ECCV 2026 Papers

Neural Radiance Fields achieve high-fidelity scene representation but suffer from costly training and rendering, while 3D Gaussian splatting offers real-time performance with strong empirical results. Recently, solutions that harness the best of both worlds by using Gaussians as proxies to guide neural field evaluations, still suffer from significant computational inefficiencies. They typically rely on stochastic volumetric sampling to aggregate features, which severely limits rendering performance. To address this issue, a novel framework named IRIS (Intersection-aware Ray-based Implicit Editable Scenes) is introduced as a method designed for efficient and interactive scene editing. To overcome the limitations of standard ray marching, an analytical sampling strategy is employed that precisely identifies interaction points between rays and scene primitives, effectively eliminating empty space processing. Furthermore, to address the computational bottleneck of spatial neighbor lookups, a continuous feature aggregation mechanism is introduced that operates directly along the ray. By interpolating latent attributes from sorted intersections, costly 3D searches are bypassed, ensuring geometric consistency, enabling high-fidelity, real-time rendering, and flexible shape editing.

3D Gaussian Splatting (3DGS) has been widely used in 3D reconstruction and 3D generation. However, the rapid adoption of 3D Gaussian Splatting raises growing concerns about information leakage and unauthorized use, urging the exploration of effective watermarking techniques. However, existing methods are limited by low capacity, fragility under geometric perturbations, and the infeasible requirement for costly fine-tuning or pipeline modifications, motivating the need for a generalizable, feed-forward framework capable of robust multi-modal embedding with minimal intrusion. In this paper, we propose a new framework X-SG2S which can simultaneously inject 1D to 3D watermarks for copyright protection, while keeping the high fidelity of original 3DGS scenes. Specifically, we first split the watermarks into message patches. A self-adaptive gate is developed to select the injection positions of the watermark messages. Then, we use an XD (multi-dimensional) injection head to inject multi-modal messages into sorted 3DGS points. To restore watermarking messages, a learnable gate is developed to recognize the watermarked locations, from which our XD-extraction heads are used to restore hidden messages. X-SG2S is the first framework to unify 1Dto-3D watermarking and enable simultaneous multi-modal watermark embedding in 3DGS, achieving this with minimal rendering interference and zero modifications to parameters or pipelines. Extensive experiments demonstrate that X-SG2S effectively preserves consistency between the watermark and the original 3DGS, exhibits robustness against model degradation, and maintains accurate judgment capabilities.

Joint video super-resolution and deblurring (VSRDB) requires both efficient long-range temporal modeling and robustness to frame-wise exposure-duration variation, which changes the extent of motion blur across video frames. We propose FMA-Net++, a non-recurrent, sequence-level framework built from Hierarchical Refinement with Bidirectional Aggregation (HRBA) blocks. By stacking HRBA blocks, FMANet++ processes video frames in parallel while hierarchically expanding the temporal receptive field, avoiding the limited temporal receptive field of sliding-window designs and the sequential bottleneck of recurrent ones. To handle exposure-duration-dependent blur, we introduce an Exposure Time-aware Modulation (ETM) layer that conditions HRBA features on exposure embeddings from an Exposure Time-aware Feature Extractor (ETE). The conditioned features guide an exposure-aware flow-guided dynamic filtering module to predict motionand exposureaware degradation kernels. FMA-Net++ decouples degradation learning from restoration: the former predicts degradation priors and the latter exploits them for efficient high-resolution restoration. To evaluate VSRDB under controlled exposure-duration variation, we introduce the REDS-ME (multi-exposure) and REDS-RE (random-exposure) benchmarks. Trained solely on synthetic data, FMA-Net++ achieves stateof-the-art accuracy and temporal consistency on these benchmarks. It further shows strong out-of-distribution performance on GoPro and challenging real-world videos, while outperforming recent methods in both restoration quality and inference speed.

One-shot Federated Learning (OFL) reduces communication costs by requiring only a single round of model exchange. However, when clients train independently on heterogeneous data, their learned representations develop different means and variances, creating distributional mismatch that prevents effective server-side aggregation. Existing methods attempt post-calibration at the server, but struggle to reconcile representations that have already diverged into incompatible statistical spaces. We propose FACE (Federated Aligned-Consistent Ensemble) to prevent distributional divergence during client training rather than correcting it afterward. FACE uses a shared fixed reference matrix to guide representation means toward consistent directions, applies a variance stabilizer to maintain consistent feature variance, and performs featurefusion ensemble inference. Extensive experiments on CIFAR-10/100 and Tiny-ImageNet demonstrate that FACE achieves state-of-the-art performance, with particularly strong improvements under extreme data heterogeneity. Comprehensive ablation studies validate the necessity of both components for achieving distributional alignment.

