AAAI 2026 论文列表

Point cloud processing has become a cornerstone technology in many 3D vision tasks. However, arbitrary rotations introduce variations in point cloud orientations, posing a long-standing challenge for effective representation learning. The core of this issue is the disruption of the point cloud's intrinsic directional characteristics caused by rotational perturbations. Recent methods attempt to implicitly model rotational equivariance and invariance, preserving directional information and propagating it into deep semantic spaces. Yet, they often fall short of fully exploiting the multiscale directional nature of point clouds to enhance feature representations. To address this, we propose the Direction-Perceptive Vector Network (DiPVNet). At its core is an atomic dot-product operator that simultaneously encodes directional selectivity and rotation invariance--endowing the network with both rotational symmetry modeling and adaptive directional perception. At the local level, we introduce a Learnable Local Dot-Product (L2DP) Operator, which enables interactions between a center point and its neighbors to adaptively capture the non-uniform local structures of point clouds. At the global level, we leverage generalized harmonic analysis to prove that the dot-product between point clouds and spherical sampling vectors is equivalent to a direction-aware spherical Fourier transform (DASFT). This leads to the construction of a global directional response spectrum for modeling holistic directional structures. We rigorously prove the rotation invariance of both operators. Extensive experiments on challenging scenarios involving noise and large-angle rotations demonstrate that DiPVNet achieves state-of-the-art performance on point cloud classification and segmentation tasks.

Textile pattern generation (TPG) aims to synthesize fine-grained textile pattern images based on given clothing images. Although previous studies have not explicitly investigated TPG, existing image-to-image models appear to be natural candidates for this task. However, when applied directly, these methods often produce unfaithful results, failing to preserve fine-grained details due to feature confusion between complex textile patterns and the inherent non-rigid texture distortions in clothing images. In this paper, we propose a novel method, SLDDM-TPG, for faithful and high-fidelity TPG. Our method consists of two stages: (1) a latent disentangled network (LDN) that resolves feature confusion in clothing representations and constructs a multi-dimensional, independent clothing feature space; and (2) a semi-supervised latent diffusion model (S-LDM), which receives guidance signals from LDN and generates faithful results through semi-supervised diffusion training, combined with our designed fine-grained alignment strategy. Extensive evaluations show that SLDDM-TPG reduces FID by 4.1 and improves SSIM by up to 0.116 on our CTP-HD dataset, and also demonstrate good generalization on the VITON-HD dataset.

Efficient visual backbone design remains crucial for resource-constrained computer vision applications. Inspired by the adaptive continuous-time dynamics observed in biological neurons, we propose FVNet, a novel lightweight architecture that integrates liquid neural dynamics for efficient and dynamic visual feature extraction. Central to FVNet is the Fluid Temporal Flow Unit (FTFU), which employs continuous-time equations with learnable time constants to capture spatio-temporal dependencies adaptively. By further stacking these units in a Multi-Phase Fluid Block (MPFB), our model processes features across parallel temporal scales, enabling context-aware feature encoding without incurring excessive computational overhead. Through a discrete closed-form solution, FVNet achieves the representational power of continuous-time models while avoiding the instability and overhead of iterative numerical solvers. Extensive experiments on various vision tasks demonstrate that FVNet achieves superior performance and efficiency over existing state-of-the-art lightweight networks.

Dense visual prediction tasks, including object detection and segmentation, inherently require precise and discriminative positional information to delineate object boundaries and pixel regions. Recent DETR-based frameworks advance dense prediction tasks through iterative attention applied to content queries, with sampled proposals as position references. However, this paradigm suffers from the misaligned sampling distribution and insufficient interaction between the content and position features, thereby limiting the encoding effectiveness. To overcome these limitations, we investigate the encoding paradigm for content-position harmonization and propose an effective predictor for dense visual tasks, termed DAPE (DETR with hArmonized content-Position Encoding). DAPE introduces explicit position encoding to facilitate content enhancement while maintaining low memory overhead. To achieves this process, DAPE comprises a Shifted Query Sampler (SQS) that enforces strict alignment between the distributions of content and position queries, and a 2D Low-Rank Position Encoder (LRPE) that progressively modulates attention maps based on the aligned representations. DAPE provides a unified solution for various dense prediction tasks. Extensive experiments on object detection, instance segmentation, and few-shot detection benchmarks demonstrate that DAPE achieves state-of-the-art performance while reducing memory consumption.

