AAAI 2026 Papers

Appearance-based gaze estimation, aiming to predict accurate 3D gaze direction from a single facial image, has made promising progress in recent years. However, most methods suffer significant performance degradation in cross-domain evaluation due to interference from gaze-irrelevant factors, such as expressions, wearables, and image quality. To alleviate this problem, we present a novel Hybrid-domain Adaptative Representation Learning (shorted by HARL) framework that exploits multi-source hybrid datasets to learn robust gaze representation. More specifically, we propose to disentangle gaze-relevant representation from low-quality facial images by aligning features extracted from high-quality near-eye images in an unsupervised domain-adaptation manner, which hardly requires any computational or inference costs. Additionally, we analyze the effect of head-pose and design a simple yet efficient sparse graph fusion module to explore the geometric constraint between gaze direction and head-pose, leading to a dense and robust gaze representation. Extensive experiments on EyeDiap, MPIIFaceGaze, and Gaze360 datasets demonstrate that our approach achieves state-of-the-art accuracy of 5.02, 3.36, and 9.26 degrees respectively, and present competitive performances through cross-dataset evaluation.

The rapid emergence of image synthesis models poses challenges to the generalization of AI-generated image detectors. However, existing methods often rely on model-specific features, leading to overfitting and poor generalization. In this paper, we introduce the Multi-Cue Aggregation Network (MCAN), a novel framework that integrates different yet complementary cues as input. MCAN employs a mixture-of-encoders adapter to dynamically process these cues, enabling more adaptive and robust feature representation. Our cues include the input image itself, which represents the overall content, and high-frequency components that emphasize edge details. Additionally, we introduce a Chromatic Inconsistency (CI) cue, which normalizes intensity values and captures noise information introduced during the image acquisition process in real images, making these noise patterns more distinguishable from those in AI-generated content. Unlike prior methods, MCAN employs a multi-cue aggregation strategy, leveraging spatial, frequency, and chromaticity-based cues. These cues are intrinsically more indicative of real images, enhancing cross-model generalization. Extensive experiments on the GenImage, Chameleon, and UniversalFakeDetect benchmark validate the state-of-the-art performance of MCAN. In the GenImage dataset, MCAN outperforms the best state-of-the-art method by up to 7.4\% in average ACC across eight different image generators.

We introduce FLAG-4D, a novel framework for generating novel views of dynamic scenes by reconstructing how 3D Gaussian primitives evolve through space and time. Existing methods typically rely on a single Multilayer Perceptron(MLP) to model temporal deformations, and they often struggle to capture complex point motions and fine-grained dynamic details consistently over time, especially from sparse input views. Our approach, FLAG-4D overcomes this by employing a dual-deformation network that dynamically warps a canonical set of 3D Gaussians over time into new positions and anisotropic shapes. This dual-deformation network consists of an Instantaneous Deformation Network (IDN) for modeling fine-grained, local deformations, and Global Motion Network (GMN) for capturing long-range dynamics, refined via mutual learning. To ensure these deformations are both accurate and temporally smooth, FLAG-4D incorporates dense motion features from a pretrained optical flow backbone. We fuse these motion cues from adjacent timeframes and use a deformation-guided attention mechanism to align this flow information with the current state of each evolving 3D Gaussian. Extensive experiments demonstrate that FLAG-4D achieves higher-fidelity and more temporally coherent reconstructions with finer detail preservation than state-of-the-art methods.

3D Vision-Language Foundation Models (VLFMs) have demonstrated strong generalization and zero-shot recognition capabilities in open-world point cloud processing tasks. However, their performance often degrades in practical scenarios where data are noisy, incomplete, or drawn from distributions that differ from the training data. To address this challenge, we propose Uni-Adapter, a novel training-free online test-time adaptation (TTA) strategy for 3D VLFMs based on dynamic prototype learning. Uni-Adapter maintains a 3D cache that stores class-specific cluster centers as prototypes, which are continuously updated to capture intra-class variability under heterogeneous data distributions. These dynamic prototypes serve as anchors for cache-based logit computation through similarity scoring. In parallel, a graph-based label smoothing module models inter-prototype similarities to enforce label consistency among related prototypes. Finally, predictions from the original 3D VLFM and the refined 3D cache are unified through entropy-weighted aggregation to ensure reliable adaptation. Without retraining, Uni-Adapter effectively mitigates distribution shifts and achieves state-of-the-art performance across diverse 3D benchmarks and multiple 3D VLFMs, improving performance on ModelNet-40C by 10.55%, ScanObjectNN-C by 8.26%, and ShapeNet-C by 4.49% over the source 3D VLFMs.

