointPillars and SECOND, show that EFA reducesdetection recall by more than 15 percentage points while requiring 25–50%fewer perturbed frustums than the SOTA non–saliency-aware baseline.These findings reveal that modern 3D detectors concentrate discrimina-tive evidence in a small subset of spatial regions, exposing a structuralrobustness vulnerability in current LiDAR perception systems. Our codeis released at https://github.com/SecMindLab/Saliency_LiDAR.

Developing effective dense crowd simulation is challengingdue to the scale gap between macroscopic collective coherence and mi-croscopic individual realism. Existing macroscopic (Eulerian) methodscapture global motion patterns but lack individual trajectories, whereasmicroscopic (Lagrangian) methods model individual behaviors yet oftenfail to preserve systemic consistency in dense scenarios. To address this,we propose the Eulerian–Lagrangian Hydrodynamics-Informed NeuralNetwork (ELHINN), a unified cross-scale framework that couples macro-scopic velocity evolution with microscopic trajectory refinement by usingevolved Eulerian velocity fields as physical priors to guide Lagrangian tra-jectories. For velocity evolution, we develop an enhanced Hydrodynamics-Informed Neural Network (HINN++) that incorporates a learnable gov-erning equation with Kolmogorov–Arnold Network (KAN) residual cor-rection and environmental boundary conditions, enabling accurate mod-eling of complex nonlinear dynamics under varying scenarios. For tra-jectory refinement, we use a Physics-Informed Neural Network (PINN)augmented with an entrance-aware resampling strategy (EARS) andcollision-avoidance constraints to ensure stability and fidelity. Experi-ments on two real-world crowd datasets demonstrate that ELHINN out-performs existing methods in simulating dense crowds across both scales.Codes are available at https://github.com/shanshan-zys/ELHINN.

End-to-End autonomous driving (E2E-AD) systems face chal-lenges in lifelong learning, including catastrophic forgetting, difficulty inknowledge transfer across diverse scenarios, and spurious correlationsbetween unobservable confounders and true driving intents. To addressthese issues, we propose DeLL, a Deconfounded Lifelong Learning frame-work that integrates a Dirichlet process mixture model (DPMM) withthe front-door adjustment mechanism from causal inference. The DPMMis employed to construct two dynamic knowledge spaces: a trajectoryknowledge space for clustering explicit driving behaviors and an implicitfeature knowledge space for discovering latent driving abilities. Leverag-ing the non-parametric Bayesian nature of DPMM, our framework en-ables adaptive expansion and incremental updating of knowledge withoutpredefining the number of clusters, thereby mitigating catastrophic for-getting. Meanwhile, the front-door adjustment mechanism utilizes theDPMM-derived knowledge as mediators to deconfound spurious correla-tions, such as those induced by sensor noise or environmental changes,and enhances the causal expressiveness of the learned representations.Additionally, we introduce an evolutionary trajectory decoder that en-ables non-autoregressive planning. To evaluate the lifelong learning per-formance of E2E-AD, we propose new evaluation protocols and metricsbased on Bench2Drive. Extensive evaluations in the closed-loop CARLAsimulator demonstrate that our framework significantly improves adapt-ability to new driving scenarios and overall driving performance, whileeffectively retaining previously acquired knowledge.

Vision-language-action models have advanced robotic ma-nipulation but remain constrained by reliance on the large, teleoperation-collected datasets dominated by the static, tabletop scenes. We propose asimulation-first framework to verify VLA architectures before real-worlddeployment and introduce MobileManiBench, a large-scale benchmarkfor mobile-based robotic manipulation. Built on NVIDIA Isaac Sim andpowered by reinforcement learning, our pipeline autonomously gener-ates diverse manipulation trajectories with rich annotations (languageinstructions, multi-view RGB–depth–segmentation images, synchronizedobject/robot states and actions). MobileManiBench features 2 mobileplatforms (parallel-gripper and dexterous-hand robots), 2 synchronizedcameras (head and right wrist), 630 objects in 20 categories, 5 skills(open, close, pull, push, pick) with over 100 tasks performed in 100 re-alistic scenes, yielding 300K trajectories. This design enables controlled,scalable studies of robot embodiments, sensing modalities, and policy ar-chitectures, accelerating research on data efficiency and generalization.We benchmark representative VLA models and report insights into per-ception, reasoning, and control in complex simulated environments, withall code, datasets, and models publicly released at our project website:https://dexhand.github.io/MobileManiBench/.

