ECCV 2026 Papers

Video diffusion transformers (DiTs) generate high-fidelityand temporally coherent videos, yet motion control remains implicit,primarily relying on text prompts. As a result, achieving desired motionoften requires extensive prompt engineering and repeated resampling.While fine-tuning models with additional spatial prompts (e.g., bound-ing boxes or point trajectories) enables explicit control, it demands sub-stantial data curation and computation, and may compromise the gener-ative capabilities of pretrained models. Consequently, training-free mo-tion control using such spatial prompts has been explored in U-Net–basedvideo diffusion models, but remains largely unexplored for DiTs. We in-troduce Qwerty, a training-free framework that enables flexible motioncontrol in pretrained image-to-video DiTs via user-defined object warpingand optical flow. We carefully manipulate the 3D full attention of DiTsby warping the frame-invariant semantic subspace of queries. We findthat the noise predicted by the query-warped DiT naturally guides thediffusion trajectory toward the desired motion, and further show thatleveraging this noise as self-guidance for latent optimization improvescontrol stability and visual quality. Experiments show that Qwertyachieves the most effective motion control among existing training-freeapproaches on a recent image-to-video DiT, with performance compara-ble to fine-tuning–based methods.

Hallucinations in medical vision language models arise whenlanguage priors override subtle visual evidence during decoding. Existingtraining-free contrastive decoding methods suppress these priors by con-trasting logits against a globally perturbed image, but global perturba-tions destroy the very diagnostic cues they are meant to protect —stan-dard VCD blur flips nearly one in five CheXpert labels on held-out chestradiographs. We propose Proximity-Constrained CounterfactualDecoding (PCCD), which admits a counterfactual view only whenit satisfies both a global and an object-masked feature-space similaritybound, constraining diagnostic drift while increasing the contrastive mar-gin for visually grounded tokens to first order under explicitly stated as-sumptions. Two complementary branches address heterogeneous halluci-nation drivers: Object-Aware VCD preserves attribution-indicated lesionregions while attenuating contextual co-occurrence priors, and Latent-Disentangled VCD applies structure-preserving style edits to suppressattribute biases. When neither branch passes the proximity gate, PCCDfalls back to greedy decoding, guaranteeing it never degrades below theunmitigated baseline. Across radiology VQA, chest X-ray report gener-ation, and a new CXR existence probe (Med-POPE), PCCD improvesGREEN AUC, QAAS, and RadGraph F1 by up to →+2.5 pp over SOTAwith one auxiliary forward pass per token and no weight updates.

We introduce PRISM-VO, a novel pure optimization-basedsparse photometric visual odometry framework for focused plenopticcameras. The core of PRISM-VO is a novel photometric plenoptic bun-dle adjustment which jointly optimizes camera poses and inverse depthvalues of points in a sliding window. By combining geometric depth froma single plenoptic image with temporal multi-view constraints, PRISM-VO achieves accurate and drift-resilient motion estimation. Through ex-plicit modeling of the plenoptic projection, PRISM-VO provides reliablemetric-scale reconstructions, overcoming the scale ambiguity of monocu-lar SLAM algorithms. Importantly, our approach relies solely on a singleplenoptic sensor and avoids complex initialization, as depth priors arecomputed directly from plenoptic imaging.Experiments show that PRISM-VO outperforms the current state-of-the-art plenoptic visual odometry method on indoor and outdoor scenes.The proposed approach rivals other optimization- and learning-basedmethods while accurately and reliably recovering a metric scale of thescene.Project page: https://prism-vo.github.io/.

Multi-View Pedestrian Detection (MVPD) aims to detectpedestrians in the form of a bird’s eye view map from multi-view images.Recent MVPD methods adopt a unified framework that projects 2D im-age features into a 3D world space and aggregates them into a singlefeature. Although they are effective, they struggle to generalize to un-seen camera configurations during training due to two main issues. First,they are difficult to capture accurate visual geometry across views in un-seen camera configurations. Second, they make detection models highlydependent on distortion patterns during training arising from their im-age feature projection. To address these, we leverage a visual geometricfoundation model and propose MV2GF. This foundation model has ex-hibited strong generalization in capturing visual geometry across viewsand predicting accurate 3D attributes in diverse camera configurations.MV2GF fuses task-specific features with general-purpose geometric fea-tures extracted by the foundation model to effectively capture the visualgeometry even in unseen camera configurations. Furthermore, MV2GFprojects each pixel in the image features to an appropriate 3D locationusing 3D pointmaps predicted by the foundation model, preventing thedetection model from depending on distortion patterns during training.Our experiments demonstrate the effectiveness of leveraging a visual geo-metric foundation model for MVPD and that MV2GF generalizes betterthan existing methods.

