AAAI 2026 论文列表

Flow Matching (FM) is an efficient generative modeling framework, but aligning it with human preferences remains underexplored.~Although applying Direct Preference Optimization (DPO) to diffusion models has yielded improvements, directly extending DPO-like methods to FM poses three challenges: 1) Incompatibility with ODE-based models, 2) Heavy computational cost from full model fine-tuning, and 3) Reliance on reference model quality. To address these limitations, we propose Preference Classifier for Flow Matching (PC-Flow), a novel reference-free preference alignment framework. Specifically, we reinterpret FM’s deterministic ODE as an equivalent SDE to enable DPO-style learning. Then, we introduce a lightweight classifier to model relative preferences exclusively. This approach decouples alignment from the generative model, eliminating the need for costly fine-tuning or a reference model. Theoretically, PC-Flow guarantees consistent preference-guided distribution evolution, achieves a DPO-equivalent objective without a reference model, and progressively steers generation toward preferred outputs. Experiments show that PC-Flow achieves DPO-level alignment with significantly lower training costs.

Real-time, high-fidelity monocular depth estimation from remote sensing imagery is crucial for numerous applications, yet existing methods face a stark trade-off between accuracy and efficiency. Although using Vision Transformer (ViT) backbones for dense prediction is fast, they often exhibit poor perceptual quality. Conversely, diffusion models offer high fidelity but at a prohibitive computational cost. To overcome these limitations, we propose Depth Detail Diffusion for Remote Sensing Monocular Depth Estimation (D³-RSMDE), an efficient framework designed to achieve an optimal balance between speed and quality. Our framework first leverages a ViT-based module to rapidly generate a high-quality preliminary depth map construction, which serves as a structural prior, effectively replacing the time-consuming initial structure generation stage of diffusion models. Based on this prior, we propose a Progressive Linear Blending Refinement (PLBR) strategy, which uses a lightweight U-Net to refine the details in only a few iterations. The entire refinement step operates efficiently in a compact latent space supported by a Variational Autoencoder (VAE). Extensive experiments demonstrate that D³-RSMDE achieves a notable 11.85% reduction in the Learned Perceptual Image Patch Similarity (LPIPS) perceptual metric over leading models like Marigold, while also achieving over a 40× speedup in inference and maintaining VRAM usage comparable to lightweight ViT models.

Bokeh is used in photography to emphasize the selected subject by smoothly blurring the out-of-focus region with appealing highlights. While recent advances have achieved impressive results in rendering realistic blur, existing frameworks typically rely on disparity maps and bokeh-relevant inputs (e.g., focal distance and blur size), and face significant challenges in video bokeh rendering due to limited temporal consistency. In this paper, we propose BokehCrafter, the first video diffusion framework that generates temporally coherent and visually pleasing bokeh effects from all-in-focus video inputs under user-friendly input conditions. Specifically, we leverage a dual-stream attention mechanism, integrating a reference image branch and a rendering instruction branch. We propose a Bokeh Image Extraction (BIE) module and a CLIP-based text encoder to extract image and text features, respectively, whose outputs are fused via a Text-Image Fusion (TIF) module to enable fine-grained and controllable bokeh rendering. To support the novel capabilities of our model, we construct Video Bokeh Scenes (VBS), a large-scale dataset containing a wide variety of bokeh videos with corresponding rendering instructions, across various scenes and rendering settings. Extensive experiments demonstrate that our method significantly outperforms state-of-the-art methods in both bokeh rendering quality and temporal consistency.

In recent years, lossy compression algorithms such as H.264/AVC, H.265/HEVC, and H.266/VVC have been proposed and widely applied in image and video encoding. However, these compression algorithms inevitably introduce various complex types of compression artifacts, which severely degrade image quality. Although existing methods have attempted to remove artifacts through filter design or probabilistic prior modeling, they are often effective only for specific types of artifacts, lacking generalization and adaptability. To address this, we propose a novel image compression artifacts removal model: ARMoE, which combines multiple frequency domain transformations with the Mixture of Experts (MoE). Considering the frequency distribution and energy distribution differences of images, we introduce various frequency domain transformations as expert branches and use the Sparse Activation Strategy to adaptively select the optimal frequency domain expert to suppress compression artifacts, achieving an efficient artifacts removal method. Furthermore, we reencode and decode multiple original uncompressed high-quality datasets, including DF2K and Kodak24, using the VTM-20.0 codec under the H.266/VVC standard, constructing a more challenging artifacts dataset. We conducted rigorous comparative experiments with current state-of-the-art image restoration methods and the results demonstrate that ARMoE exhibits outstanding image restoration capability.