Visual Active Tracking (VAT) requires an embodied agent tocontrol cameras to follow a designated target, which is essential for ap-plications like robot navigation and security monitoring. However, trans-ferring VAT trackers to the real world faces two primary bottlenecks: (1)severe visual domain shifts between simulated environments and the realworld, leading to performance collapse during deployment; and (2) tar-get distribution shift caused by similar-looking distractors in real-worldscenes. To address these, we propose VATA, a novel test time adaptationmethod with two complementary strategies. First, we introduce a CriticValue Maximization strategy that leverages value signals from a criticmodel to filter reliable test samples, enabling unsupervised model up-dates to mitigate visual shift. Second, we propose an Action Smooth-ness Regularization term that exploits the continuity of real-worldtarget motion to bound action variations, correcting irrational actionsunder target shift. Experiments demonstrate VATA achieves a 24.4%SR improvement on the EVT-Benchmark and reaches 80% TSR in 10real-world scenarios (vs. 30% by the SOTA TTA method), providing arobust and plug-and-play solution for VAT deployment.

Diffusion models have become a dominant paradigm for conditional image generation, yet existing approaches generally follow two directions: task-specific designs that can improve performance but limit generalization, and training-free loss guidance that compresses rich conditions into scalar objectives and applies stepwise guidance, leading to information bottlenecks and error accumulation along the sampling trajectory. Given the urgent need for an effective unified framework across diverse conditional image generation tasks, we propose Data Injection and Contrastive Trajectory Refinement (DICT), a training-free inference method that enhances conditional image generation without introducing task-dependent architectures. DICT introduces Data Injection, where noise-perturbed conditional signals are integrated into early denoising stages; by performing guided denoising on these injected signals, DICT adaptively selects and distills task-salient information from the raw condition, effectively preserving spatial richness and ensuring precise condition-to-generation alignment. Furthermore, DICT applies Contrastive Trajectory Refinement across adjacent denoising states, enabling pairwise comparisons that progressively improve sample quality. These designs keep inference simple while improving cross-task transfer under a unified diffusion formulation. Extensive experiments on conditional image generation tasks (e.g., style transfer, image super-resolution, and image deblurring) show consistent gains in fidelity and perceptual quality over representative task-specific and loss-guided baselines.

Multispectral object detection enhances perceptual robust-ness in challenging environments by fusing complementary informationfrom RGB and infrared modalities. However, existing methods predomi-nantly follow a parameter-expanding paradigm, introducing customizedfusion modules that inevitably increase computational overhead and dis-rupt the topological consistency with unimodal pre-trained weights. Inthis paper, we challenge this convention and propose ZPA-MDETR,a zero-parameter adaptive multispectral framework without additionallearnable fusion parameters. By re-examining the DETR architecture,we reveal that its inherent query-memory interaction mechanism pro-vides a natural foundation for cross-modal fusion. Specifically, ZPA-MDETR strictly preserves the architecture of the unimodal DETR andachieves flexible and effective fusion via an asymmetric input organiza-tion strategy, where one modality serves as the memory, and the otherinitializes object queries. To bridge the cross-modal distribution dis-crepancy, we introduce parameter-free semantic alignment constraintsduring training. Extensive experiments on four public datasets acrossdifferent scenes demonstrate that ZPA-MDETR outperforms state-of-the-art methods while maintaining the same parameter scale as theunimodal framework, establishing a favorable trade-off between detec-tion accuracy and deployment efficiency. The source code is available athttps://github.com/UserXiangYang/ZPA-MDETR.

Achieving reliable single-shot high dynamic range (HDR)imaging under extreme illumination conditions remains a long-standingchallenge, yet no comprehensive benchmark exist for evaluating HDRperception in multi-sensor robotic systems. To fill this gap, we intro-duce a large-scale dataset collected via a custom robotic vision platformand an iPhone 13 Pro: 121 real-world scenes spanning modest and ultra-high dynamic range conditions, alongside 20 synthetic video sequencesfrom the CARLA simulator. As a reference pipeline for this dataset,we propose Depth-guided Multi-view Exposure Bracketing (DMEB), asingle-shot HDR method that distributes drastically different exposuresacross multi-view low-bit-depth cameras and fuses them via depth-guidedconfidence-aware fusion. Evaluations on our dataset show that DMEBestablishes a strong reference point and highlight the promise of thissensor configuration for robust HDR perception in diverse multi-cameraand depth sensor system.