Recently, Multimodal Large Language Models (MLLMs) have been widely integrated into diffusion frameworks primarily as text encoders to tackle complex tasks such as spatial reasoning. However, this paradigm suffers from two critical limitations: (i) MLLMs text encoder exhibit insufficient reasoning depth. Single-step encoding fails to activate the Chain-of-Thought process, which is essential for MLLMs to provide accurate guidance for complex tasks. (ii) The guidance remains invariant during the decoding process. Invariant guidance during decoding prevents DiT from progressively decomposing complex instructions into actionable denoising steps, even with correct MLLM encodings. To this end, we propose Endogenous Chain-of-Thought (EndoCoT), a novel framework that facilitates structured visual reasoning in MLLMs by iteratively refining latent thought states through an iterative thought guidance module, and then bridges these states to the DiT’s denoising process. Second, a terminal thought grounding module is applied to ensure the reasoning trajectory remains grounded in textual supervision by aligning the final state with ground-truth answers. With these two components, the MLLM text encoder delivers meticulously reasoned guidance, enabling the DiT to execute it progressively and ultimately solve complex tasks in a step-by-step manner. Extensive evaluations across diverse benchmarks (e.g., Maze, TSP, VSP, and Sudoku) achieve an average accuracy of 92.1%, outperforming the strongest baseline by 8.3 percentage points.

Video shadow detection (VSD) aims to robustly identify shadow regions in videos and serves as a key task for improving scene understanding. Although substantial progress has been made in recent years, existing methods still suffer notable performance degradation in challenging scenarios involving fast motion and drastic illumination changes. To address these limitations, we introduce event cameras into the VSD task and propose an Event-guided Diffusion Model (EDM) tailored for complex dynamic environments. Based on the high temporal resolution and asynchronous sensing characteristics of event cameras, we design a Motion Trajectory Attention (MTA) module and a Bidirectional Mask Generation (BMG) module to capture event-driven temporal cues. These cues are further integrated through a Dual-Modal Temporal Guidance (DMTG) module and injected into the diffusion model as conditional guidance to steer the denoising process. By incorporating event streams as a complementary modality, EDM effectively captures fast scene dynamics and abrupt lighting variations, leading to more robust and accurate video shadow detection. Furthermore, to address the lack of event modalities in existing VSD benchmarks, we construct a multimodal extension of the ViSha dataset, termed Event-ViSha, and collect a new real-world multimodal dataset, the Event-based Video Shadow Dataset (EVSD). Extensive experiments demonstrate that our method consistently outperforms state-of-the-art approaches across multiple benchmarks.

Multimodal document retrieval aims to retrieve relevant pages while preserving both textual and visual content from the original document. However, existing benchmarks primarily evaluate simple lexical or semantic matching, and most methods encode pages independently. Consequently, they overlook the contextual information in the document required to resolve queries that aggregate information across multiple pages. In this paper, we introduce CMDR and CMDR-Bench, a new multimodal document retrieval task and benchmark that require modeling document context. To address this challenge, we propose CMDR-Embed, a contextual multimodal embedding framework that explicitly incorporates document context by jointly encoding multiple pages and deriving page-level embeddings from a shared contextual representation. Furthermore, we introduce CMCL, a contextual multimodal contrastive learning objective that effectively trains CMDR-Embed and balances contextual modeling with page-level discriminability. Experiments demonstrate that CMDR-Embed significantly outperforms non-contextual embeddings, highlighting the importance of context-aware multimodal embeddings for advancing document retrieval1.