Fair incomplete multi-view clustering (FIMVC) confronts a critical yet unresolved challenge, as existing methods often fail to address the intertwined issues of data missingness and algorithmic bias simultaneously. In this paper, we propose a novel FIMVC method named Adversarial Fair Incomplete Multi-View Clustering (AFIMVC). The core of AFIMVC is a new adaptive adversarial disentanglement mechanism. This mechanism trains the feature encoder to produce representations that are invariant to sensitive attributes by adversary learning, where the adversarial intensity is dynamically controlled by the model's real-time bias. Additionally, we develop a probabilistic cross-view contrastive learning strategy to achieve semantic consistency in latent space. To handle missing data, AFIMVC employs a context-aware fusion strategy that leverages cross-sample attention to robustly synthesize a unified representation from incomplete views. Extensive experiments demonstrate that AFIMVC achieves a state-of-the-art balance between clustering accuracy and fairness, significantly outperforming existing methods.

Purple flare, a diffuse chromatic aberration artifact commonly found around highlight areas, severely degrades the tone transition and color of the image. Existing traditional methods are based on hand-crafted features, which lack flexibility and rely entirely on fixed priors, while the scarcity of paired training data critically hampers deep learning. To address this issue, we propose a novel network built upon decoupled HSV Look-Up Tables (LUTs). The method aims to simplify color correction by adjusting the Hue (H), Saturation (S), and Value (V) components independently. This approach resolves the inherent color coupling problems in traditional methods. Our model adopts a two-stage architecture: First, a Chroma-Aware Spectral Tokenizer (CAST) converts the input image from RGB space to HSV space and independently encodes the Hue (H) and Value (V) channels into a set of semantic tokens describing the Purple flare status; second, the HSV-LUT module takes these tokens as input and dynamically generates independent correction curves (1D-LUTs) for the three channels H, S, and V. To effectively train and validate our model, we built the first large-scale purple flare dataset with diverse scenes. We also proposed new metrics and a loss function specifically designed for this task. Extensive experiments demonstrate that our model not only significantly outperforms existing methods in visual effects but also achieves state-of-the-art performance on all quantitative metrics.

Accurately simulating existing 3D objects and a wide variety of materials often demands expert knowledge and time-consuming physical parameter tuning to achieve the desired dynamic behavior. We introduce MotionPhysics, an end‑to‑end differentiable framework that infers plausible physical parameters from a user-proved natural language prompt for a chosen 3D scene of interest, removing the need for guidance from ground‑truth trajectories or annotated videos. Our approach first utilizes a multimodal large language model to estimate material parameter values, which are constrained to be within plausible ranges. We further propose a learnable motion distillation loss, which extracts robust motion priors from pretrained video diffusion models while minimizing appearance and geometry inductive biases to guide the simulation. We evaluate MotionPhysics across more than thirty scenarios, including real-world, human-designed, and AI-generated 3D objects, spanning a wide range of materials such as elastic solids, metals, foams, sand, and both Newtonian and non-Newtonian fluids. We demonstrate that it produces visually realistic dynamic simulations guided by natural language, surpassing the state of the art, with physically plausible parameters that are automatically determined.

Face super-resolution (FSR) aims to reconstruct high-resolution (HR) face images from low-resolution (LR) inputs. While recent methods have advanced this task through architectural innovations and generative modeling, but they often leads to semantically inconsistent structures and unrealistic textures, particularly under high magnification. To mitigate these limitations, we draw inspiration from the human artistic process of “structuring before detailing” and propose a progressive prior-guided restoration strategy. Specifically, we first introduce a Sketching Structure Prior (SSP) module that embeds global semantics and refines local geometry through implicit parsing guidance and explicit spatial modulation. Then, an Associative Texture Prior (ATP) module leverages a High-Quality Dictionary (HD) learned from high-quality reconstruction to guide fine-grained detail recovery. Finally, to unify structure and detail features, we design a Holistic Prior Fusion (HPF) module that adaptively integrates them within semantically consistent facial regions. Our method surpasses state-of-the-art on CelebA and Helen in both structural fidelity and texture realism.