Food cutting is a representative non-rigid manipulation taskinvolving deformation, contact forces, and material separation, yet itRobot Simulation remains largely unexplored in VLA research. Existing robot manipula-tion datasets are primarily built around rigid objects and struggle tojointly capture the spatial precision, deformation, topology changes, andCut the banana force interactions required for cutting. To systematically study theseat the center. challenges, we introduce CulinaryCut, a benchmark that integrates anniskill MPM-based deformable simulator with a robot environment to generatedataset rather than provided as policy inputs, enabling evaluation of ex-isting VLAs while exposing where physical grounding is needed. Usingthree representative VLA baselines, we identify two core limitations innon-rigid cutting. First, a geometry gap: VLAs struggle to map ratio-and direction-based instructions onto object geometry, especially whensequential cuts change the object’s topology. Second, a physics gap: be-cause policies output motion without a notion of material stiffness orcontact resistance, a geometrically accurate path may still fail to severthe object. We show that training VLAs on physics-grounded trajecto-ries, without modifying the policy architecture or adding force inputs,improves cutting success and sim-to-real transfer. Together, CulinaryCutprovides a foundation for evaluating spatial reasoning and physical plau-sibility in deformable manipulation.

Recent 3D Gaussian Splatting (3DGS) GANs for humanheads synthesize and render photorealistic 3D models in real-time andoffer a vast variety in identity and appearance. However, controlling spe-cific semantic attributes such as hair color or glasses remains challenging,as edits in the entangled latent space often induce unintended changesin identity or appearance. Although there are several methods that aimto disentangle the latent space post training by estimating directionsthat only modify certain features, these methods cannot guarantee com-plete disentanglement and often require pre-trained classifiers. In ourapproach, we propose a new generator architecture that synthesizes com-ponents, such as hair, skin, glasses, and torso, completely independently.This allows for changing the latent vector for one region while keepingthe remaining parts fixed. Further, we achieve this separation using onlysparse information such as the hair or skin color, eliminating the re-quirement of segmentation masks or geometric priors, often seen in priorwork. To ensure matching shape and lighting conditions during editing,we allow minimal shared information via context tokens between the in-dependent generators. These tokens even allow us to control the shapeand light, without any prior annotation. Compared to existing works onGAN-based generation and editing, our method shows better disentan-glement, more precise editing control, and competitive visual quality.

Recent multi-view 3D human pose estimation methods de-compose the problem into per-image 2D pose detection and 3D lift-ing stages, which can naturally scale to multi-frame input and achievestrong results by leveraging temporal context. However, this 2D-3D lift-ing paradigm introduces two levels of geometric inconsistency: (i) projec-tion inconsistency, i.e., the detected 2D keypoints may disagree with the2D projections of the inferred 3D pose, and (ii) cross-view inconsistency,i.e., 2D poses across views are mutually inconsistent because off-the-shelfdetectors operate independently per camera, so there may exist no sin-gle 3D pose whose projections match all detections. We observe that 2Ddetection errors closely follow a Gaussian distribution, meaning that de-tector outputs can be naturally viewed as noisy samples from a diffusionforward process. Based on this insight, we propose CoDePose, a coupled2D-3D denoising diffusion framework that jointly performs multi-viewdenoising of the detected 2D poses and estimates the underlying 3Dpose. As training proceeds, the generated joint distribution progressivelyapproaches the real data distribution, leading to increased projection andcross-view consistency and more reliable 3D estimation. Unlike existingdiffusion-based methods that condition on fixed 2D detections, CoDe-Pose refines the 2D observations during the reverse process, preventing2D detection errors and cross-view inconsistencies from being reinforcedthroughout the whole 3D pose generation. Experimental results on Hu-man3.6M show that CoDePose achieves 12.1 mm MPJPE, a 19.9% im-provement over the previous state-of-the-art method. Cross-dataset eval-uation on HumanEva and MPI-INF-3DHP further demonstrates stronggeneralization.