Lifting perspective images and videos to 360◦ panoramas enables immersive 3D world generation. Existing approaches often rely on explicit geometric alignment between the perspective and the equirectangular projection (ERP) space. Yet, this requires known camera metadata, obscuring the application to in-the-wild data where such calibration is typically absent or noisy. We propose 360Anything, a geometry-free framework built upon pre-trained diffusion transformers. By treating the perspective input and the panorama target simply as token sequences, 360Anything learns the perspective-to-equirectangular mapping in a purely data-driven way, eliminating the need for camera information. Our approach achieves state-of-the-art performance on both image and video perspective-to-360◦ generation, outperforming prior works that use ground-truth camera information. We also trace the root cause of the seam artifacts at ERP boundaries to zero-padding in the VAE encoder, and introduce Circular Latent Encoding to facilitate seamless generation. Finally, we show competitive results in zero-shot camera FoV and orientation estimation benchmarks, demonstrating 360Anything’s deep geometric understanding and broader utility in computer vision tasks. Additional results are available at https://360anything.github.io.

Generating temporally coherent 4D content from monocularvideo is a challenging yet practically important problem. Recent advancesin layout-aware 3D generation have enabled producing objects with theirspatial arrangements from a single image, naturally providing scene-levelunderstanding that is valuable for 4D generation. However, applyingsuch single-image models independently to each video frame yields severetemporal inconsistencies—flickering textures, structural discontinuities,and jittering poses. We present Tempo-SAM3D, a training-free frame-work that extends layout-aware 3D generation to coherent 4D generationfrom monocular video. Our approach introduces Dual-Path TemporalAttention, which maintains importance-driven KV caches in both cross-attention and self-attention to propagate visual memory and structuralmemory across frames, respectively. We further incorporate gradient-guided velocity correction that constrains the flow matching samplingtrajectory, and HexPlane-based temporal smoothing that regularizes la-tent representations via low-rank spatiotemporal decomposition. Tempo-SAM3D requires no model retraining and inherits layout awareness forspatially-grounded 4D generation. Experiments demonstrate clear im-provements over per-frame baselines and existing video-to-4D methods.

Video super-resolution (VSR) using large-scale Diffusion Transformer (DiT) priors achieves exceptional perceptual quality but is often impractical due to the quadratic computational cost of processing dense spatio-temporal token sequences. Existing efficiency-oriented methods risk irreversible detail loss and temporal flickering, a vulnerability especially pronounced in one-step diffusion models. To address this, we propose TRaM-VSR, a Token Routing and Merging framework for adaptive token allocation, leveraging both context-aware video priors and network-level priors. First, token importance is estimated by fusing motion-sensitive temporal cues with semantic text similarity, isolating dynamic objects and structural boundaries. Next, this importance is further calibrated and adjusted by an offline planner to guide routing across optimally grouped network blocks. Technically, within each routed group, structurally critical tokens are processed in a highfidelity local stream, while less informative tokens are aggregated into a compact global stream, both modulated by network depth and aligned with the multigranular nature of diffusion models. Extensive experiments show that TRaM-VSR accelerates inference significantly while preserving state-of-the-art reconstruction quality and robust temporal consistency. The code is available at https://github.com/Ree1s/TRaM-VSR.

Dense 3D sensors in various real-world fields produce point clouds that are geometrically redundant for real-time processing. In this paper, we propose an efficient and scalable learning-based anisotropic surface approximation framework, HD-PEA, that operates directly on unstructured point clouds, integrating anisotropic optimization into reconstruction to produce compact, geometry-aligned surface representations with higher fidelity, fewer elements, and improved numerical stability compared to isotropic and adaptive meshes. Firstly, we develop a novel learning-based high-dimensional (high-d) Euclidean point embedding method to map the input point clouds into a high-d manifold embedding space. For handling large-scale point clouds without retraining and fine-tuning, a patch-based meta-embedding scheme is designed during the inference stage. Then, we develop a new tangent subspace estimation for the high-d embedding manifold approximation and anisotropic manifold reconstruction in high-d space. The main contribution of this work is to propose a scalable deep learning framework and a variety of datasets for constructing a high-d Euclidean point embedding space aimed to 3D anisotropic surface mesh approximation and Riemannian curvature tensor estimation from point clouds. We extensively evaluate our method against state-of-the-art surface reconstruction approaches using several datasets, such as Thingi10K dataset, AIM@SHAPE and Stanford 3D Scanning Repository, ScanNet dataset, and further demonstrate its generalization and usability on diverse unseen shapes and applications from these datasets.