Unsupervised visible-infrared person re-identification (USVI-ReID) aims to match individuals across visible and infrared cameras without relying on any annotation. Given the significant gap across visible and infrared modality, estimating reliable cross-modality association becomes a major challenge in USVI-ReID. Existing methods usually adopt optimal transport to associate the intra-modality clusters, which is prone to propagating the local cluster errors, and also overlooks global instance-level relations. By mining and attending to the visible-infrared modality bias, this paper focuses on addressing cross-modality learning from two aspects: bias-mitigated global association and modality-invariant representation learning. Motivated by the camera-aware distance rectification in single-modality re-ID, we propose modality-aware Jaccard distance to mitigate the distance bias caused by modality discrepancy, so that more reliable cross-modality associations can be estimated through global clustering. To further improve cross-modality representation learning, a `split-and-contrast' strategy is designed to obtain modality-specific global prototypes. By explicitly aligning these prototypes under global association guidance, modality-invariant yet ID-discriminative representation learning can be achieved. While conceptually simple, our method obtains state-of-the-art performance on benchmark VI-ReID datasets and outperforms existing methods by a significant margin, validating its effectiveness.

Recent advances in audio-driven portrait animation have demonstrated impressive capabilities. However, existing methods struggle to align with fine-grained human preferences across multiple dimensions, such as motion naturalness, lip-sync accuracy, and visual quality. This is due to the difficulty of optimizing among competing preference objectives, which often conflict with one another, and the scarcity of large-scale, high-quality datasets with multidimensional preference annotations. To address these, we first introduce Talking-Critic, a multimodal reward model that learns human-aligned reward functions to quantify how well generated videos satisfy multidimensional expectations. Leveraging this model, we curate Talking-NSQ, a large-scale multidimensional human preference dataset containing 410K preference pairs. Finally, we propose Timestep-Layer adaptive multi-expert Preference Optimization (TLPO), a novel framework for aligning diffusion-based portrait animation models with fine-grained, multidimensional preferences. TLPO decouples preferences into specialized expert modules, which are then fused across timesteps and network layers, enabling comprehensive, fine-grained enhancement across all dimensions without mutual interference. Experiments demonstrate that Talking-Critic significantly outperforms existing methods in aligning with human preference ratings. Meanwhile, TLPO achieves substantial improvements over baseline models in lip-sync accuracy, motion naturalness, and visual quality, exhibiting superior performance in both qualitative and quantitative evaluations.

Recent advances in point cloud analysis have increasingly leveraged large-scale unlabeled data through self-supervised representation learning. Autoregressive models based on next-token prediction have shown strong performance, but they usually model point clouds as linear sequences, ignoring their inherent spatial structure. To address this limitation, we propose PointChain, a novel autoregressive paradigm inspired by human perception mechanisms, designed to better align with the structural properties of point cloud. Specifically, we introduce structural chain encoding, which models the understanding process as a global-to-local structural chain inference, preserving spatial relationships throughout the prediction sequence. During pre-training, we design two auxiliary tasks: a next-scale prediction task that encourages cross-scale reasoning, and a scale-level contrastive learning task that promotes semantic consistency across scales. These components guide the model to learn more discriminative and generalizable point cloud representations. Experiments on multiple benchmarks, using both Transformer and Mamba backbones, validate the effectiveness of our approach. PointChain achieves state-of-the-art performance on several downstream tasks, including 93.75% accuracy on the hardest split of ScanObjectNN without voting strategy.

To facilitate the large-scale deployment of autonomous driving in real-world scenarios, developing low-cost and high-performance 3D object detection systems has become a critical technical challenge. Although high-beam LiDARs provide denser point cloud data, their prohibitive hardware cost and high power consumption limit their practicality. In contrast, low-beam LiDARs offer advantages in terms of affordability and energy efficiency, but often suffer from inadequate perception accuracy due to their sparser point cloud data. This paper focuses on the task of multimodal 3D object detection with low-beam LiDARs, and proposes a novel approach that integrates temporal and spatial representation learning to enhance detection accuracy under sparser sensor conditions. Specifically, our approach comprises: (1) a Temporal Feature Prediction Learning (TFPL) module, which predicts the current BEV representation based on a sequence of historical BEV features; (2) a Spatial Feature Observation Learning (SFOL) module, which aligns BEV (Bird's-Eye-View) features from high-beam and low-beam LiDAR to enforce the low-beam features to approximate high-beam representations; (3) an Uncertainty-Aware Fusion (UAF) strategy, which performs feature-wise weighting between the predicted and observed BEV features by leveraging channel-wise variances, effectively mitigating perturbations in the learned BEV representations. Extensive experiments on the KITTI and nuScenes 3D object detection datasets demonstrate that the proposed approach significantly improves detection performance under low-beam LiDAR configurations.

