CVPR 2026 Papers

Trajectory prediction is critical for autonomous driving, enabling safe and efficient planning in dense, dynamic traffic. Most existing methods optimize prediction accuracy under fixed-length observations. However, real-world driving often yields variable-length, incomplete observations, posing a challenge to these methods. A common strategy is to directly map features from incomplete observations to those from complete ones. This one-shot mapping, however, struggles to learn accurate representations for short trajectories due to significant information gaps. To address this issue, we propose a Progressive Retrospective Framework (PRF), which gradually aligns features from incomplete observations with those from complete ones via a cascade of retrospective units. Each unit consists of a Retrospective Distillation Module (RDM) and a Retrospective Prediction Module (RPM), where RDM distills features and RPM recovers previous timesteps using the distilled features. Moreover, we propose a Rolling-Start Training Strategy (RSTS) that enhances data efficiency during PRF training. PRF is plug-and-play with existing methods. Extensive experiments on datasets Argoverse 2 and Argoverse 1 demonstrate the effectiveness of PRF. Code is available at https://github.com/zhouhao94/PRF.

Multi-image diffusion models can generate images like multi-views or videos to describe static or dynamic scenes, yet texture and structure drift persist, severely undermining the spatiotemporal consistency. Addressing this issue remains challenging, especially without any external geometric or semantic priors during the pure generative inference. In this paper, we introduce CorrAdapter, a plug-and-play adapter that discovers and exploits an innate property of the multi-image diffusion itself, aligning all output images before they are in fact generated. Specifically, CorrAdapter designs a bypass branch for transformer blocks in the multi-image diffusion model, encompassing a native correspondence constructor that builds reliable correspondences from the diffusion model's intermediate features, and an aligned area aggregator that integrates messages from only matching regions to avoid ambiguous information interactions. Given the native correspondences as guidance, CorrAdapter can enhance spatiotemporal consistency without any auxiliary inputs, and remains training-free and baseline-agnostic, which enables it to generalize seamlessly to various generation tasks. Additionally, we provide an optional training scheme to explore further-improved possibilities. Experiments on both static multi-view generation and dynamic video generation show that CorrAdapter consistently improves spatiotemporal consistency and perceptual quality over strong baselines, offering a simple yet versatile drop-in approach to geometrically faithful multi-image diffusion. Code is available at https://github.com/SuhZhang/CorrAdapter.

Procedural sequence generation aims to create intermediate images through multi-step processes, which is applied in industrial design, educational tutorial, and book illustration. However, existing methods often focus on a specific domain or initialize several expert networks for different domains, which face three challenges. First, the poor generalization to unseen domains. Second, the parameter redundancy due to multiple expert networks.Third, the difficulty in adaptively determining the number of generation steps for different processes.To address these challenges, we propose ProcessMaker, a novel framework that harnesses the inherent generalization capabilities in Diffusion Transformers (DiTs) for procedural sequence generation. Concretely, we introduce three key innovations: (1) Self-supervised Representation Alignment to explore the generalized ability for unseen processes. (2) Sparse Masks for different domains without additional expert networks. (3) A sliding window strategy, which dynamically accommodates the generation steps based on the process complexity. Extensive experiments validate that our ProcessMaker achieves procedural sequence generation with generalization ability and adaptive steps, while using only 7.3% trainable parameters compared with the state-of-the-art method.

YOLO series lead object detection with superior accuracy and speed. However, both convolutional and self-attention based architectures suffer from parameter redundancy and insufficient computational efficiency. Existing lightweight methods excessively pursue speed while ignoring the loss of important information during feature extraction and spatial transformation across different stages. Thus, effective lightweighting is crucial for detection performance. We propose YOLO-ULM, an ultra-lightweight real-time detector that achieves accelerated inference while preserving high accuracy. We innovatively design a variety of dual efficiency- and accuracy-driven modules, including efficient feature aggregation and parallel downsampling modules, as well as a more focused complete-IoU loss function. To validate our approach, we train it from scratch on COCO dataset without pretrained weights. By refining backbone parameters, we extend it to YOLO-ULM-Turbo for accelerated inference. YOLO-ULM surpasses state-of-the-art real-time detectors like YOLOv11/YOLOv12/YOLOv13 and RT-DETR. On a T4 GPU, YOLO-ULM-N achieves 41.6% mAP with an inference latency of 1.52 ms, outperforming YOLOv11-N (2.2%\uparrow) and YOLOv12-N (1.0%\uparrow). YOLO-ULM-S exceeds RT-DETR-R18 by 1.6% mAP with 64.7% fewer FLOPs and 63% fewer parameters. YOLO-ULM-L / X surpass YOLOv13-L / X by 0.7% and 0.8% respectively in mAP. YOLO-ULM-Turbo matches YOLOv12-Turbo's performance but uses less computation, with Turbo-N variant achieving 0.3% higher mAP and 16% fewer parameters than YOLOv12-Turbo-N.

