ECCV 2026 论文列表

We present DeWorldSG, a novel framework that generatesspatio-temporally robust 3D Semantic Scene Graphs from RGB-D se-quences. Existing methods often struggle to construct reliable 3D scenegraphs due to unstable 3D object representations and missing relationscaused by frame-wise inference. DeWorldSG addresses these issues byestimating instance-level geometric 3D Gaussian distributions throughdepth-guided filtering and representing each object as a probabilistic 3Dnode rather than a single projected point. To mitigate relational sparsityfrom frame-wise inference, our framework further aggregates spatiotem-poral evidence across object pairs and refines relations using contextualpriors derived from a world model (V-JEPA 2). Experiments on the3DSSG and ReplicaSSG datasets demonstrate state-of-the-art (SoTA)performance in both object and predicate prediction, while producingtemporally consistent scene structures. In particular, our method im-proves triplet recall by 77.4% and predicate recall by 23.2% over priorSoTA approaches, making it suitable for robotic manipulation and ARapplications. Our code and models are open-sourced.

Multimodal large language models (MLLMs) have demon-strated strong performance in offline video understanding tasks, but mostremain constrained to offline inference or exhibit weak online reasoningability, rendering online multi-turn interaction over continuously arriv-ing video streams challenging. Existing streaming approaches adopt aninterleaved perception-generation paradigm, which precludes concurrentperception and generation and induces early memory decay with grow-ing streams, degrading long-range dependency modeling. We proposeThink While Watching, a memory-anchored streaming video reason-ing framework that maintains continuous segment-level memory duringmulti-turn interaction. We construct a three-stage, multi-round, chain-of-thought (CoT) dataset with a stage-matched training strategy while en-forcing strict causality in streaming reasoning via a segment-level stream-ing causal mask and streaming positional encoding. At inference, we de-sign an efficient pipeline that overlaps watching and thinking and adap-tively selects the optimal attention backend. We evaluate our methodunder single-round and multi-round streaming input protocols. Basedon Qwen3-VL, we improve single-round accuracy by 2.6% on Streaming-Bench and 3.79% on OVO-Bench. In the multi-round protocol, we main-tain performance while reducing output tokens by 56%. Code is avail-able at: https://github.com/wanglu-cs/Think_While_Watching.

Autoregressive video diffusion models have enabled the gener-ation of arbitrarily long videos by removing conditioning on future frames,thus greatly improving computational efficiency. Yet, they suffer from er-ror accumulation over time, as the denoised sequence gradually drifts awayfrom the conditioning distribution seen during training. Recent advancesattempt to reduce this error by anchoring each generated frame to thelearned manifold of real ones. However, even when all generated individualframes lie close to the real manifold, there are trajectories which the modellacks sufficient knowledge to continue without exiting it, thus reachinga terminal point. To prevent the model from being trapped in terminalpoints, we start from the hypothesis that for well-modeled future trajec-tories the distribution of the predicted noise should match the one of theforward noising process. To enforce such a prior at test time, we introduceTerminal points Avoidance through Noise Guided Optimization (TANGO),which uses the diffusion model as a critic of its own outputs, by predictingone step forward and requiring an isotropic Gaussian noise prediction. Weuse the deviation from this expected noise distribution to search for an al-ternative trajectory that does not lead to a terminal point. Our approachachieves a 3.1% absolute improvement on VBench over state-of-the-art,while reducing Fréchet Video Distance by 28.3% on average across 15 svideos. Our code is available on https://mever-team.github.io/tango.

Accurate landslide segmentation from high-resolution remote sensing imagery remains challenging due to weak spectral contrast and strong dependence on terrain structure. Unlike appearance-defined objects, landslides follow geomorphological patterns governed by slope continuity and elevation discontinuities. However, most existing segmentation methods treat landslides as generic semantic regions and incorporate Digital Elevation Models (DEM) through simple feature fusion without explicitly enforcing terrain-consistent spatial reasoning. To address this limitation, we reformulate landslide segmentation as a terrainconditioned structured prediction problem and propose TerrainGraphNet, a terrain-aware graph reasoning framework that embeds geomorphological constraints into representation learning and spatial propagation. The proposed method introduces terrain-modulated feature interaction, where elevation structure adaptively regulates visual representations, and terrain-aware graph construction, where spatial connectivity is defined jointly by feature similarity and slope continuity. This formulation can be interpreted as learning a terrain-weighted smoothness prior that encourages predictions to respect geomorphological coherence. Extensive experiments on three benchmark datasets demonstrate consistent improvements over strong CNN and transformer baselines. In addition to higher IoU and F1 scores, TerrainGraphNet significantly improves boundary accuracy and topological consistency, highlighting the importance of terrain-constrained reasoning for reliable landslide delineation.