Universal Photometric Stereo is a promising approach for recovering surface normals without strict lighting assumptions. However, it struggles when multi-illumination cues are unreliable, such as under biased lighting or in shadows or self-occluded regions of complex in-the-wild scenes. We propose GeoUniPS, a universal photometric stereo network that integrates synthetic supervision with high-level geometric priors from large-scale 3D reconstruction models pretrained on massive in-the-wild data. Our key insight is that these 3D reconstruction models serve as visual-geometry foundation models, inherently encoding rich geometric knowledge of real scenes. To leverage this, we design a Light-Geometry Dual-Branch Encoder that extracts both multi-illumination cues and geometric priors from the frozen 3D reconstruction model. We also address the limitations of the conventional orthographic projection assumption by introducing the PS-Perp dataset with realistic perspective projection to enable learning of spatially varying view directions. Extensive experiments demonstrate that GeoUniPS delivers state-of-the-arts performance across multiple datasets, both quantitatively and qualitatively, especially in the complex in-the-wild scenes.

Contrastive pre-trained vision-language models, such as CLIP, demonstrate strong generalization abilities in zero-shot classification by leveraging embeddings extracted from image and text encoders. This paper aims to robustly fine-tune these vision-language models on in-distribution (ID) data without compromising their generalization abilities in out-of-distribution (OOD) and zero-shot settings. Current robust fine-tuning methods tackle this challenge by reusing contrastive learning, which was used in pre-training, for fine-tuning. However, we found that these methods distort the geometric structure of the embeddings, which plays a crucial role in the generalization of vision-language models, resulting in limited OOD and zero-shot performance. To address this, we propose Difference Vector Equalization (DiVE), which preserves the geometric structure during fine-tuning. The idea behind DiVE is to constrain difference vectors, each of which is obtained by subtracting the embeddings extracted from the pre-trained and fine-tuning models for the same data sample. By constraining the difference vectors to be equal across various data samples, we effectively preserve the geometric structure. Therefore, we introduce two losses: average vector loss (AVL) and pairwise vector loss (PVL). AVL preserves the geometric structure globally by constraining difference vectors to be equal to their weighted average. PVL preserves the geometric structure locally by ensuring a consistent multimodal alignment. Our experiments demonstrate that DiVE effectively preserves the geometric structure, achieving strong results across ID, OOD, and zero-shot metrics.

Identification of fine-grained embryo developmental stages during In Vitro Fertilization (IVF) is crucial for assessing embryo viability. Although recent deep learning methods have achieved promising accuracy, existing discriminative models fail to utilize the distributional prior of embryonic development to improve accuracy. Moreover, their reliance on single-focal information leads to incomplete embryonic representations, making them susceptible to feature ambiguity under cell occlusions. To address these limitations, we propose EmbryoDiff, a two-stage diffusion-based framework that formulates the task as a conditional sequence denoising process. Specifically, we first train and freeze a frame-level encoder to extract robust multi-focal features. In the second stage, we introduce a Multi-Focal Feature Fusion Strategy that aggregates information across focal planes to construct a 3D-aware morphological representation, effectively alleviating ambiguities arising from cell occlusions. Building on this fused representation, we derive complementary semantic and boundary cues and design a Hybrid Semantic-Boundary Condition Block to inject them into the diffusion-based denoising process, enabling accurate embryonic stage classification. Extensive experiments on two benchmark datasets show that our method achieves state-of-the-art results. Notably, with only a single denoising step, our model obtains the best average test performance, reaching 82.8% and 81.3% accuracy on the two datasets, respectively.