Multi-stage generative models have shown great promise in 3D content creation due to focused generation of structure or texture in different stages, but their outputs often fail to align with human preferences. The key bottleneck to apply alignment methods is the presence of non-differentiable operations between generative stages.This disconnection stops preference signals applied to the final output from being backpropagated to the crucial, early stages of generation, while simple separated stage-wise alignment leads to texture-geometry inconsistency.To address this challenge, we introduce Circular-DPO, which builds a preference feedback loop to align multi-stage 3D generation models to human preference.Our method first applies Direct Preference Optimization (DPO) to refine the final 3D asset.We then construct new preference pairs by sampling and decoding the assets generated by the optimized model.These newly-formed pairs are used to train the preceding generative stage, effectively creating a feedback loop that bridges the non-differentiable gap. Furthermore, to enhance robustness against noisy data, we introduce a quality-aware weighting mechanism that prioritizes reliable preference pairs during training. Experiments demonstrate that our approach improves the alignment of generated 3D content with human preferences by enabling holistic, multi-stage optimization.

Recent feed-forward Gaussian reconstruction models adopt a pixel-aligned formulation that maps each 2D pixel to a 3D Gaussian, entangling Gaussian representations tightly with the input images. In this paper, we propose AnchorSplat, a novel feed-forward 3DGS framework for scene-level reconstruction that represents the scene directly in 3D space. AnchorSplat introduces an anchor-aligned Gaussian representation guided by 3D geometric priors (e.g., sparse point clouds, voxels, or RGB-D point clouds), enabling a more geometry-aware renderable 3D Gaussians that is independent of image resolution and number of views. This design substantially reduces the number of required Gaussians, improving computational efficiency while enhancing reconstruction fidelity. Beyond the anchor-aligned design, we utilize a Gaussian Refiner to adjust the intermediate Gaussians via merely a few forward passes. Experiments on the ScanNet++ v2 NVS benchmark demonstrate the SOTA performance, outperforming previous methods with more view-consistent and substantially fewer Gaussian primitives.

Hand-object interaction (HOI) involves dynamics where human manipulations produce spatio-temporal effects on objects. However, existing semantic HOI benchmarks focus on either manipulation or effects at a coarse level, lacking fine-grained spatio-temporal reasoning to capture HOI dynamics. We introduce HanDyVQA, a fine-grained video QA benchmark that comprehensively covers both the manipulation and effect aspects of HOI. HanDyVQA comprises six complementary question types (Action, Process, Objects, Location, State Change, and Object Parts), totaling 11.1K multiple-choice QA pairs. Collected QA pairs require recognizing manipulation styles, hand/object motions, and part-level state changes. HanDyVQA also includes 10.3K segmentation masks for Objects and Object Parts, enabling the evaluation of object/part-level reasoning in video object segmentation. We evaluated recent video foundation models on our benchmark and found that even the best, Gemini-2.5-Pro, achieved only 73% accuracy, well below human performance (97%). Further analysis shows the remaining challenges in spatial relationships, motion, and part-level geometric understanding. We also found that incorporating explicit HOI cues into visual features improves performance, providing insights for future HOI-aware models.