Multi-modal Salient Object Detection (SOD) shows an improvement over its uni-modal counterpart by exploiting the complementary benefits between modalities. However, this improvement relies on complete multi-modal information, which is difficult to be guaranteed in practice due to sensor failures and transmission errors. To address this issue, we propose a robust multi-modal SOD framework that enhances the adaptability to modality-missing conditions, while maintaining comparable performance in the modality-complete condition. Nevertheless, flexibly handling modality-missing and modality-complete cases and integrating their corresponding multi-modal features in a unified framework is non-trivial. To this end, we achieve this framework by designing a Cascaded Mixture-of-Experts (CMoE) network that sequentially incorporates missing-aware and multi-modal MoE. Specifically, the missing-aware MoE employs three modality-reconstruction experts with a soft router to adaptively reconstruct feature representations for both missing and available modalities, assisted by an expert modulation loss that guides the router to assign expert weights according to missing conditions. The multi-modal MoE adopts two homogeneous uni-modal experts with learned modality-specific knowledge tailored for integrating modality features, which are dynamically combined via the soft router. The cascaded architecture fully empowers CMoE with the flexibility across varying input cases. Extensive experiments on modality-missing and modality-complete conditions demonstrate the effectiveness of the proposed method.

The scarcity of paired data severely limits the performance and generalization of learning-based underwater image enhancement (UIE) methods. This challenge is particularly prominent in scenes with complex degradations. Semi-supervised learning has emerged as a promising solution by enabling the utilization of large-scale unlabeled data. However, its effectiveness is limited by the use of static, model-agnostic metrics for pseudo-label reliability assessment. To address this, we propose SEA-PACE, a novel semi-supervised framework that integrates model-aware uncertainty modeling and self-paced consistency learning to fully exploit unlabeled data for UIE. Specifically, we design a Model-Aware Reliability Estimator (MARE) that quantifies the uncertainty of the teacher model's predictions through Gaussian Process Regression in latent feature space. The resulting uncertainty is then transformed into reliability weights via a rank-based mapping. Additionally, we apply the Self-Paced Consistency Learning (SPCL) strategy that employs a loss-aware schedule to dynamically prioritize high-confidence pseudo-labels, gradually incorporating more challenging samples during training. Extensive experiments on several public UIE benchmarks demonstrate that SEA-PACE consistently surpasses state-of-the-art methods in both visual quality and generalization capability.

Image deraining is crucial for vision applications but is challenged by the complex multi-scale physics of rain and its coupling with scenes. To address this challenge, a novel approach inspired by multi-stage image restoration is proposed, incorporating Point Spread Function (PSF) mechanisms to reveal the image degradation process while combining dynamic physical modeling with sequential feature fusion transfer, named SD-PSFNet. Specifically, SD-PSFNet employs a sequential restoration architecture with three cascaded stages, allowing multiple dynamic evaluations and refinements of the degradation process estimation. The network utilizes components with learned PSF mechanisms to dynamically simulate rain streak optics, enabling effective rain-background separation while progressively enhancing outputs through novel PSF components at each stage. Additionally, SD-PSFNet incorporates adaptive gated fusion for optimal cross-stage feature integration, enabling sequential refinement from coarse rain removal to fine detail restoration. Our model achieves state-of-the-art PSNR/SSIM metrics on Rain100H (33.12dB/0.9371), RealRain-1k-L (42.28dB/0.9872), and RealRain-1k-H (41.08dB/0.9838). In summary, SD-PSFNet demonstrates excellent capability in complex scenes and dense rainfall conditions, providing a new physics-aware approach to image deraining.