Pretrained Vision Transformers (ViTs) such as DINOv2 andMAE provide generic image features that can be applied to a varietyof downstream tasks such as retrieval, classification, and segmentation.However, such representations tend to focus on the most salient visualcues in the image, with no way to direct them toward less prominentconcepts of interest. In contrast, Multimodal LLMs can be guided withtextual prompts, but the resulting representations tend to be language-centric and lose their effectiveness for generic visual tasks. To addressthis, we introduce Steerable Visual Representations, a new class of vi-sual representations, whose global and local features can be steered withnatural language. While most vision-language models (e.g., CLIP) fusetext with visual features after encoding (late fusion), we inject text di-rectly into the layers of the visual encoder (early fusion) via lightweightcross-attention. We introduce benchmarks for measuring representationalsteerability, and demonstrate that our steerable visual features can focuson any desired object in an image while preserving the underlying rep-resentation quality. Our method also matches or outperforms dedicatedapproaches on anomaly detection and personalized object discrimination,exhibiting zero-shot generalization to out-of-distribution tasks.Project Website: jonaruthardt.github.io/project/SteerViTPrompt CLS AttentionSteerable Visual RepresentationsGoal: control what vision features encode New Pareto FrontierEval: retrieving images w/ prompted objects Previous SoTAUse: task-specific adaptationFig. 2: SteerViT produces high-quality visual representations that can besteered by text. Left: Traditional (non-steerable) representations like DINOv2 tendto focus on the dominant object in an image and retrieve images with the same object.SteerViT can adapt to a text prompt, enabling retrieval of images even with smallobjects of interest. Right: We compare SteerViT to prior work in terms of its ability toadapt to text (measured by text-guided image retrieval (cf. Sec. 4.1)) and the qualityof the visual representation (measured by the accuracy of linear probing for the CLSfeature and semantic segmentation for patch features). While models typically tradeoff steerability for representation quality, SteerViT preserves both. By modulating agating factor (Eq. (2)), SteerViT achieves a new Pareto frontier.

Cross-attention is a crucial component in learning-based imageto-point-cloud (I2P) registration. Although existing cross-attention mechanisms have achieved promising progress, attention ambiguity remains a fundamental challenge that hinders the learning of discriminative 2D- 3D correspondences. To address this problem, we revisit cross-attention and establish ordinary differential equations (ODEs) to model the ideal I2P feature interaction. Based on this formulation, we develop an ODEdriven cross-attention (OCA) module that refines feature representations and attention matrices through ODEs. In practice, OCA can be seamlessly integrated into existing I2P registration frameworks. To validate its effectiveness, we incorporate OCA into five state-of-the-art baselines and evaluate on four public benchmark datasets. Experimental results demonstrate that OCA improves registration recall by up to 5%, 9%, and 15% under the standard, fine-tuning, and zero-shot settings, respectively. Code is released at github.com/anpei96/oca-i2p-demo.

Tokenizers are a key component of state-of-the-art generativeimage models, extracting the most important features from the signalwhile reducing data dimension and redundancy. Most current imagetokenizers are based on KL-regularized variational autoencoders (KL-VAE), trained with reconstruction, perceptual and adversarial losses.Diffusion decoders have been proposed as a more principled alternativeto model the distribution over images conditioned on the latent. However,matching the performance of KL-VAE still required adversarial losses,as well as a higher decoding time due to iterative sampling. To addressthese limitations, we introduce a new pixel diffusion decoder architecturefor improved scaling and training stability, benefiting from transformercomponents and GAN-free training. We use distillation to replicate theperformance of the diffusion decoder in an efficient single-step decoder.This makes SSDD the first GAN-free single-step diffusion decoder toreach state-of-the-art continuous tokenization, with higher reconstructionquality and faster sampling than KL-VAE. In particular, SSDD improvesreconstruction FID from 0.87 to 0.46 with 1.4× higher throughput andpreserves generation quality of DiTs with 3.8× faster sampling. As such,SSDD can be used as a drop-in replacement for KL-VAE, and for buildinghigher-quality and faster generative models.