Reconstructing and semantically interpreting 3D scenes from sparse 2D views remains a fundamental challenge in computer vision. Conventional methods often decouple semantic understanding from reconstruction or necessitate costly per-scene optimization, thereby restricting their scalability and generalizability. In this paper, we introduce Uni3R, a novel feed-forward framework that jointly reconstructs a unified 3D scene representation enriched with open-vocabulary semantics, directly from unposed multi-view images. Our approach leverages a Cross-View Transformer to robustly integrate information across arbitrary multi-view inputs, which then regresses a set of 3D Gaussian primitives endowed with semantic feature fields. This unified representation facilitates high-fidelity novel view synthesis, open-vocabulary 3D semantic segmentation, and depth prediction--all within a single, feed-forward pass. Extensive experiments demonstrate that Uni3R sets a new state of the art across multiple benchmarks, including in-domain datasets such as RE10K and ScanNet, as well as the out-of-domain dataset Mip-NeRF360. This work represents a new paradigm toward generalizable and unified 3D scene reconstruction and understanding.

Merging multiple Low-Rank Adaptation (LoRA) modules is promising for constructing general-purpose systems, yet challenging because LoRA update directions span different subspaces and contribute unevenly. When merged naively, such mismatches can weaken the directions most critical to certain task losses while overemphasizing relatively less important ones, ultimately reducing the model's ability to represent all tasks faithfully. We revisit this problem through two perspectives: subspace coverage, which captures how broadly LoRA directions cover diverse representational directions, and anisotropy, which reflects the imbalance of influence across those directions. We propose TARA-Merging (Task-Rank Anisotropy Alignment), which aligns merging weights using a preference-weighted cross-entropy pseudo-loss while preserving task-relevant LoRA subspaces. This ensures broad subspace coverage and mitigates anisotropy via direction-wise reweighting. Across eight vision and six NLI benchmarks, TARA-Merging consistently outperforms vanilla and LoRA-aware baselines, demonstrating strong robustness and generalization, and highlighting the importance of addressing both subspace coverage and anisotropy in LoRA merging.

Most vision-language models assume images have a single literal meaning, even though images are inherently polysemous. We propose a retrieval paradigm that models many-to-many relationships between images and text using interpretive lenses and introduce Lenses, a multi-prompt embedding model and dataset for polysemous image-text retrieval. The Lenses dataset contains 105,669 images and 732,405 captions, with each image paired with multiple captions and image-side prompts annotated across five categories: Literal, Figurative, Abstract, Background, and Emotional. Building on a multimodal large language model, the Lenses model uses learned lens tokens to extract lens-specific embeddings for every image and caption and compares these using a lens-masking similarity function with a global fallback that prioritizes same-lens matches while retaining a global pathway. Training uses a category-aware multi-positive contrastive loss and intra-set diversity regularization to align corresponding perspectives while preventing semantic collapse across lenses. We further propose lens-aware evaluation protocols, including category-aware ranking, that better reflect how humans match images and text. Experiments on the Lenses dataset and public benchmarks show that our model outperforms baselines on literal and non-literal retrieval and reduces over-reliance on literal cues.

Generating RGB-A videos, which include alpha channels for transparency, has wide applications. However, current methods often suffer from low quality due to confusion between RGB and alpha. In this paper, we address this problem by learning shiftable RGB-A distributions. We adjust both the latent space and noise space, shifting the alpha distribution outward while preserving the RGB distribution, thereby enabling stable transparency generation without compromising RGB quality. Specifically, for the latent space, we propose a transparency-aware bidirectional diffusion loss during VAE training, which shifts the RGB-A distribution according to likelihood. For the noise space, we propose shifting the mean of diffusion noise sampling and applying a Gaussian ellipse mask to provide transparency guidance and controllability. Additionally, we construct a high-quality RGB-A video dataset. Compared to state-of-the-art methods, our model excels in visual quality, naturalness, transparency rendering, inference convenience, and controllability. The released model is available on our website: https://donghaotian123.github.io/Wan-Alpha/.