Test-time adaptation (TTA) aims to mitigate distributionshifts by adapting models with unlabeled target data at inference time.While TTA with vision-language models (VLMs) has shown promisingresults in classification, extending it to medical image segmentation re-mains challenging. In this setting, the adaptation gains from optimizingon VLM-generated predictions are often outweighed by the degradationto the VLM’s strong pretrained features caused by noisy, update-drivenlearning, resulting in limited and unstable improvements. We thereforepropose Memory-Supported Synergistic Adaptation (MSSA), anovel training-free TTA framework for medical image segmentation.Without updating model parameters, MSSA dynamically selects reli-able image–text predictions to construct an online memory, uses themas text-guided semantic priors, and couples them with cross-image struc-tural alignment for robust adaptation. Specifically, MSSA consists of (i) anoise-aware memory construction module that filters and stabilizes cross-modal predictions, and (ii) a relevance-driven prototype alignment mod-ule that aligns the target sample with structurally consistent memorysamples and their reliable predictions to improve adaptation. Extensiveexperiments on multiple medical segmentation benchmarks demonstratethat MSSA consistently improves VLM-based segmentation models andoutperforms existing fine-tuning-based TTA methods by a clear mar-gin, with gains of up to 12.2% DSC and 11.7% mIoU. Project Page:https://lingrayy.github.io/MSSA/.

Image restoration aims to reconstruct high-quality images from degraded low-quality inputs. As the computational demands of image restoration models continue to rise, there is growing interest in lightweight architectures optimized for fast and efficient inference. Logic gate networks (LGNs), which operate using fundamental logic operations such as NAND and XOR, have recently emerged as a promising direction for achieving highly efficient computation. However, their potential remains largely untapped in the domain of image restoration. In this work, we introduce LogicIR, the first LGN specifically designed for image restoration tasks. LogicIR incorporates a UNet-inspired architecture composed entirely of logic gates. In addition, we propose a differentiable bit decoding layer and an index shuffling mechanism that improves information propagation across logic gates. Experimental results across multiple image restoration benchmarks demonstrate that LogicIR achieves strong performance with significantly reduced computational cost, establishing LogicIR as a viable and efficient alternative for image restoration. The source code is available at our project page https://github.com/jimmy9704/LogicIR.

Unified multimodal models have recently shown remarkablegains in both capability and versatility, yet most leading systems are stilltrained from scratch and require substantial computational resources. Inthis paper, we show that competitive performance can be obtained farmore efficiently by strategically fusing publicly available models special-ized for either generation or understanding. Our key design is to re-tain the original blocks while additionally interleaving multimodal self-attention blocks throughout the networks. This double fusion mechanism(1) effectively enables rich multi-modal fusion while largely preservingthe original strengths of the base models, and (2) catalyzes synergisticfusion of high-level semantic representations from the understanding en-coder with low-level spatial signals from the generation encoder. By train-ing with only ∼35B tokens, this approach achieves strong results acrossmultiple benchmarks: 0.91 on GenEval for compositional text-to-imagegeneration, 82.16 on DPG-Bench for complex text-to-image generation,6.06 on GEditBench, and 3.77 on ImgEdit-Bench for image editing. Wewill fully release the entire suite of code, model weights, and datasets tosupport future research on unified multimodal modeling.

Neural rendering techniques allow for accurate reconstruc-tion of the geometry and color appearance of 3D scenes. Some methodshave extended their use to additional imaging modalities, such as multi-spectral, infrared, or polarimetric data. However, all of these approachesrequire expensive sensors and calibrated setups to capture new multi-modal frames for each new scene. We propose Spectral and PolarimetricImplicit Learned Representation (SPoILeR), a novel method to obtainmulti-view consistent renderings of unconventional modalities for sceneswhere either only RGB frames or very few of the additional modali-ties are available. Thanks to a multimodal pre-training phase, the modellearns the mutual correlation between different modalities. This step al-lows predicting accurate renderings of unconventional modalities duringa fine-tuning phase supervised only by RGB images. Experimental resultsshow that the approach can accurately render infrared, polarimetric, andmultispectral frames for scenes where no input sample captured by thesetypes of sensors is provided.