Gait recognition has emerged as a promising biometric technique for long-distance and non-intrusive human identification. While Transformers have revolutionized vision tasks, their adaptation to gait recognition remains underexplored due to domain-specific challenges such as sparse silhouette modality, spatial-temporal dynamics, fine-grained motion cues, and limited training data. In this paper, we propose Gait Transformer (GaT), an end-to-end Transformer backbone specifically tailored for silhouette-based gait recognition. GaT introduces three key components: (1) a hybrid patch embedding module that combines convolutional stems with group-batch normalization to enhance structural preservation; (2) a decomposed token mixer that explicitly models both short-range and long-range dependencies across spatial-temporal dimensions; and (3) a hybrid positional encoding strategy that integrates absolute, relative, and rotary embeddings to support efficient training under data scarcity. Without relying on any pretraining, GaT achieves state-of-the-art performance on Gait3D, GREW, and CCGR-MINI.

Resolution generalization in image generation tasks enables the production of higher-resolution images with lower training resolution overhead. However, a key obstacle for diffusion transformers in addressing this problem is the mismatch between positional encodings seen at inference and those used during training. Existing strategies such as positional encodings interpolation, extrapolation, or hybrids, do not fully resolve this mismatch. In this paper, we propose a novel two-dimensional randomized positional encodings, namely RPE-2D, that prioritizes the order of image patches rather than their absolute distances, enabling seamless high- and low-resolution generation without training on multiple resolutions. Concretely, RPE-2D independently samples positions along the horizontal and vertical axes over an expanded range during training, ensuring that the encodings used at inference lie within the training distribution and thereby improving resolution generalization. We further introduce a simple random resize-and-crop augmentation to strengthen order modeling and add micro-conditioning to indicate the applied cropping pattern. On the ImageNet dataset, RPE-2D achieves state-of-the-art resolution generalization performance, outperforming competitive methods when trained at 256^2 and evaluated at 384^2 and 512^2, and when trained at 512^2 and evaluated at 768^2 and 1024^2. RPE-2D also exhibits outstanding capabilities in low-resolution image generation, multi-stage training acceleration, and multi-resolution inheritance.

Pansharpening is a powerful technique for generating high-resolution multispectral (HRMS) images by fusing currently available image pairs of low-resolution multispectral (LRMS) and texture-rich panchromatic (PAN) data, effectively addressing the physical constraints of satellite sensors. While recent generative diffusion models have demonstrated impressive performance gains in this domain, their prohibitive computational demands and training costs hinder practicality in resource-constrained remote sensing satellite systems. In this work, we propose NODiff, a novel diffusion framework that replaces the conventional attention-based denoising backbone with a neural operator, seamlessly integrating operator learning and generative modeling into an efficient yet effective solution for pansharpening. In practice, we implement our approach through a two-stage learning paradigm: First, we pretrain the proposed Neural Operator-based diffusion model to learn the high-resolution texture priors essential for pansharpening. Afterward, we freeze the pretrained parameters, and design a lightweight conditional detail guidance adapter to enable efficient fine-tuning for generating desired HRMS images. Meanwhile, a time-aware low-rank adaptation is introduced to dynamically refine high-frequency details potentially affected by spectral mode truncation. Extensive experiments on multiple benchmark datasets demonstrate that NODiff achieves competitive pansharpening performance while significantly reducing training and inference costs. Beyond pansharpening, our method provides new insights into building resource-efficient generative models.

Modern preference alignment methods, such as DPO, rely on divergence regularization to a reference model for training stability—but this creates a fundamental problem we call "reference mismatch." In this paper, we investigate the negative impacts of reference mismatch in aligning text-to-image (T2I) diffusion models, showing that larger reference mismatch hinders effective adaptation given the same amount of data, e.g., as when learning new artistic styles, or personalizing to specific objects. We demonstrate this phenomenon across text-to-image (T2I) diffusion models and introduce margin-aware preference optimization (MaPO), a reference-agnostic approach that breaks free from this constraint. By directly optimizing the likelihood margin between preferred and dispreferred outputs under the Bradley-Terry model without anchoring to a reference, MaPO transforms diverse T2I tasks into unified pairwise preference optimization. We validate MaPO's versatility across five challenging domains: (1) safe generation, (2) style adaptation, (3) cultural representation, (4) personalization, and (5) general preference alignment. Our results reveal that MaPO's advantage grows dramatically with reference mismatch severity, outperforming both DPO and specialized methods like DreamBooth while reducing training time by 15%. MaPO thus emerges as a versatile and memory-efficient method for generic T2I adaptation tasks.