Personalized Federated Learning (PFL) aims to train customized models for clients with highly heterogeneous data distributions while preserving data privacy. Existing approaches often rely on heuristics like clustering or model interpolation, which lack principled mechanisms for balancing heterogeneous client objectives. Serving M clients with distinct data distributions is inherently a multi-objective optimization problem, where achieving optimal personalization ideally requires M distinct models on the Pareto front. However, maintaining M separate models poses significant scalability challenges in federated settings with hundreds or thousands of clients. To address this challenge, we reformulate PFL as a few-for-many optimization problem that maintains only K shared server models (K << M) to collectively serve all M clients. We prove that this framework achieves near-optimal personalization: the approximation error diminishes as K increases and each client's model converges to each client's optimum as data grows. Building on this reformulation, we propose FedFew, a practical algorithm that jointly optimizes the K server models through efficient gradient-based updates. Unlike clustering-based approaches that require manual client partitioning or interpolation-based methods that demand careful hyperparameter tuning, FedFew automatically discovers the optimal model diversity through its optimization process. Experiments across vision, NLP, and real-world medical imaging datasets demonstrate that FedFew, with just 3 models, consistently outperforms other state-of-the-art approaches. Code is available at https://github.com/pgg3/FedFew.

Document AI has advanced rapidly and is attracting increasing attention. Yet, while most efforts have focused on document layout analysis (DLA), its generative counterpart, layout generation, remains underexplored. Distinct from traditional graphic layout design and room layout planning, document layout generation typically involves a larger number of elements per page and exhibits greater structural diversity and complexity. Currently, a major obstacle lies in the scarcity of diverse document layouts: academic papers with Manhattan-style structures dominate existing studies, while open-world genres such as newspapers and magazines remain severely underrepresented. To address this gap, we curate OmniDocLayout-1M, the first million-scale dataset of diverse document layouts, covering six common document types and comprising contemporary layouts collected from multiple sources. Moreover, since existing methods struggle in complex domains and often fail to arrange long sequences coherently, we introduce OmniDocLayout-LLM, a 0.5B model with designed two-stage Coarse-to-Fine learning paradigm: 1) learning universal layout principles from our dataset with coarse category definitions, and 2) transferring the knowledge to a specific domain with few fine-grained annotated samples. Extensive experiments demonstrate that our approach achieves strong performance on multiple domains in M^6Doc dataset, substantially surpassing both existing layout generation experts and several latest general-purpose LLMs. More information can be found at https://github.com/opendatalab/OmniDocLayout.

This paper presents an innovative approach for non-destructive 3D modeling of plant root structures, which are essential for nutrient and water uptake. While Ground Penetrating Radar (GPR) has been used for detecting subsurface objects with well-defined shapes, such as pipes, accurately reconstructing complex root structures remains a significant challenge. To address this, we propose a novel framework that leverages GPR signal shape priors for target signal detection and curve parameter regression across multiple B-scans. By integrating these detection and regression results, we obtain precise hyperbolic curves representing root structures. To further assess complete and detailed 3D root systems, we design a root shape modeling network that processes sparse 3D slices using a specialized point graph network and an upsampling module. The method can be extended to many applications, including civil engineering, geology, and environmental monitoring.

Understanding social interaction in video requires reasoning over a dynamic interplay of verbal and non-verbal cues: who is speaking, to whom, and with what gaze or gestures.While Multimodal Large Language Models (MLLMs) are natural candidates, simply adding visual inputs yields surprisingly inconsistent gains on social tasks. Our quantitative analysis of cross-modal attention inside state-of-the-art MLLMs reveals a core failure mode: in multi-speaker scenes, visual and textual tokens lack speaker-consistent alignment, exhibiting substantially weaker cross-modal attention than in object-centric images.To address this, we propose a multimodal multi-speaker attention alignment method that can be integrated into existing MLLMs. First, we introduce dynamic cross-modal head selection to identify attention heads most responsible for grounding. Then, an adaptive social-aware attention bias, computed from existing attention patterns and speaker locations, is injected into the attention mechanism. This bias reinforces alignment between a speaker's visual representation and their utterances without introducing trainable parameters or architectural changes.We integrate our method into three distinct MLLMs (LLaVA-NeXT-Video, Qwen2.5-VL, and InternVL3) and evaluate on three benchmarks (TVQA+, MMSI, OnlineMMSI). Across four social tasks, results demonstrate that our approach improves the ability of MLLMs and achieves state-of-the-art results.Attention visualizations confirm our method successfully focuses the model on speaker-relevant regions, enabling more robust multi-party social reasoning.