Document comprehension is a challenging yet impactful taskfor Multimodal Large Language Models, especially as these systems seegrowing adoption in real-world, human-centric applications. However,this adoption is limited for low-resource languages such as Bangla dueto the scarcity of high-quality annotated data. To address this gap, weintroduce BaFCo, a benchmark dataset for Bangla form comprehensionwith a focus on Document Layout Analysis (DLA) and Key InformationExtraction (KIE). BaFCo curates 200 multi-page complex Bangladeshigovernment forms, sourced from across diverse sectors including agricul-ture, education, banking, and land management. To accurately capturethe structural and contextual complexity of these forms, we define a fine-grained annotation schema comprising 26 types of form entities, alongwith a separate coarse form entity set consisting of 5 types. We evaluatethe latest MLLMs from the ChatGPT, Gemini, Claude, Qwen, and Kimiseries using zero-shot and chain-of-thought prompts under both low andhigh reasoning setups. Our results reveal limitations in current MLLMs’ability in comprehending Bangla forms, particularly in accurately local-izing highly granular form entities. Our dataset and code is available at:https://huggingface.co/datasets/Mausul/bafco

Enabling reliable long-horizon robotic manipulation is a cru-cial step toward open-world embodied intelligence. However, VLM-basedplanners treat each step as an isolated observation-to-action mapping,forcing them to reinfer scene geometry from raw pixels at every decisionstep while remaining unaware of how prior actions have reshaped the en-vironment. Despite strong short-horizon performance, these systems lackthe spatio-temporal reasoning required for persistent geometric anchoringand memory of action-triggered state transitions. Without persistentstate tracking, perceptual errors accumulate across the execution horizon,temporarily occluded objects are catastrophically forgotten, and com-pounding failures lead to precondition violations that cascade throughsubsequent steps. In contrast, humans maintain a persistent mental modelthat continuously tracks spatial relations and action consequences acrossinteractions rather than reconstructing them at each instant. Inspiredby this human capacity for causal spatio-temporal reasoning with persis-tent memory, we propose RoboStream, a training-free framework thatachieves geometric anchoring through Spatio-Temporal Fusion Tokens(STF-Tokens), which bind visual evidence to 3D geometric attributesfor persistent object grounding, and maintains causal continuity via aCausal Spatio-Temporal Graph (CSTG) that records action-triggeredstate transitions across steps. This design enables the planner to tracecausal chains and preserve object permanence under occlusion withoutadditional training or fine-tuning. RoboStream achieves a 90.5% successrate on long-horizon RLBench tasks and a 44.4% success rate on challeng-ing real-world block-building tasks, where both SoFar and VoxPoser score11.1%, demonstrating that spatio-temporal reasoning and causal memoryare critical missing components for reliable long-horizon manipulation.

In person re-identi_x001C_cation, neighbor-based methods haveachieved signi_x001C_cant success by interacting with neighbor samples to ob-tain more robust representations. However, existing methods rely onlyon a_x001E_nity relations, causing their success to depend heavily on the re-liability of selected neighbors. We _x001C_nd that a_x001E_nity-only interaction of-ten fails in challenging scenarios due to the inevitable presence of noisyneighbors. To enable e_x001B_ective interactions under noisy neighborhoods, werevisit neighbor-based methods under distinct reliability conditions andpropose a novel Adaptive Neighbor Feature Interaction (ANFI)method. The core idea of ANFI is to account for negative e_x001B_ects fromnoisy neighbors, allowing samples to remain distinguishable from falsepositive neighbors. Unlike existing methods, ANFI models not only a_x001E_n-ity relations but also discrepancy relations, and employs sample-wiseadaptive weighting for these two types of relations. Given that capturingnegative e_x001B_ects from noisy neighbors di_x001B_ers signi_x001C_cantly from traditionalrelation learning, we derive discrepancy relations from a new neighbor-hood similarity, which provides more information than pairwise sim-ilarity. In addition, we propose Noisy Relation Supervision (NRS)to train ANFI, gradually injecting robustness to noisy relations into themodel. Extensive experiments conducted under standard, cross-modal,and cross-domain settings, including comparisons with neighbor-basedmethods and re-ranking methods, demonstrate the superiority of ourmethod across various neighbor distributions.