Generative vision-language models (VLMs) such as Qwen-VLand LLaVA achieve strong zero-shot performance on tasks overlappingwith their pretraining distribution, yet fail on specialized domains wherethe required discriminative features were never learned, a regime weterm distant out-of-distribution (OOD). Standard adaptation methodscannot overcome this representational absence because they operatewithin the encoder’s existing feature space. However, VLMs retain arobust descriptive capacity even when discrimination collapses: a modelthat cannot classify a medical scan can still articulate its visual patterns.Exploiting this asymmetry, we introduce Inductive Visual Logic (IVL), atraining-free framework that constructs classification knowledge from themodel’s surviving descriptive ability. IVL extracts visual traits from few-shot support images through dual-mode prompting, combining semanticdescriptions with primitive visual observations, and organizes them intoper-class trait dictionaries. At inference, hierarchical filtering identifiesspatially grounded trait evidence for classification. Across multiple distant-OOD benchmarks, IVL achieves the highest aggregate accuracy under twoVLM backbones while producing interpretable, trait-traceable predictions.

Self-collision remains a persistent challenge in SMPL-basedhuman pose estimation and motion generation. Under extreme articula-tions or stochastic motion synthesis, generated meshes frequently exhibitself-penetrations, leading to physically implausible results. We proposePoseShield, a neural collision constraint defined directly in SMPL posespace. We formulate collision correction as a constrained optimizationproblem and connect the learned constraint with the Eikonal equation.Enforcing Eikonal regularization ensures non-vanishing gradients nearthe collision boundary, improving numerical stability and robustness ofthe optimization process. Unlike prior methods that operate in the meshspace or rely on heuristic penalties, our approach operates directly in thelow-dimensional space of human poses and is theoretically grounded. Thesame learned constraint extends to human motion sequences, providinga generator-agnostic post-hoc collision corrector without retraining theunderlying motion model. Experiments on a newly constructed SMPLpose benchmark show that our method achieves a 95.8% success rateand outperforms state-of-the-art baselines.

Generating clean line art from conti, rough storyboard sketchesused in webtoon production, remains a labor-intensive process requiringsubstantial artistic expertise. Conti sketches provide only coarse geom-etry and sparse semantic cues, making it difficult to recover detailedline art while preserving consistent character identity across sequentialpanels. Existing sketch-to-image or diffusion-based translation modelsprocess images independently and therefore struggle to maintain cross-panel identity coherence. We present PanelDiff, a diffusion transformerframework for conti-to-line-art generation that explicitly models cross-panel character consistency. PanelDiff integrates three components. Amulti-reference character conditioning module encodes character labelsand multiple line-art exemplars into compact identity tokens for robustidentity guidance under ambiguous conti inputs. A memory-augmentedrepresentation bank accumulates character-aware features from previ-ously generated panels, while a similarity interpreter dynamically re-trieves relevant entries for the current panel. A panel-aware diffusiontransformer then jointly attends to conti structure, textual descriptions,identity tokens, and retrieved memory features to produce coherent line-art sequences. To support this task, we construct a conti–line-art datasetwith character identity annotations and sequential panel structures. Ex-periments show that PanelDiff improves line-art fidelity, identity preser-vation, and cross-panel consistency over strong baselines, and ablationstudies verify the contribution of each component. The code and datasetare available online (huggingface) ⋆ .