Text-to-image diffusion models exhibit remarkable generative capabilities, yet their internal operations remain opaque, particularly when handling prompts that are not fully descriptive. In such scenarios, models must make implicit decisions to generate details not explicitly specified in the text. This work investigates the hypothesis that this decision-making process is not diffuse but is computationally localized within the model's architecture. While existing localization techniques focus on prompt-related interventions, we notice that such explicit conditioning may differ from implicit decisions. Therefore, we introduce a probing-based localization technique to identify the layers with the highest attribute separability for concepts. Our findings indicate that the resolution of ambiguous concepts is governed principally by self-attention layers, identifying them as the most effective point for intervention. Based on this discovery, we propose ICM (Implicit Choice-Modification) - a precise steering method that applies targeted interventions to a small subset of layers. Extensive experiments confirm that intervening on these specific self-attention layers yields superior debiasing performance compared to existing state-of-the-art methods, minimizing artifacts common to less precise approaches.

This paper focuses on enhancing the grasping precision and generalization of manipulation policies learned via imitation learning. Diffusion-based policy learning methods have recently become the mainstream approach for robotic manipulation tasks. As grasping is a critical subtask in manipulation, the ability of imitation-learned policies to execute precise and generalizable grasps merits particular attention. Existing imitation learning techniques for grasping often suffer from imprecise grasp executions, limited spatial generalization, and poor object generalization. To address these challenges, we incorporate grasp prior knowledge into the diffusion policy framework. In particular, we employ a latent diffusion policy to guide action chunk decoding with grasp pose prior, ensuring that generated motion trajectories adhere closely to feasible grasp configurations. Furthermore, we introduce a self-supervised reconstruction objective during diffusion to embed the graspness prior: at each reverse diffusion step, we reconstruct wrist-camera images back-projected the graspness from the intermediate representations. Both simulation and real robot experiments demonstrate that our approach significantly outperforms baseline methods and exhibits strong dynamic grasping capabilities.

Although 3D Gaussian Splatting (3DGS) has achieved impressive performance in real-time rendering, its unordered Gaussians make level-of-detail (LoD) construction and model compression highly challenging, limiting its applicability in customized scenarios.In this work, we propose a learning-based Gaussian hierarchy representation that ranks Gaussians by their contribution to the scene, enabling flexible LoD representations across arbitrary Gaussian counts.We first introduce a unified, continuous formulation and metric for Gaussian hierarchy. Then, we introduce a hierarchy-based modulated rendering method built upon a Differentiable Decreasing Step Function, which enables efficient hierarchy learning while maintaining approximately equivalent rendering. Moreover, we develop a PDF-Guided Active-Region Sampling strategy that encourages the learned hierarchy to become widely distributed within its value range.Our method requires no additional training stages and produces Gaussian hierarchies within training time comparable to classical 3DGS. Experiments on multiple datasets show that our approach achieves performance comparable to or surpassing state-of-the-art methods in both LoD rendering and model pruning.

Wide-baseline matching (WBM) requires integrating geometric understanding, viewpoint changes, fine-grained perception, and occlusion reasoning, making it a challenging testbed for spatial reasoning in multimodal large language models (MLLMs) deployed in physical environments. However, current MLLMs lack systematic evaluation and training frameworks for these capabilities. We introduce ReasonMatch-Bench, a benchmark stratified by viewpoint displacement and matching granularity across indoor, outdoor, and object-centric scenarios, and show that current MLLMs still struggle with fine-grained wide-baseline correspondence: on a difficult 90-sample subset, human annotators achieve 84.0 F1, while the best existing baseline reaches 37.2. To bridge this gap, we build a scalable data-generation pipeline that automatically extracts wide-baseline view pairs from large-scale video-3D corpora, including RGB-D videos and SfM reconstructions, yielding diverse and verifiable supervision. We further propose Dynamic Correspondence Reinforcement Learning (DCRL), which combines Image-Level Viewpoint Progression and Point-Level Correspondence Curriculum to improve WBM training through verifiable rewards without explicit CoT supervision. Extensive experiments show that DCRL substantially improves ReasonMatch-Bench and transfers to related spatial benchmarks, while maintaining general visual understanding performance with modest gains on several benchmarks.