Image editing has become an important capability of modernvisual generative models. Most existing approaches generate the targetimage directly given an input image and an editing instruction. Whileeffective for simple appearance modifications, this paradigm often failswhen edits require multi-step inference over spatial, temporal, causal, orcommonsense relationships, which entails reasoning about how the sceneshould evolve through a series of plausible intermediate states from thesource image to the desired result. Notably, modern video generationmodels naturally capture such transformation processes: by modelinghow scenes evolve over time, they encode strong temporal coherence andimplicit physical priors. However, directly applying video models to imageediting remains challenging due to the mismatch between editing instruc-tions and video generation objectives. To bridge this gap, we proposeVGEdit, a reinforcement learning framework that empowers video gener-ation models for image editing. Our method consists of two components:(1) Instruction transfer : an MLLM converts the editing instruction into avideo description, reframing the edit as a plausible transformation process.(2) Reinforcement learning: we adapt the video model with reward-basedoptimization, treating intermediate frames as implicit reasoning chainsand the last frame as the target edited image. This formulation eliminatesthe need for ground-truth editing videos and enables direct optimizationfrom rewards that capture instruction faithfulness, contextual consistency,and user preference alignment. Experiments show that VGEdit signifi-cantly outperforms prior methods on reasoning-informed image editingbenchmarks, demonstrating that connecting image editing and videogeneration in a process-centric manner provides a scalable path towardmore physically grounded and reasoning-capable visual editing.

Camera control has been extensively studied in conditional video generation; however, stable altering camera trajectories while faithfully preserving video content remains a challenging task. The mainstream approach to achieving camera control is warping a 3D representation according to the target trajectory. However, such methods fail to fully leverage the 3D priors of video diffusion models (VDMs) and often fall into the Inpainting Trap, resulting in subject inconsistency and degraded generation quality. To address this problem, we propose DepthDirector, a video re-rendering framework with stable camera controllability. Our key insight is to decompose the conditional injection of video content and camera trajectories by leveraging the depth video from explicit 3D representation as camera-control guidance. Specifically, we design a View-Content Dual-Stream Condition mechanism, enabling VDMs to comprehend camera movements and leverage their 3D understanding capabilities. Additionally, we construct a large-scale multi-camera synchronized dataset named MultiCamWarp Dataset using Unreal Engine 5. Extensive experiments show that DepthDirector outperforms existing methods. Our code and dataset will be publicly available on our project website: https://eleanor6725.github.io/DepthDirector/.

Embodied Visual Tracking (EVT) requires an agent to con-tinuously follow a language-specified target from egocentric observations.Most existing methods adopt single-agent imitation learning, which de-pends on expert demonstrations and can struggle to generalize to sce-narios beyond the training distribution. Inspired by competition-drivencapability evolution in nature, we propose CoMaTrack, a competitivemulti-agent reinforcement learning (RL) framework that trains agentsin a dynamic adversarial setting with competitive subtasks, yieldingstronger adaptive planning and interference-resilient strategies. We fur-ther propose CoMaTrack-Bench, a Habitat-based benchmark for language-conditioned competitive EVT, featuring dynamic dueling scenarios be-tween a tracker and adaptive opponents across diverse environments andinstructions, enabling standardized robustness evaluation under activeadversarial interactions. Experiments show that CoMaTrack achievesstate-of-the-art results on both standard benchmarks and CoMaTrack-Bench. Notably, a 3B VLM trained with our framework surpasses pre-vious single-agent imitation learning methods based on 7B models onthe challenging EVT-Bench, achieving 92.1% in STT, 74.2% in DT, and57.5% in AT.

Pixel-space diffusion has recently re-emerged as a strong al-ternative to latent diffusion, enabling high-quality generation withoutpretrained autoencoders. However, standard pixel-space diffusion mod-els receive relatively weak semantic supervision and are not explicitlydesigned to capture high-level visual structure. Recent representation-alignment methods (e.g., REPA) suggest that pretrained visual featurescan substantially improve diffusion training, and visual co-denoising hasemerged as a promising direction for incorporating such features intothe generative process. However, existing co-denoising approaches of-ten entangle multiple design choices, making it unclear which are trulyessential. We therefore present V-Co, a systematic study of visual co-denoising in a unified JiT-based framework. This controlled setting allowsus to isolate the ingredients that make visual co-denoising effective. Ourstudy reveals two main ingredients. First, co-denoising benefits from pre-serving feature-specific computation while enabling flexible cross-streaminteraction, which leads to a fully dual-stream architecture together witha structurally defined unconditional prediction for classifier-free guid-ance. Second, it requires both stronger semantic supervision and propercross-stream calibration, which we realize through a perceptual-driftinghybrid loss and RMS-based feature rescaling. Together, these findingsyield a simple recipe for visual co-denoising. Experiments on ImageNet-256 show that, at comparable model sizes, V-Co outperforms the un-derlying pixel-space diffusion baseline and strong prior pixel-diffusionmethods while using fewer training epochs, offering practical guidancefor future representation-aligned generative models.