How can we teach large multimodal models (LMMs) newskills without erasing prior abilities? We study sequential fine-tuningon five target skills while monitoring general ability on eight held-outbenchmarks across three model families. Surprisingly, we find that per-formance lost on held-out tasks after fine-tuning on one skill can partlyrecover when the model is subsequently tuned on a different skill. Wetrace this behavior to a measurable shift in the output token distribu-tion, manifested through a simple counting-bias probe that shows theshift co-varies with forgetting. Guided by this insight, we identify twosimple, robust tuning recipes that learn strongly while limiting drift: (i)updating only the self-attention projection layers (SA Proj., ∆ learning+24.9 / ∆ held-out forgetting -0.6), and (ii) updating only the MLPGate&Up while freezing the Down projection (+30.5 / -2.1). Both sub-stantially outperform full-LLM tuning (+31.8 / -23.3) in the learning–forgetting trade-off. We also compare against common forgetting mitiga-tion methods—Learning without Forgetting (LwF), LoRA, Mixture-of-Experts, and weight-space interpolation (WiSE-FT)—and find that ourselective tuning recipes match or exceed their learning–stability balancewhile remaining simpler, requiring no replay, auxiliary parameters, orper-stage tuning. These results hold across LLaVA-OneVision, LLaVA-NeXT, and Qwen2.5-VL, confirming that the key to teaching LMMs newskills without forgetting lies in controlling output distribution shift bychoosing which components to tune. Code will be made available.

The advancement of Video Object Segmentation (VOS) isimpeded by its heavy reliance on expensive annotations. This has led toa landscape of compromised alternatives: unsupervised methods provideweak information, few-shot learners are plagued by referential ambiguityfrom distractors, and referring techniques are limited by a lack of visionpriors. We revisit Few-Shot Referring Video Object Segmentation (FSR-VOS) under a streamlined and practical formulation that segments tar-get objects using only query-level textual descriptions and a small set ofannotated support images. This dual conditioning enables coarse-to-finedisambiguation and precise boundary refinement with minimal annota-tion cost. To address the challenge of complex multi-source informationfusion, we propose the QST-SAM framework. Specifically, a SupportInstruction Compressor is introduced to refine and condense the sup-port information into compact representations. The Q-S-T Transformerfurther facilitates comprehensive integration of multi-modal supervisorycues from query, support, and text branches. The Multi-Priors Prompt-ing module provides the SAM2 prompt encoder with enriched guidancefor precise, context-aware segmentation. We further establish a new FS-RVOS benchmark derived from Ref-Youtube-VOS and Ref-DAVIS17, onwhich QST-SAM achieves state-of-the-art performance, validating its ro-bustness and effectiveness in segmentation.

Current Large Multimodal Models (LMMs) struggle withhigh-resolution visual inputs during the reasoning process, as the num-ber of image tokens increases quadratically with resolution, introducingsubstantial redundancy and irrelevant information. A common practiceis to identify key image regions and refer to their high-resolution counter-parts during reasoning, typically trained with external visual supervision.However, such visual supervision cues require costly grounding labelsfrom human annotators. Meanwhile, it remains an open question how toenhance a model’s grounding abilities to support reasoning without rely-ing on additional annotations. In this paper, we propose High-resolutionAnnotation-free Reasoning Technique (HART), a closed-loop frameworkthat enables LMMs to focus on and self-verify key regions of high-resolution visual inputs. HART incorporates a post-training paradigmin which we design Advantage Preference Group Relative Policy Opti-mization (AP-GRPO) to encourage accurate localization of key regionswithout external visual annotations. Notably, HART provides explain-able reasoning pathways and enables efficient optimization of localiza-tion. Extensive experiments on MME-RealWorld-Lite, TreeBench, V*Bench, HR-Bench-4K/8K, and MMStar demonstrate that HART im-proves performance across a wide range of high-resolution visual tasks,consistently outperforming strong baselines. Code will be available athttps://github.com/RL-MIND/HART.

Designing functional and aesthetically coherent floor plansrequires exploring a vast space of possible room arrangements, a taskthat quickly becomes overwhelming for human designers. In this pa-per, we propose GRE-Diff, a controllable and interactive diffusion-basedframework that automates the creation and editing of apartment floorplans under user-specified constraints. By combining AI-generated sug-gestions with real-time, human-in-the-loop editing, the system enablesusers to specify room types, room counts, boundary shapes, and editingoperations through LLM-parsed instructions or GUI-based interaction.It then generates a diverse set of plausible and well-structured designs forrefinement. At the core of our approach is Gaussian Room Embedding(GRE), a continuous latent representation that models each room as aspatial Gaussian distribution capturing its location and extent. Exten-sive experiments on the RPLAN dataset show that GRE-Diff produceshigh-quality, constraint-aware, and editable polygonal layouts, offeringa practical step toward bridging AI-driven automation and human cre-ativity in spatial design.