Multimodal Large Language Models (MLLMs) have shown remarkable progress in temporal or spatial localization tasks, but struggle with joint spatio-temporal video grounding (STVG). We identify two key bottlenecks hindering this capability: (1) the sheer number of visual tokens makes long-range and fine-grained visual modeling challenging; (2) generating a long sequence of bounding boxes in text makes it hard to accurately align each box with its specific video frame. Distinct from prior efforts that rely on attaching complex modules, we argue for a more elegant paradigm that unlocks the inherent potential of MLLMs and leverages their strengths. To this end, we propose \textbf{\textit{SpaceVLLM}}, a MLLM equipped with spatio-temporal video grounding capabilities. Specifically, we propose Spatio-Temporal Aware Queries, interleaved with video frames, to guide the MLLM in capturing both static appearance and dynamic motion features. We further present a lightweight Query-Guided Space Head that maps queries to precise spatial coordinates, bypassing the need for direct textual coordinate generation and enabling the MLLM to focus on video understanding. To further facilitate research in this area, we propose an automated data synthesis pipeline to construct \textbf{V-STG} dataset, comprising 110K STVG instances. Extensive experiments show that \textit{SpaceVLLM} achieves the state-of-the-art performance on STVG benchmarks and maintains strong performance on various video understanding tasks, validating our approach's effectiveness.

Modern autonomous driving (AD) systems leverage 3D object detection to perceive foreground objects in 3D environments for subsequent prediction and planning. Visual 3D detection based on RGB cameras provides a cost-effective solution compared to the LiDAR paradigm. While achieving promising detection accuracy, current deep neural network-based models remain highly susceptible to adversarial examples. The underlying safety concerns motivate us to investigate realistic adversarial attacks in AD scenarios. Previous work has demonstrated the feasibility of placing adversarial posters on the road surface to induce hallucinations in the detector. However, the unnatural appearance of the posters makes them easily noticeable by humans, and their fixed content can be readily targeted and defended. To address these limitations, we propose the AdvRoad to generate diverse road-style adversarial posters. The adversaries have naturalistic appearances resembling the road surface while compromising the detector to perceive non-existent objects at the attack locations. We employ a two-stage approach, termed Road-Style Adversary Generation and Scenario-Associated Adaptation, to maximize the attack effectiveness on the input scene while ensuring the natural appearance of the poster, allowing the attack to be carried out stealthily without drawing human attention. Extensive experiments show that AdvRoad generalizes well to different detectors, scenes, and spoofing locations. Moreover, physical attacks further demonstrate the practical threats in real-world environments.

Random Sample Consensus (RANSAC) is a fundamental approach for robustly estimating parametric models from noisy data. Existing learning-based RANSAC methods utilize deep learning to enhance the robustness of RANSAC against outliers. However, these approaches are trained and tested on the data generated by the same algorithms, leading to limited generalization to out-of-distribution data during inference. Therefore, in this paper, we introduce a novel diffusion-based paradigm that progressively injects noise into ground-truth data, simulating the noisy conditions for training learning-based RANSAC. To enhance data diversity, we incorporate Monte Carlo sampling into the diffusion paradigm, approximating diverse data distributions by introducing different types of randomness at multiple stages. We evaluate our approach in the context of feature matching through comprehensive experiments on the ScanNet and MegaDepth datasets. The experimental results demonstrate that our Monte Carlo diffusion mechanism significantly improves the generalization ability of learning-based RANSAC. We also develop extensive ablation studies that highlight the effectiveness of key components in our framework.

Recent approaches for few-shot 3D point cloud semantic segmentation typically require a two-stage learning process, i.e., a pre-training stage followed by a few-shot training stage. While effective, these methods face overreliance on pre-training, which hinders model flexibility and adaptability. Some models tried to avoid pre-training yet failed to capture ample information. In addition, current approaches focus on visual information in the support set and neglect or do not fully exploit other useful data, such as textual annotations. This inadequate utilization of support information impairs the performance of the model and restricts its zero-shot ability. To address these limitations, we present a novel pre-training-free network, named Efficient Point Cloud Semantic Segmentation for Few- and Zero-shot scenarios. Our EPSegFZ incorporates three key components. A Prototype-Enhanced Registers Attention (ProERA) module and a Dual Relative Positional Encoding (DRPE)-based cross-attention mechanism for improved feature extraction and accurate query-prototype correspondence construction without pre-training. A Language-Guided Prototype Embedding (LGPE) module that effectively leverages textual information from the support set to improve few-shot performance and enable zero-shot inference.Extensive experiments show that our method outperforms the state-of-the-art method by 5.68% and 3.82% on the S3DIS and ScanNet benchmarks, respectively.