Drone image–text retrieval (DITR) aims to associate aerialimages with textual descriptions in complex low-altitude environments.However, intricate spatial layouts and long-range structural dependen-cies in drone imagery challenge conventional Euclidean embedding mod-els, often leading to cross-modal misalignment. To address this issue, wepropose SARA, a structure-aware Riemannian-guided alignment frame-work that incorporates Riemannian manifold modeling to complementsemantic embeddings. Specifically, the Manifold Structural Feature Ex-traction (MSFE) module captures second-order structural dependenciesby encoding local feature correlations as symmetric positive definite rep-resentations. The Log-Euclidean Structural Alignment (LESA) moduleperforms geometry-consistent alignment in the tangent space, integratingsemantic correspondence with structural coherence through distributionalignment and feature fusion. We further provide theoretical analysisdemonstrating convergence and minimization of an upper bound on theexpected alignment error. Experiments on ERA and UDV benchmarksdemonstrate that SARA consistently improves retrieval performance overstate-of-the-art methods, particularly in structurally complex or seman-tically ambiguous drone scenarios.

Existing methods for adapting 2D foundation models such asSAM to 3D volumes either process slices independently—ignoring inter-slice context—or require substantial architectural changes and retraining.In this paper, we present SAM+D, a parameter-efficient framework thatlifts SAM-family models by one spatial dimension—enabling 3D volu-metric segmentation from 2D SAM and, for the first time via parameter-efficient fine-tuning, end-to-end 4D (3D+T) spatiotemporal segmenta-tion from video-based SAM2—while keeping the vast majority of pre-trained parameters frozen. SAM+D introduces two lightweight, model-agnostic modules into frozen transformer blocks: (1) Depth-RoutedLoRA (DRLoRA) experts with learned routing for spatially adaptivelow-rank updates, and (2) Depth Shift Modules (DSM) for cross-slice feature exchange at zero additional parameter cost. Together, theyprovide volume-level context while tuning only ∼2.8% of parameters forSAM and ∼3.7% for SAM2. We evaluate SAM+D in two distinct set-tings, each lifting the base model by one spatial dimension: 3D segmenta-tion, where SAM (2D → 3D) is evaluated on four CT benchmarks (KiTS,Pancreas, LiTS, Colon), and 4D segmentation, where SAM2 (2D+T →3D+T) is evaluated on a cell tracking challenge (CTC) dataset (Fluo-N3DH-SIM+). In both settings SAM+D achieves competitive or superiorresults under the single-point prompt setting while using fewer trainableparameters than existing methods, demonstrating that SAM+D general-izes across SAM-family architectures, target dimensionalities (3D, 4D),and domains spanning medical imaging and bio-scene understanding.Code is publicly available at https://github.com/JerrySongCST/SAM-Plus-D.

Micro-expressions are brief, involuntary facial movementsthat reveal subtle affective states. While most computational work fo-cuses on micro-expression recognition, which classifies clips into dis-crete emotion categories, Micro-Expression Action Unit (ME-AU) de-tection offers a more theory-consistent formulation grounded in the Fa-cial Action Coding System. Despite its finer granularity, ME-AU de-tection remains comparatively under-explored due to the limited avail-ability of high-quality data for efficient automatic models. Recent volu-metric micro-expression datasets provide temporally coherent 3D facialrecordings with reliable AU labels, creating new opportunities for fine-grained analysis. However, their adoption remains limited due to largedata volumes, heterogeneous formats, and the absence of standardizedbenchmarks, which together hinder reproducibility and cross-study com-parison. Moreover, because of the small spatio-temporal spanning of theAUs, designing models that effectively combine volumetric representa-tions with temporal dynamics remains a non-trivial research challenge.To address these barriers, we introduce an extended and standardizedrelease of the 4DME dataset with cleaned data and normalized format-ting, accompanied by a compressed version to reduce storage overhead.We further design VoluME, a family of dual-stream optical-flow-basedarchitecture for volumetric ME-AU detection. To determine appropriatesettings for compression, we combine state-of-the-art compression base-lines, perceptual evaluation, and downstream task-specific benchmark-ing. As the first work dealing with volumetric ME-AU detection undercompression, our results show that a compression ratio of approximately3000× can be achieved with ≈ 1% F1 degradation. The extended releaseadditionally introduces explicit two-way cultural balancing, supportingfuture cross-cultural micro-expression studies.

We address the problem of discovering repeated elementsfrom a single image. In contrast to existing approaches that depend onlarge annotated datasets, curated multi-image collections, or object seg-mentation masks, we show that a single image can suffice to learn ameaningful object model in a completely bottom-up fashion, without anyprior knowledge beyond a coarse scale prior. Our method learns a tunableimage-space prototype of the repeated elements through a reconstructionobjective, enabling the model to identify and synthesize consistent objectinstances within the same image. Experiments on 116 real images fromthe FSC-147 dataset demonstrate that our method successfully learnscoherent element models and captures intra-category variation on chal-lenging images. Qualitative results reveal superior reconstructions andinterpretable decompositions compared to classical decomposition, jointalignment, and 3D object modeling methods, while maintaining a simple2D formulation. These results suggest that meaningful object discoverycan emerge from single-image learning alone.