Reinforcement Learning from Verifiable Rewards (RLVR)has substantially enhanced the reasoning capabilities of large languagemodels in abstract reasoning tasks. However, its application to LargeVision-Language Models (LVLMs) remains constrained by a structuralrepresentational bottleneck. Existing approaches generally lack explicitmodeling and effective utilization of visual information, preventing vi-sual representations from being tightly coupled with the reinforcementlearning optimization process and thereby limiting further improvementsin multimodal reasoning performance. To address this limitation, wepropose KAWHI (Key-Region Aligned Weighted Harmonic Incentive),a plug-and-play reward reweighting mechanism that explicitly incorpo-rates structured visual information into uniform reward policy optimiza-tion methods (e.g., GRPO and GSPO). The method adaptively localizessemantically salient regions through hierarchical geometric aggregation,identifies vision-critical attention heads via structured attribution, andperforms paragraph-level credit reallocation to align spatial visual ev-idence with semantically decisive reasoning steps. Extensive empiricalevaluations on diverse reasoning benchmarks substantiate KAWHI as ageneral-purpose enhancement module, consistently improving the perfor-mance of various uniform reward optimization methods. Code is availableat https://github.com/kawhiiiileo/KAWHI.

Probabilistic point cloud registration has attracted increasing attention due to its robustness to noise, outliers and occlusions. However, existing methods often suffer from high computational cost and neglect the role of informative priors. In this paper, we propose a new probabilistic registration method based on graph signal processing (GSP), called graph coherent point drift (GraphCPD). Specifically, we design a high-pass graph filter to extract high-frequency components, which are theoretically proven to be invariant under rigid transformations. These components are combined with point coordinates and normals to form a high-dimensional graph signal. We construct a local graph based on the graph signal and use the graph Laplacian model for registration. Compared with the classical Gaussian mixture models (GMMs), graph Laplacian models provide more discriminative geometric representations and enhance the model’s ability to capture graph structure. Furthermore, we exploit the invariance of high-frequency components to define prior probabilities, significantly reducing the corresponding search space and improving the speed of registration. Experimental results demonstrate that our method improves runtime efficiency over most existing probabilistic methods, while maintaining competitive registration accuracy, especially on large-scale point clouds. The source code is available at https://github.com/YCLai1997/GraphCPD.

Blood oxygen saturation (SpO2 ) is a vital indicator of respi-ratory and cardiovascular health, increasingly targeted by camera-based,contactless monitoring systems. Existing video-based approaches typi-cally rely on predefined frequency filtering to separate pulsatile (AC)and baseline (DC) components. Such spectral decomposition propagatesband-limited noise and assumes signal stationarity, limiting robustnessunder motion, illumination variation, and subject diversity.We propose FUSE, a filter-free, physics-guided framework that learnsAC and DC representations via structured differential constraints ratherthan static spectral filtering. The pulsatile component is regularized as aband-limited harmonic field through a second-order ordinary differentialequation, while spatial coherence is enforced by modeling it as a trans-ported density field under an advection-based partial differential con-straint. This coupled wave-transport formulation enforces physiologicallyconsistent spatiotemporal dynamics without explicit frequency-domain

Partial label learning in the instance-dependent setting (ID-PLL) poses significant challenges, as candidate labels are highly corre-lated with sample features, and strong semantic overlap among candi-dates often leads to error-prone disambiguation. Meanwhile, the infor-mative dark knowledge inherent in non-candidate labels remains largelyunderexplored. To address these issues, we propose DKDP, a dual dis-tillation method that extracts and transfers more stable and reliableknowledge from the teacher model. Specifically, we first use constrainedprojection to structurally rectify the teacher’s prediction distribution,enforcing the prior that the minimum value of candidate labels exceedsthe maximum value of non-candidate labels. On this basis, we guide thestudent model to fit the corrected teacher outputs. We further derive theminimum modification required to satisfy the prior, which theoreticallyguarantees the interpretability of the rectification process. Moreover, weintroduce class prototypes to model the distribution of fine-grained se-mantic relations, extending reliable knowledge distillation from the out-put space to the embedding space. This yields a stable, separable geomet-ric structure in the feature space. Extensive experiments on benchmarkdatasets demonstrate the effectiveness of our proposed method.