AAAI 2026 论文列表

Temporal Action Detection (TAD) aims to identify specific actions in long, untrimmed videos by determining their start, end times and categories, yet existing models suffer from performance degradation under out-of-distribution scenarios due to unrealistic i.i.d. assumptions. While domain generalization (DG) offers a promising solution, image-based DG methods fail to address the unique spatiotemporal challenges in video-based TAD, including the spatiotemporal complexities and significant variations in action instance scales and densities across domains. To bridge this gap, we propose the first DG framework tailored for TAD. We propose Scene-Aware Video Segmentation, which segments videos based on semantic similarity, addressing cross-domain action instance density and scale discrepancies. Additionally, we present Temporal-Aware Normalization Perturbation to generate diverse video features while preserving temporal integrity. We establish the first DG-TAD benchmark, evaluating 11 state-of-the-art DG methods across four datasets. The experiments demonstrate that our framework consistently outperforms existing approaches, achieving superior generalization on unseen domains. The proposed modules are architecture-agnostic, offering plug-and-play compatibility for broader video understanding tasks.

Open-world 3D scene understanding is fundamentally challenging for vision and robotics, due to the constraints of closed-vocabulary supervision and static annotations. To address this, we propose a unified framework for Open-World 3D Scene Graph Generation with Retrieval-Augmented Reasoning, which enables generalizable and interactive 3D scene understanding. Our method integrates vision-language models with retrieval-based reasoning to support multimodal exploration and language-guided interaction. The framework comprises two key components: (1) a dynamic scene graph generation module that detects objects and infers semantic relationships without fixed label sets, and (2) a retrieval-augmented reasoning pipeline that encodes scene graphs into a vector database to support text/image-conditioned queries. We evaluate our method on 3DSSG and Replica benchmarks across four tasks—scene question answering, visual grounding, instance retrieval, and task planning—demonstrating robust generalization and superior performance in diverse environments. Our results highlight the effectiveness of combining open-vocabulary perception with retrieval-based reasoning for scalable 3D scene understanding.

Driven by the wave of large language models, Video-Language Models (VLMs) have become a significant yet challenging technology to bridge the gap between videos and texts. Although previous VLM works have made significant progress, almost all of them implicitly assume that all the texts are predefined by the specific template. In real-world applications, such a strict assumption is impossible to satisfy since 1) predefining all the texts is extremely time-consuming and labor-intensive. 2) these predefined text inputs are too restrictive and user-unfriendly, limiting their applications. It is observed that given a video input, texts with similar semantics but different templates lead to various performances. To this end, in this paper, we propose a novel plug-and-play framework for various VLM-based methods to fully bridge videos and texts. Specifically, we first generate positive and negative texts from the original ones to target specific text components. Then, we propose an attribute-based text reasoning strategy to mine fine-grained textual semantics of generated texts. Finally, we utilize videos as guidance to conduct cross-modal bridging by designing a self-weighted loss. Extensive experiments show that the proposed method can serve as the plug-and-play module to effectively improve the performance of state-of-the-art VLMs.

Large Language Models (LLMs) are increasingly vulnerable to adversarial prompts that exploit semantic ambiguities to bypass safety mechanisms, resulting in harmful or inappropriate outputs. Such attacks, including jailbreaking and prompt injection, pose significant risks to the integrity and availability of LLMs in security-critical applications. This paper proposes the Adversarial Prompt Disentanglement (APD) framework, a novel defense mechanism that proactively identifies and neutralizes malicious components in input prompts before they are processed by the LLM. The APD framework integrates three key innovations: (1) a mutual information- based semantic decomposition method to isolate adversarial and benign prompt components, ensuring statistical in- dependence; (2) a graph-based intent classification approach that leverages spectral analysis to detect malicious patterns in prompt semantics; and (3) a lightweight transformer-based classifier trained on real-world datasets of toxic and jailbreaking prompts, enabling efficient and accurate adversarial intent detection. Evaluated on diverse datasets containing adversarial prompts, APD demonstrates superior robustness, reducing harmful output generation by over 85% while maintaining negligible impact on model performance. The framework’s computational efficiency supports real-time deploy- ment, making it a practical solution for securing LLMs. Our work addresses critical challenges in machine learning security on novel attacks and integrity methods for ML systems, and offers a scalable, ethically grounded defense against prompt-based adversarial threats.