Event vision exploration holds significant achievements in various vision tasks, such as object recognition, action recognition, etc. However, improving the model’s robustness against overfitting remains a critical challenge. To address this, we propose augmenting event data from the noise rather than applying fixed data transformations. Specifically, we first decouple the quadruplet event data across spatial, temporal, and polarity dimensions to construct a representation, breaking existing frame-tied manners. Then, an event diffusion decoder is integrated with a condition extractor to gradually optimize the noise to the event data (N2E). The condition extractor aims to yield guidance vectors from the original data, which are fed into each optimization step of the diffusion process. During data augmentation, we can directly generate new event streams by adjusting the noise, diffusion parameters (e.g., number of steps or noise intensity), or conditioning vectors, thereby boosting the performance of downstream event vision models. Extensive experiments on two downstream tasks among five event vision models demonstrate that N2E achieves greater improvements than existing augmentation baselines. Code is available at https://github.com/rfww/n2e.

Transformer-based models have made remarkable progress in image restoration (IR) tasks. However, the quadratic complexity of self-attention in Transformer hinders its applicability to high-resolution images. Existing methods mitigate this issue with sparse or windowbased attention, yet inherently limit global context modeling. Linear attention, a variant of softmax attention, demonstrates promise in global context modeling while maintaining linear complexity, offering a potential solution to the above challenge. Despite its efficiency benefits, vanilla linear attention suffers from a significant performance drop in IR, largely due to the low-rank nature of its attention map. To counter this, we propose Rank Enhanced Linear Attention (RELA), a simple yet effective method that enriches feature representations by integrating a lightweight depthwise convolution. Building upon RELA, we propose an efficient and effective Vision Transformer, named LAformer. LAformer eliminates hardware-inefficient operations such as softmax and window shifting, enabling efficient processing of high-resolution images. Extensive experiments across 7 IR tasks and 21 benchmarks demonstrate that LAformer outperforms SOTA methods and offers significant computational advantages. Furthermore, we extend LAformer to diffusion-based and flow-based visual generation, showcasing its strong potential as a competitive alternative to DiT and SiT. Code and models are available at https://github.com/shallowdream204/LAformer.

Critical mineral discovery is a positive-only problem: deposits are observed as sparse locations, while unlabeled regions are not reliable negatives, and similar geophysical signatures can arise from different subsurface states. We therefore model mineral targeting as learning a conditional spatial distribution over occurrence locations, π(p | d), given geo-images d, rather than predicting a deterministic per-pixel score map. We introduce GeoCFM, a conditional flow-matching model that generates mineral occurrence point sets conditioned on multi-channel geo-images; GeoCFM learns a point-wise transport field in R2, using UNet features with point-conditioned velocity prediction to bridge dense rasters and sparse supervision without pseudo-negatives. On a synthetic magnetics–geochemistry benchmark with latent activation and on USGS Earth MRI data with a spatially disjoint tile split, GeoCFM improves geometric agreement with observed occurrences over score-map and nonconditional baselines, while representing epistemic uncertainty through conditional sampling.

Existing person search methods assume access to complete visual queries or exhaustive tracking, yet real-world witness accounts are vague, partial, and spread across cameras and time. We introduce ARGOS (Agentic Retrieval with Grounded Observational Search), a benchmark and agent framework that recasts multi-camera person search from one-shot retrieval on a complete query into interactive reasoning from partial clues. To our knowledge, ARGOS is the first interactive benchmark to couple witness dialogue with camera-network topology, requiring an agent to plan, question, and eliminate under information asymmetry. An ARGOS agent receives a vague witness statement and must decide what to ask, when to invoke spatial or temporal tools, and how to interpret ambiguous natural-language responses, all within a limited turn budget. To ground reasoning in physical constraints, the agent accesses a Spatio-Temporal Topology Graph (STTG) encoding camera connectivity and empirically validated transition times. The benchmark comprises 2,691 tasks across 14 real-world scenarios in three progressive tracks: semantic perception (Who, 989 tasks), spatial reasoning (Where, 550 tasks), and temporal reasoning (When, 1,152 tasks). We propose TurnWeighted Success (TWS) as the primary metric, jointly measuring correctness and turn efficiency. Experiments with four LLM backbones show the benchmark is far from solved: the best agent achieves TWS of 0.383 (Track 2) and 0.590 (Track 3). Ablations confirm each component is essential: removing domain-specific tools drops Top-1 accuracy by up to 49.6 percentage points, and removing strategic reasoning halves TWS while barely affecting Top-1.