Distribution matching is central to many vision and graphicstasks, where the widely used Wasserstein distance is too costly to com-pute for high-dimensional distributions. The Sliced Wasserstein Distance(SWD) offers a scalable alternative, yet its Monte Carlo estimator suffersfrom high variance, resulting in noisy gradients and slow convergence. Weintroduce Reservoir SWD (ReSWD), which integrates Weighted Reser-voir Sampling into SWD to adaptively retain informative projectiondirections in optimization steps, resulting in stable gradients while re-maining unbiased. Experiments on synthetic benchmarks and real-worldtasks such as color correction and diffusion guidance show that ReSWDconsistently outperforms standard SWD and other variance reductionbaselines. Project page with code: https://ReservoirSWD.github.io

Ternary transformers offer extreme memory and compute ef-ficiency, but existing low-bit LoRA-based methods cannot directly fine-tune ternary weights. Current approaches either require dequantization,restoring low-bit base weights to higher precision to merge with adap-tation weight, or update only quantization parameters, preventing amerged model that remains ternary. We propose ternary multiplica-tive adaptation, which represents discrete updates of ternary weightssuch as sign flips or zeroing through a low-rank Kronecker factoriza-tion into two small ternary matrices applied element-wise to ternaryweights. This design is parameter-efficient and expressive, preserves theternary domain, and supports direct merging without dequantization.Experiments on six models across language and vision, including ternar-ized LLaMA-3 1B and 3B and a ternary ViT-B/16, demonstrate thatour method recovers much of the performance lost to quantization andoutperforms strong low-bit and ternary baselines. Code is available athttps://github.com/alexmanoo/ternary_adaptation.

Procedural material creation underpins applications in digitalcontent creation, visual effects, and 3D asset design. Achieving high-quality results requires more than reproducing node graphs—it demandsunderstanding the process by which experts construct materials. We for-mulate procedural material generation as retrieval-time process reasoningover expert demonstrations, elevating process to a first-class represen-tation beyond graph-only synthesis. Concretely, we represent expertworkflows as process traces: textual records of construction steps, param-eters, and design intent. To instantiate this idea, we use a pretrainedLLM-based ProcessSynthesizer to synthesize a process trace alignedwith a user’s intent and a pretrained LLM-based Compiler to groundthe process trace into an executable Blender material graph. Becauseprocedural expertise is most naturally conveyed through demonstrations,we leverage tutorial videos as a source of process knowledge and extracttextual, LLM-compatible traces using automated video analysis tools.In an expert study with five Blender artists (avg. 7.5 years of expe-rience), materials generated by reflecting expert demonstrations werefound to produce workflows requiring fewer edits, and more closely matchprofessional design strategies than methods operating solely on staticartifacts. A user study with 150 participants further shows that ourapproach achieves superior generation and editing performance comparedto prior procedural systems. All code, models, and data will be availableat https://materialapprentice.github.io.

Single-image reflection removal (SIRR) seeks to recover thetransmission layer from a mixture corrupted by reflections—a severelyill-posed problem. Existing methods operate in pixel space, where thenonlinear sRGB formation model entangles the two layers and limits gen-eralization. We observe that pretrained VAE latent spaces exhibit sub-stantially lower coherence between image layers compared to pixel space,providing a more favorable working space for decomposition. Buildingon this finding, we propose PRISM (Pretrained-latent Reflection ImageSeparation Model), which reinterprets SIRR as a latent linear separa-tion problem. Under an approximate additive formulation in latent space,PRISM learns a flow matching velocity field on a pretrained FLUX back-bone that recovers both transmission and reflection in a single forwardpass. To enforce robust disentanglement, we introduce a Latent Com-position Consistency (LCC) strategy that constructs synthetic mixturesby swapping reflection latents across samples and enforces consistent de-composition via a cycle loss. We further propose a Layer ContrastiveSeparation (LCS) loss that promotes semantic separation between layersthrough patch-level contrastive learning, without requiring explicit reflec-tion targets. Experiments on six benchmarks demonstrate that PRISMconsistently outperforms state-of-the-art methods by significant margins,with strong generalization to in-the-wild images.