Infrared and visible image fusion generates all-weather perception-capable images by combining complementary modalities, enhancing environmental awareness for intelligent unmanned systems. Existing methods either focus on pixel-level fusion while overlooking downstream task adaptability or implicitly learn rigid semantics through cascaded detection/segmentation models, unable to interactively address diverse semantic target perception needs. We propose CtrlFuse, a controllable image fusion framework that enables interactive dynamic fusion guided by mask prompts. The model integrates a multi-modal feature extractor, a reference prompt encoder (RPE), and a prompt-semantic fusion module (PSFM). The RPE dynamically encodes task-specific semantic prompts by fine-tuning pre-trained segmentation models with input mask guidance, while the PSFM explicitly injects these semantics into fusion features. Through synergistic optimization of parallel segmentation and fusion branches, our method achieves mutual enhancement between task performance and fusion quality. Experiments demonstrate state-of-the-art results in both fusion controllability and segmentation accuracy, with the adapted task branch even outperforming the original segmentation model.

Despite Video Large Language Models (Video-LLMs) having rapidly advanced in recent years, perceptual hallucinations pose a substantial safety risk, which severely restricts their real-world applicability. While several methods for hallucination mitigation have been proposed, they often compromise the model’s capacity for video understanding and reasoning. In this work, we propose SmartSight, a pioneering step to address this issue in a training-free manner by leveraging the model’s own introspective capabilities. Specifically, SmartSight generates multiple candidate responses to uncover low-hallucinated outputs that are often obscured by standard greedy decoding. It assesses the hallucination of each response using the Temporal Attention Collapse score, which measures whether the model over-focuses on trivial temporal regions of the input video when generating the response. To improve efficiency, SmartSight identifies the Visual Attention Vanishing point, enabling more accurate hallucination estimation and early termination of hallucinated responses, leading to a substantial reduction in decoding cost. Experiments show that SmartSight substantially lowers hallucinations for QwenVL-2.5-7B by 10.59% on VRIPT-HAL, while simultaneously enhancing video understanding and reasoning, boosting performance on VideoMMMU by 8.86%. These results highlight SmartSight’s effectiveness in improving the reliability of open-source Video-LLMs.

Microaneurysms (MAs), the earliest pathognomonic signs of Diabetic Retinopathy (DR), present as sub-60 μm lesions in fundus images with highly variable photometric and morphological characteristics, rendering manual screening not only labor-intensive but inherently error-prone. While diffusion-based anomaly detection has emerged as a promising approach for automated MA screening, its clinical application is hindered by three fundamental limitations. First, these models often fall prey to "identity mapping", where they inadvertently replicate the input image. Second, they struggle to distinguish MAs from other anomalies, leading to high false positives. Third, their suboptimal reconstruction of normal features hampers overall performance. To address these challenges, we propose a Wavelet Diffusion Transformer framework for MA Detection (WDT-MD), which features three key innovations: a noise-encoded image conditioning mechanism to avoid "identity mapping" by perturbing image conditions during training; pseudo-normal pattern synthesis via inpainting to introduce pixel-level supervision, enabling discrimination between MAs and other anomalies; and a wavelet diffusion Transformer architecture that combines the global modeling capability of diffusion Transformers with multi-scale wavelet analysis to enhance reconstruction of normal retinal features. Comprehensive experiments on the IDRiD and e-ophtha MA datasets demonstrate that WDT-MD outperforms state-of-the-art methods in both pixel-level and image-level MA detection. This advancement holds significant promise for improving early DR screening.

Diffusion-based or flow-based models have achieved significant progress in video synthesis but require multiple iterative sampling steps, which incurs substantial computational overhead. While many distillation methods that are solely based on trajectory-preserving or distribution-matching have been developed to accelerate video generation models, these approaches often suffer from performance breakdown or increased artifacts in few-step settings. To address these limitations, we propose SwiftVideo, a unified and stable distillation framework that combines the advantages of trajectory-preserving and distribution-matching strategies. Our approach introduces continuous-time consistency distillation to ensure precise preservation of ODE trajectories. Subsequently, We propose a dual-perspective alignment encompassing distribution alignment between synthetic and real data along with trajectory alignment across different inference steps. Our method maintains high-quality video generation while substantially reducing the number of inference steps. Quantitative evaluations on the OpenVid-1M benchmark demonstrate that our method significantly outperforms existing approaches in few-step video generation.