In supervised learning, traditional image masking faces two key issues: (i) discarded pixels are underutilized, leading to a loss of valuable contextual information; (ii) masking may remove small or critical features, especially in fine-grained tasks. In contrast, masked image modeling (MIM) has demonstrated that masked regions can be reconstructed from partial input, revealing that even incomplete data can exhibit strong contextual consistency with the original image. This highlights the potential of masked regions as sources of semantic diversity. Motivated by this, we revisit the image masking approach, proposing to treat masked content as auxiliary knowledge rather than ignored. Based on this, we proposed MaskAnyNet, which combines masking with a relearning mechanism to exploit both visible and masked information. It can be easily extended to any model with an additional branch to jointly learn from the recomposed masked region. This approach leverages the semantic diversity of masked regions to enrich features and preserve fine-grained details. Experiments on CNN and Transformer backbones show consistent gains across multiple benchmarks. Further analysis confirms that the proposed method improves semantic diversity through the reuse of masked content.

Recently, 3D Gaussian Splatting (3DGS), an explicit scene representation technique, has shown significant promise for dynamic novel-view synthesis from monocular video input. However, purely data-driven 3DGS often struggles to capture the diverse physics-driven motion patterns in dynamic scenes. To fill this gap, we propose Physics‑Informed Deformable Gaussian Splatting (PIDG), which treats each Gaussian particle as a Lagrangian material point with time-varying constitutive parameters and is supervised by 2D optical flow via motion projection. Specifically, we adopt static-dynamic decoupled 4D decomposed hash encoding to reconstruct geometry and motion efficiently. Subsequently, we impose the Cauchy momentum residual as a physics constraint, enabling independent prediction of each particle’s velocity and constitutive stress via a time-evolving material field. Finally, we further supervise data fitting by matching Lagrangian particle flow to camera-compensated optical flow, which accelerates convergence and improves generalization. Experiments on a custom physics-driven dataset as well as on standard synthetic and real-world datasets demonstrate significant gains in physical consistency and monocular dynamic reconstruction quality.

For artificial intelligence to be safely deployed in high-risk domains, it must reliably know its limits. Selective prediction, or learning with a reject option, addresses this by enabling a model to abstain from prediction on inputs it deems unreliable, deferring them to a human expert. While deep ensembles have emerged as a leading approach for uncertainty estimation, their potential is often squandered by rejection methods that rely on static thresholds applied to the mean prediction. In this paper, we propose to learn a dynamic rejection policy directly from the rich behavioral signals of the ensemble itself. Our framework, DEGRE (Dynamic Ensembles Gating for REjection), is a novel meta-learning approach that trains a lightweight gating network on the ensemble’s consensus confidence and its internal disagreement (variance)— to explicitly discriminate between correct and incorrect predictions. Through rigorous evaluation across twelve diverse medical imaging benchmarks (MRI, X-ray, CT), DEGRE significantly advances selective prediction, achieving an average risk-coverage (AURC) reduction of 68.2% compared to the standard ensemble baseline. By providing a more reliable method for a model to recognize its own limitations, this learned, adaptive rejection mechanism paves the way for safer and more responsible integration of AI into critical clinical workflows.

We focus on the automatic evaluation of image captions in both reference-based and reference-free settings. Existing metrics based on large language models (LLMs) favor their own generations; therefore, the neutrality is in question. Most LLM-free metrics do not suffer from such an issue, whereas they do not always demonstrate high performance. To address these issues, we propose Pearl, an LLM-free supervised metric for image captioning, which is applicable to both reference-based and reference-free settings. We introduce a novel mechanism that learns the representations of image--caption and caption--caption similarities. Furthermore, we construct a human-annotated dataset for image captioning metrics that comprises approximately 333k human judgments collected from 2,360 annotators across over 75k images. Pearl outperformed other existing LLM-free metrics on the Composite, Flickr8K-Expert, Flickr8K-CF, Nebula, and FOIL datasets in both reference-based and reference-free settings.