Unified fashion generation integrates tasks like virtual try-onand garment reconstruction into a single model to reduce task-specificadaptation costs. However, naive parameter sharing across semanticallydistinct tasks induces negative transfer through severe inter-task gradientconflict. We propose OrthoTryOn, a unified framework mitigating thisinterference within a shared Low-Rank Adaptation (LoRA) module. ItsOrthogonal Subspace Projection (OSP) applies task-specific orthogonalrotations to bottleneck features, mapping them into decorrelated coordi-nate frames. To address residual semantic coupling at inference time, wefurther propose Fisher-guided Negative Guidance (FNG), a parameter-free strategy that utilizes diagonal Fisher information to quantify inter-task sensitivity overlap and explicitly repels generation trajectories fromthe most confusable task via Classifier-Free Guidance. Extensive exper-iments demonstrate that OrthoTryOn avoids the severe performancedegradation typical of naive unified training and even surpasses inde-pendently trained task-specific models, achieving state-of-the-art resultsacross multiple benchmarks while generalizing robustly across diversediffusion backbones. Code is available at https://github.com/NJU-PCALab/OrthoTryOn.

Previous feed-forward 4D reconstruction methods either pre-dict per-frame static point clouds, ignoring foreground motion, or esti-mate point cloud trajectories while being limited to small camera mo-tions. This limits their ability to aggregate observations over time andreconstruct complete dynamic scenes under large viewpoint changes. Toaddress this, we propose OmniX, a feed-forward 4D reconstruction frame-work that predicts dense 3D point trajectories for every pixel fromvideos with large camera motion. 1) OmniX separates dynamic mo-tion modeling from static geometry prediction and represents motionwith a small set of dynamic tokens. Leveraging the sparse and low-rankstructure of 3D motion, these tokens generate trajectory fields for allpixels in all images while efficiently preserving global interactions. 2) Tofacilitate training, we build an automatic UE5-based 4D data engine andintroduce a dataset of 80k scenes and 1.28M multi-view videos with fullgeometric annotations. OmniX achieves state-of-the-art results on dense3D point trajectory prediction and 3D point tracking, with competitiveperformance on video depth estimation and camera pose estimation.

Recent advancements in LVLMs necessitate robust bench-marks for complex, visually grounded reasoning. A critical limitation isidentified in many document understanding benchmarks: visual contentis often reducible to text, enabling high performance without genuine vi-sual grounding. To address this limitation, OmniMapBench is introducedto foster visual-centric reasoning for map documents. The benchmarkcomprises 2,096 manually annotated question-answer pairs across 1,603map documents from nine categories. It is designed to probe a hierar-chy of skills, ranging from perception to multi-step visual reasoning. Toquantify benchmark properties, a simple yet effective benchmark-levelmetric is proposed: the Visual Dependency Index (VDI), defined as theaccuracy drop when images are replaced with question-agnostic descrip-tions. OmniMapBench exhibits higher VDI than established benchmarks,which quantitatively validates its focus on irreducible visual reasoning.Comprehensive evaluations of 25 leading LVLMs are conducted on Om-niMapBench. A significant performance gap is observed, with the top-performing model achieving only 75.03% accuracy. This result under-scores the challenges posed by OmniMapBench to current LVLMs. Thiswork aims to catalyze progress in visual-centric reasoning for documentunderstanding of LVLMs. The dataset and code are publicly available athttps://github.com/SIGMME/OmniMapBench.

Data augmentation is fundamental to training modern deepvision and multimodal models. While individual methods, such as Ran-dAug, CutMix, Mixup, RandErase, and DropPath, offer strong regular-ization effects, their combined use has saturated in performance due tooverlapping functionalities, and aggressive pixel-level manipulations maydisrupt delicate cross-modal alignment. This saturation motivates thesearch for a new augmentation axis within the embedding space ratherthan the input space. We introduce Aether, a simple plug-in method thatapplies diffusion-style random perturbations in the embedding space viacontrolled alpha-mixing, specifically designed to provide isotropic regu-larization that remains semantically consistent. Inspired by feature-spaceperturbations in language models and image degradation in generativepretraining, Aether induces mild yet effective perturbations that smooththe representations without compromising the fine-grained structuralinformation required for strong vision-language encoders. Across diversearchitectures and across multiple recognition tasks, Aether delivers consis-tent gains over the advanced recipe combining CutMix, Mixup, DropPath,and RandAug—a level of improvement rarely observed with modern aug-mentation alternatives. Notably, Aether demonstrates superior effective-ness in multi-modal alignment, succeeding where traditional pixel-spaceaugmentations fail by providing a stable, isotropic regularization signalthat respects the integrity of the high-dimensional feature space. Code isavailable at https://github.com/naver-ai/aether.