Accurately localizing and segmenting relevant objects from optical remote sensing images (ORSIs) is critical for advancing remote sensing applications. Existing methods are typically built upon moderate-scale pre-trained models and employ diverse optimization strategies to achieve promising performance under full-parameter fine-tuning. In fact, deeper and larger-scale foundation models can provide stronger support for performance improvement. However, due to their massive number of parameters, directly adopting full-parameter fine-tuning leads to pronounced training difficulties, such as excessive GPU memory consumption and high computational costs, which result in extremely limited exploration of large-scale models in existing works. In this paper, we propose a novel dynamic wavelet expert-guided fine-tuning paradigm with fewer trainable parameters, dubbed WEFT, which efficiently adapts large-scale foundation models to ORSIs segmentation tasks by leveraging the guidance of wavelet experts. Specifically, we introduce a task-specific wavelet expert extractor to model wavelet experts from different perspectives and dynamically regulate their outputs, thereby generating trainable features enriched with task-specific information for subsequent fine-tuning. Furthermore, we construct an expert-guided conditional adapter that first enhances the fine-grained perception of frozen features for specific tasks by injecting trainable features, and then iteratively updates the information of both types of feature, allowing for efficient fine-tuning. Extensive experiments show that our WEFT not only outperforms 21 state-of-the-art (SOTA) methods on three ORSIs datasets, but also achieves optimal results in camouflage, natural, and medical scenarios.

Cross-domain Few-shot Segmentation (CD-FSS) aims to segment novel classes from target domains that are not involved in training and have significantly different data distributions from the source domain, using only a few annotated samples, and recent years have witnessed significant progress on this task. However, existing CD-FSS methods primarily focus on style gaps between source and target domains while ignoring segmentation granularity gaps, resulting in insufficient semantic discriminability for novel classes in target domains. Therefore, we propose a Hierarchical Semantic Learning (HSL) framework to tackle this problem. Specifically, we introduce a Dual Style Randomization (DSR) module and a Hierarchical Semantic Mining (HSM) module to learn hierarchical semantic features, thereby enhancing the model's ability to recognize semantics at varying granularities. DSR simulates target domain data with diverse foreground-background style differences and overall style variations through foreground and global style randomization respectively, while HSM leverages multi-scale superpixels to guide the model to mine intra-class consistency and inter-class distinction at different granularities. Additionally, we also propose a Prototype Confidence-modulated Thresholding (PCMT) module to mitigate segmentation ambiguity when foreground and background are excessively similar. Extensive experiments are conducted on four popular target domain datasets, and the results demonstrate that our method achieves state-of-the-art performance.

3D human motion generation has seen substantial advancement in recent years. While state-of-the-art approaches have improved performance significantly, they still struggle with complex and detailed motions unseen in training data, largely due to the scarcity of motion datasets and the prohibitive cost of generating new training examples. To address these challenges, we introduce CoMA, an agent-based solution for complex human motion generation, editing, and comprehension. CoMA leverages multiple collaborative agents powered by large language and vision models, alongside a mask transformer-based motion generator featuring body part-specific encoders and codebooks for fine-grained control. Our framework enables generation of both short and long motion sequences with detailed instructions, text-guided motion editing, and self-correction for improved quality. Evaluations on the HumanML3D dataset demonstrate competitive performance against state-of-the-art methods. Additionally, we create a set of context-rich, compositional, and long text prompts, where user studies show our method significantly outperforms existing approaches.

Spiking Neural Networks (SNNs) offer a promising direction for energy-efficient event-based vision by leveraging sparse, temporally precise spikes. We propose a directly trained, fully spiking model for optical flow estimation, featuring a novel Spike GRU and membrane potential carryover for improved temporal modeling. On the DSEC-Flow benchmark, our model achieves competitive accuracy while reducing energy consumption by 42.88× over EV-FlowNet and 38× over TIDNet. Building on the predicted motion field, we infer camera rotation and, to the best of our knowledge, are the first to construct panoramic event images from SNN-based flow. We further introduce an optional unsupervised SO(3) refinement step that improves rotation accuracy by maximizing panorama consistency—without IMU or pose supervision. Our results achieve comparable visual quality to CMax-SLAM, showing that SNNs can enable fast and high-level spatial perception using only event-based input.