Large Vision-Language Models (LVLMs) have transformed multi-modal understanding, excelling in tasks like image captioning and visual question answering by integrating visual and textual inputs. However, their robustness against adversarial attacks—particularly those exploiting both modalities—remains underexplored, posing risks to critical applications like autonomous driving and content moderation. Existing attacks focus on single modalities or require impractical white-box access, limiting their real-world relevance. In this paper, we introduce Multi-Modal Adversarial Synergy (MMAS), a groundbreaking framework that crafts universal, black-box multi-modal attacks against LVLMs. MMAS simultaneously generates a texture scale-constrained Universal Adversarial Perturbation (UAP) for images and a learnable prompt perturbation for text, optimized jointly using only model queries. The image perturbation, bounded by an L∞-norm, leverages wavelet-based texture constraints to ensure imperceptibility and robustness across diverse visual inputs. The text perturbation, constrained by an L2-norm in the embedding space, maintains semantic coherence while steering outputs toward a target. A novel cross-modal regularization term aligns the perturbations’ gradient directions, enhancing their synergistic impact and transferability across tasks and models. Extensive experiments are conducted to verify the strong universal adversarial capabilities of our proposed attack with prevalent LVLMs, spanning a spectrum of tasks on various datasets, all achieved without delving into the details of the model structures.

Video-Language Models (VLMs) have demonstrated impressive multi-modal reasoning capabilities across diverse computer vision applications. However, these VLMs are task-specific and assume that both video and language inputs are complete. However, real-world VLM applications might face challenges due to deactivated sensors (e.g., cameras are unavailable due to data privacy), yielding modality-incomplete data and leading to inconsistency between training and testing data. While straightforward incomplete input can boast training generalization-ability and lead to training failure, its potential risks to VLMs regarding safety and trustworthiness have been largely neglected. To this end, we make the first attempt to propose a unified incomplete video-language model to process the incomplete multi-modal inputs. Extensive experimental results show that our method can serve as a plug-and-play module for previous works to improve their performance in various multi-modal tasks.

Image reflection separation aims to disentangle the transmission layer and the reflection layer from a blended image. Existing methods rely on limited information from a single image, tending to confuse the two layers when their contrasts are similar, a challenge more severe at night. To address this issue, we propose the Depth-Memory Decoupling Network (DMDNet). It employs the Depth-Aware Scanning (DAScan) to guide Mamba toward salient structures, promoting information flow along semantic coherence to construct stable states. Working in synergy with DAScan, the Depth-Synergized State-Space Model (DS-SSM) modulates the sensitivity of state activations by depth, suppressing the spread of ambiguous features that interfere with layer disentanglement. Furthermore, we introduce the Memory Expert Compensation Module (MECM), leveraging cross-image historical knowledge to guide experts in providing layer-specific compensation. To address the lack of datasets for nighttime reflection separation, we construct the Nighttime Image Reflection Separation (NightIRS) dataset. Extensive experiments demonstrate that DMDNet outperforms state-of-the-art methods in both daytime and nighttime.