Most privacy evaluations in computer vision treat datasets and models as isolated, static disclosures, failing to account for how risk accumulates in practice. We formalize Compositional Non-Face ReIdentification Risk, where identities remain linkable through non-face cues, such as clothing, gait, and scene context that aggregate across sequential releases of data expansions, model checkpoints, and metadata. We introduce the Vision Re-Identification Pressure Index (vRPI ), a foundational, information-theoretic measure derived from α Arimoto conditional Rényi entropy that quantifies the systemic erosion of anonymity. We prove that vRPI is monotone under cu- α mulative releases, establish an exact bridge to Bayes-optimal guessing probability, and derive additive decompositions under conditionalindependence benchmarks. Through the NFLeak protocol, we provide illustrative experiments on Market-1501-Tau and a CUHK03- Tau closed-world replication using a structured Attacker Ladder, together with masking-level, dependence-overlap, temperature-calibration, and non-uniform-prior diagnostics, demonstrating that vRPI accurately α tracks realized re-identification success even after explicit face removal. Our framework enables principled auditing of the “leaky bucket” effect in vision benchmarks, shifting the focus from single-release compliance to ecosystem-level privacy hygiene.

We address the challenge of novel view synthesis from only two input images under large viewpoint changes. Existing regression-based methods lack the capacity to reconstruct unseen regions, while camera-guided diffusion models often deviate from intended trajectories due to noisy point cloud projections or insufficient conditioning from camera poses. To address these issues, we propose ConfCtrl, a confidence-aware video interpolation framework that enables diffusion models to follow prescribed camera poses while completing unseen regions. ConfCtrl initializes the diffusion process by combining a confidenceweighted projected point cloud latent with noise as the conditioning input. It then applies a Kalman-inspired predict–update mechanism, treating the projected point cloud as a noisy measurement and using learned residual corrections to balance pose-driven predictions with noisy geometric observations. This allows the model to rely on reliable projections while down-weighting uncertain regions, yielding stable, geometry-aware generation. Experiments on multiple datasets show that ConfCtrl produces geometrically consistent and visually plausible novel views, effectively reconstructing occluded regions under large viewpoint changes.

The rapid rise of generative image synthesis has intensified interest in the graphics and vision community in producing realistic, controllable illumination. We present a generative method for screenspace global illumination (GI) that produces geometrically aligned, temporally stable indirect lighting using a single-step latent diffusion model. Unlike iterative and video diffusion methods that are costly and hard to control, our model conditions a one-step image generator on sceneintrinsic signals and sparse lighting hints for the current frame, encoded via adapter modules. To address flicker without the overhead of video diffusion, we introduce a temporal VAE decoder that plugs into a latent diffusion pipeline at inference time, delivering long-range temporal coherence while preserving spatial fidelity. On test data, our approach synthesizes high-quality indirect illumination well-aligned to inputs and improves temporal stability. Our quality evaluation demonstrates that our method generates visually and quantitatively plausible illumination in generative forward rendering, outperforming state-of-the-art methods.

Autoregressive (AR) diffusion offers a promising framework for generating videos of theoretically infinite length. However, a major challenge is maintaining temporal continuity while preventing the progressive quality degradation caused by error accumulation. To ensure continuity, existing methods typically condition on highly denoised contexts; yet, this practice propagates prediction errors with high certainty, thereby exacerbating degradation. In this paper, we argue that a highly clean context is unnecessary. Drawing inspiration from bidirectional diffusion models, which denoise frames at a shared noise level while maintaining coherence, we propose that conditioning on context at the same noise level as the current block provides sufficient signal for temporal consistency while effectively mitigating error propagation. Building on this insight, we propose HiAR, a hierarchical denoising framework that reverses the conventional generation order: instead of completing each block sequentially, it performs causal generation across all blocks at every denoising step, so that each block is always conditioned on context at the same noise level. This hierarchy naturally admits pipelined parallel inference, yielding a ∼1.8× wall-clock speedup in our 4-step setting. We further observe that self-rollout distillation under this paradigm amplifies a low-motion shortcut inherent to the mode-seeking reverse-KL objective. To counteract this, we introduce a forward-KL regulariser in bidirectional-attention mode, which preserves motion diversity for causal inference without interfering with the distillation loss. On VBench (20s generation), HiAR achieves the best overall score and the lowest temporal drift among all compared methods.