Retrieval Augmented Generation enhances the response accuracy of Large Language Models (LLMs) by integrating retrieval and generation modules with external knowledge, demonstrating particular strength in real-time queries and Visual Question Answering tasks. However, the effectiveness of RAG is frequently hindered by the precision of the retriever: many retrieved samples fed into the generation phase are irrelevant or misleading, posing a critical bottleneck to LLMs’ performance. To address this challenge, we introduce \textbf{VaccineRAG}, a novel Chain-of-Thought-based retrieval-augmented generation dataset. On one hand, VaccineRAG employs a benchmark to evaluate models using data with varying positive/negative sample ratios, systematically exposing inherent weaknesses in current LLMs. On the other hand, it enhances models’ sample-discrimination capabilities by prompting LLMs to generate explicit Chain-of-Thought (CoT) analysis for each sample before producing final answers. Furthermore, to enhance the model’s ability to learn long-sequence complex CoT content, we propose \textbf{Partial-GRPO}. By modeling the outputs of LLMs as multiple components rather than a single whole, our model can make more informed preference selections for complex sequences, thereby enhancing its capacity to learn complex CoT. Comprehensive evaluations and ablation studies on VaccineRAG validate the effectiveness of the proposed scheme.

Reconstructing fine-grained geometry of clothed human from single-view image is a challenging task, particularly in accurately recovering complex shapes and generating clothes details. To address these limitations, we propose a novel approach named HumanPro, which estimates high-quality human normals via a generative model, and progressively deforms a parametric body into the final clothed human mesh guided by normals. First, we propose a geometry-aware latent diffusion model with a normal enhancer to estimate high-quality human normals from four views. Then, we propose a progressive mesh optimization consisting of shape-aware deformation alignment and global-to-patch detail refinement for human mesh reconstruction. The shape-aware deformation alignment applies image morphing to learn the shape-level gap of normals, addressing large-scale deformation of complex clothes. It can recover the overall silhouette of a clothed human, and serves as an initialization for the global-to-patch detail refinement. Our detail refinement combines global and patch-wise optimization strategies to iteratively produce the clothed human mesh by minimizing the pixel-level difference of normals. This way effectively recovers fine-grained details while avoiding local minima. Extensive experiments demonstrate that HumanPro can deal with various challenging scenarios and outperforms state-of-the-art methods.

Most existing RGB-Event trackers rely on strictly aligned datasets, overlooking the asynchronous spatio-temporal resolutions common in real-world scenarios. This methodological limitation impedes effective RGB-Event feature alignment and ultimately degrades tracking performance. To overcome this limitation, we propose AlignTrack, a novel tracking framework built upon a Top-Down Alignment (TDA) strategy inspired by the human visual system. Our TDA framework follows an encode-decode-align paradigm: it first encodes multimodal features to generate target-related priors, which are then progressively decoded to guide a subsequent feature alignment pass. Within this framework, we introduce two key innovations: (1) a Cross-Prior Attention (CPA) module that effectively generates and integrates cross-modal priors, and (2) a Cross-Modal Semantic Alignment (CSA) loss that maximizes mutual information to enforce semantic consistency between modalities. Extensive experiments show that AlignTrack achieves state-of-the-art performance on four challenging RGB-Event tracking benchmarks, demonstrating its robustness in both aligned and unaligned scenarios. Ablation studies further validate the significant contribution of each proposed component.

Adverse weather conditions—such as rain, fog, and snow—significantly degrade LiDAR point cloud quality, causing substantial performance deterioration in detection models trained on clean data. To address this, we propose LTDNet, a novel point cloud quality improvement net-work that restores degraded LiDAR scans by learning an end-to-end mapping from corrupted to clean geometry. LTDNet leverages position encoding, spatial–frequency joint feature extraction, weather-aware refinement, and probabilistic pruning to effectively recover structural in-tegrity while suppressing weather-induced noise. To fa-cilitate standardized evaluation, we introduce IQA3D, a new benchmark comprising both synthetic and real-world sequences under adverse weather. This dual-design benchmark serves two complementary purposes: synthet-ic sequences provide pixel-wise correspondences between degraded and clean point clouds for quantitatively as-sessing restoration fidelity, while real-world sequences enable evaluation of the practical impact of improvement methods on downstream 3D object detection under au-thentic weather conditions. This makes IQA3D particular-ly suitable for jointly measuring both perceptual quality and task-level robustness of point cloud improvement models. Extensive experiments on IQA3D demonstrate that LTDNet significantly improves detection perfor-mance across various state-of-the-art 3D detectors and three tested weather conditions, making it a practical and effective solution for robust LiDAR-based detection.