Scaling Vision Transformers (ViTs) has yielded remarkable advancements in diverse vision tasks, albeit at the cost of escalating computational, memory, and parameter demands. Existing efficiency techniques typically address only one dimension, computation, memory, or parameters, lacking a cohesive approach. In this paper, we introduce Vision-MoR, a novel ViT architecture that unifies parameter sharing, spatially adaptive computation, and memory-efficient design into a single framework. Vision-MoR employs a spatial-aware router with shifted-window attention to dynamically assign per-patch recursion depths, coupled with a recursive Transformer loop enabling token-wise early exiting. This facilitates content-adaptive processing and recursive parameter reuse while preserving spatial locality. On ImageNet-1K, Vision-MoR Small attains 74.6% Top-1 accuracy with 140M FLOPs and 5.7M parameters, outperforming EfficientViT-M2 (70.8%) and SHViT-S1 (72.8%) at superior throughput. The Vision-MoR X-Large variant achieves 80.4% Top-1 and 95.2% Top-5 accuracy using 14.3M parameters and 2044M FLOPs, surpassing ResNet-50 and EfficientNet-B1. On COCO object detection, Vision-MoR X-Large yields 39.1 AP with the lowest latency among comparable models. These results underscore Vision-MoR's state-of-the-art accuracy-efficiency trade-offs, positioning it as a scalable, deployment-friendly backbone for real-time vision applications.

Perceiving threats is an innate human instinct. During driving, humans naturally focus their attention on objects that pose real potential risks. Motivated by this observation, we shift the focus from traditional class-based detection to a novel task termed threat-oriented reasoning detection in autonomous driving. This task aims to localize threat objects and reason about their threat levels from a driver-centric perspective. To support this task, we build a benchmark comprising diverse corner-case scenarios, annotated by multiple experienced drivers to reflect human-aligned threat cognition. Given the reasoning demands of this task, we then explore the capabilities of multi-modal large language models (MLLMs) and introduce two methods based on whether the MLLM supports object detection: 1) For MLLMs lacking detection capability, we introduce ThreatCoT, a plug-and-play training-free method that combines chain-of-thought (CoT) with a visual expert toolchain to support step-by-step reasoning. 2) For MLLMs with detection support, we introduce ThreatReasoner, an end-to-end reinforcement learning (RL)-based method built on the GRPO algorithm, which enables per-object reasoning through a fully unsupervised reward strategy. Both quantitative and qualitative experiments show that our methods can effectively unlock the new capabilities of MLLM in threat-oriented reasoning detection.

The demonstrated success of sparsely-gated Mixture-of-Experts (MoE) architectures, exemplified by models such as DeepSeek and Grok, has motivated researchers to investigate their adaptation to diverse domains. In real-world image super-resolution (Real-ISR), existing approaches mainly rely on fine-tuning pre-trained diffusion models through Low-Rank Adaptation (LoRA) module to reconstruct high-resolution (HR) images. However, these dense Real-ISR models are limited in their ability to adaptively capture the heterogeneous characteristics of complex real-world degraded samples or enable knowledge sharing between inputs under equivalent computational budgets. To address this, we investigate the integration of sparse MoE into Real-ISR and propose a Mixture-of-Ranks (MoR) architecture for single-step image super-resolution. We introduce a fine-grained expert partitioning strategy that treats each rank in LoRA as an independent expert. This design enables flexible knowledge recombination while isolating fixed-position ranks as shared experts to preserve common-sense features and minimize routing redundancy. Furthermore, we develop a degradation estimation module leveraging CLIP embeddings and predefined positive-negative text pairs to compute relative degradation scores, dynamically guiding expert activation. To better accommodate varying sample complexities, we incorporate zero-expert slots and propose a degradation-aware load-balancing loss, which dynamically adjusts the number of active experts based on degradation severity, ensuring optimal computational resource allocation. Comprehensive experiments validate our framework's effectiveness and state-of-the-art performance.

3D LiDAR scene completion from point clouds is a fundamental component of perception systems in autonomous vehicles. Previous methods have predominantly employed diffusion models for high‑fidelity reconstruction. However, their multi-step iterative sampling incurs significant computational overhead, limiting its real-time applicability. To address this, we propose LiNeXt: a lightweight, non‐diffusion network optimized for rapid and accurate point cloud completion. Specifically, LiNeXt first applies the Noise‑to‑Coarse (N2C) Module to denoise the input noisy point cloud in a single pass, thereby obviating the multi‑step iterative sampling of diffusion‑based methods. The Refine Module then takes the coarse point cloud and its intermediate features from the N2C Module to perform more precise refinement, further enhancing structural completeness. Furthermore, we observe that LiDAR point clouds exhibit a distance-dependent spatial distribution, being densely sampled at proximal ranges and sparsely sampled at distal ranges. Accordingly, we propose the Distance‑aware Selected Repeat strategy to generate a more uniformly distributed noisy point cloud. On the SemanticKITTI dataset, LiNeXt achieves a 199.8 times speedup in inference, reduces Chamfer Distance by 50.7 percent, and uses only 6.1 percent of the parameters compared with LiDiff. These results demonstrate the superior efficiency and effectiveness of LiNeXt for real-time scene completion.

Dynamic Vision Sensor (DVS) asynchronously records sparse events triggered by changes in pixel intensity, offering high temporal resolution and low latency. Existing frame-based methods process event data densely, violating its inherent sparsity and introducing computational redundancy. While asynchronous models preserve the event stream's native format, they often neglect spatial information, compromising their adaptability and efficiency. To address these limitations, we propose a Spatiotemporally Separated Sparse Network (S3Net) for efficient event stream encoding and learning. Specifically, we employ a learnable sparse encoding scheme to construct a voxel-structured representation that effectively extracts spatiotemporal relationships among event data. After that, we propose a dual-branch architecture to capture localized spatial dependencies and dynamic temporal patterns of event data. By explicitly decoupling spatial and temporal modeling, S3Net enables end-to-end asynchronous processing of variable-length event sequences, achieving both strong representational capacity and high computational efficiency. Experimental results on six event-based datasets demonstrate that S3Net achieves state-of-the-art performance. Compared to frame-based methods, it significantly reduces computational overhead and model complexity, while also outperforming existing asynchronous approaches in inference speed without compromising accuracy. Extensive experiments across six event-based datasets show that S3Net establishes new state-of-the-art performance. Our method reduces computational costs by 35% and model parameters by 27% compared to frame-based approaches, while delivering 1.58× faster inference than existing point-based methods at comparable accuracy levels.

Understanding 3D scene-level affordances from natural language instructions is essential for enabling embodied agents to interact meaningfully in complex environments. However, this task remains challenging due to the need for semantic reasoning and spatial grounding. Existing methods mainly focus on object-level affordances or merely lift 2D predictions to 3D, neglecting rich geometric structure information in point clouds and incurring high computational costs. To address these limitations, we introduce Task-Aware 3D Scene-level Affordance segmentation (TASA), a novel geometry-optimized framework that jointly leverages 2D semantic cues and 3D geometric reasoning in a coarse-to-fine manner. To improve the affordance detection efficiency, TASA features a task-aware 2D affordance detection module to identify manipulable points from language and visual inputs, guiding the selection of task-relevant views. To fully exploit 3D geometric information, a 3D affordance refinement module is proposed to integrate 2D semantic priors with local 3D geometry, resulting in accurate and spatially coherent 3D affordance masks. Experiments on SceneFun3D demonstrate that TASA significantly outperforms the baselines in both accuracy and efficiency in scene-level affordance segmentation.

Video generative models pre-trained on large-scale internet datasets have achieved remarkable success, excelling at producing realistic synthetic videos. However, they often generate clips based on static prompts (e.g., text or images), limiting their ability to model interactive and dynamic scenarios. In this paper, we propose Dynamic World Simulation (DWS), a novel approach to transform pre-trained video generative models into controllable world simulators capable of executing specified action trajectories. To achieve precise alignment between conditioned actions and generated visual changes, we introduce a lightweight, universal action-conditioned module that seamlessly integrates into any existing model. Instead of focusing on complex visual details, we demonstrate that consistent dynamic transition modeling is the key to building powerful world simulators. Building upon this insight, we further introduce a motion-reinforced loss that enhances action controllability by compelling the model to capture dynamic changes more effectively. Experiments demonstrate that DWS can be versatilely applied to both diffusion and autoregressive transformer models, achieving significant improvements in generating action-controllable, dynamically consistent videos across games and robotics domains. Moreover, to facilitate the applications of the learned world simulator in downstream tasks such as model-based reinforcement learning, we propose prioritized imagination to improve sample efficiency, demonstrating competitive performance compared with state-of-the-art methods.