Moving infrared small target detection (IRSTD) plays a critical role in practical applications, such as surveillance of unmanned aerial vehicles (UAVs) and UAV-based search system. Moving IRSTD still remains highly challenging due to weak target features and complex background interference. Accurate spatio-temporal feature modeling is crucial for moving target detection, typically achieved through either temporal differences or spatio-temporal (3D) convolutions. Temporal difference can explicitly leverage motion cues but exhibits limited capability in extracting spatial features, whereas 3D convolution effectively represents spatio-temporal features yet lacks explicit awareness of motion dynamics along the temporal dimension. In this paper, we propose a novel moving IRSTD network (TDCNet), which effectively extracts and enhances spatio-temporal features for accurate target detection. Specifically, we introduce a novel temporal difference convolution (TDC) re-parameterization module that comprises three parallel TDC blocks designed to capture contextual dependencies across different temporal ranges. Each TDC block fuses temporal difference and 3D convolution into a unified spatio-temporal convolution representation. This re-parameterized module can effectively capture multi-scale motion contextual features while suppressing pseudo-motion clutter in complex backgrounds, significantly improving detection performance. Moreover, we propose a TDC-guided spatio-temporal attention mechanism that performs cross-attention between the spatio-temporal features extracted from the TDC-based backbone and a parallel 3D backbone. This mechanism models their global semantic dependencies to refine the current frame’s features, thereby guiding the model to focus more accurately on critical target regions. To facilitate comprehensive evaluation, we construct a new challenging benchmark, IRSTD-UAV, consisting of 15,106 real infrared images with diverse low signal-to-clutter ratio scenarios and complex backgrounds. Extensive experiments on IRSTD-UAV and public infrared datasets demonstrate that our TDCNet achieves state-of-the-art detection performance in moving target detection.

Recent advancements in personalized Text-to-Video (T2V) generation have made significant strides in synthesizing character-specific content. However, these methods face a critical limitation: the inability to perform fine-grained control over motion intensity. This limitation stems from an inherent entanglement of action semantics and their corresponding magnitudes within coarse textual descriptions, hindering the generation of nuanced human videos and limiting their applicability in scenarios demanding high precision, such as animating virtual avatars or synthesizing subtle micro-expressions. Furthermore, existing approaches often struggle to preserve high identity fidelity when other attributes are modified. To address these challenges, we introduce MotionCharacter, a framework for high-fidelity human video generation with precise motion control. At its core, MotionCharacter explicitly decouples motion into two independently controllable components: action type and motion intensity. This is achieved through two key technical contributions: (1) a Motion Control Module that leverages textual phrases to specify the action type and a quantifiable metric derived from optical flow to modulate its intensity, guided by a region-aware loss that localizes motion to relevant subject areas; and (2) an ID Content Insertion Module coupled with an ID-Consistency loss to ensure robust identity preservation during dynamic motions. To facilitate training for such fine-grained control, we also curate Human-Motion, a new large-scale dataset with detailed annotations for both motion and facial features. Extensive experiments demonstrate that MotionCharacter achieves substantial improvements over existing methods. Our framework excels in generating videos that are not only identity-consistent but also precisely adhere to specified motion types and intensities.

Semantic scene completion simultaneously reconstructs the shapes of missing regions and predicts semantic labels for the entire 3D scene. Although point cloud-based methods are more efficient than voxel-based methods, existing point cloud-based approaches largely fail to fully leverage semantic information. To address this challenge, we propose a Prototype-Guided Transformer (ProtoFormer) that encodes semantic information into a set of semantic prototypes to guide the underlying Transformer for semantic scene completion. Specifically, we leverage semantic prototypes to enhance information from both geometric and semantic perspectives, and integrate the top-K attention mechanisms to guide scene completion and semantic awareness. Extensive qualitative and quantitative experimental results demonstrate that ProtoFormer outperforms state-of-the-art approaches with low complexity.

Segment Anything (SAM) has recently pushed the boundaries of segmentation by demonstrating remarkable zero-shot generalization and flexible prompting after training on over one billion masks. Despite this, its mask prediction accuracy often falls short of the precision required in real-world applications. While several refinement modules have been proposed to boost SAM’s segmentation quality, achieving highly accurate object delineation within a single, unified framework remains an open challenge. Furthermore, interactive image matting—which aims to generate fine-grained alpha mattes guided by diverse user hints—has not yet been explored in the context of SAM. Insights from recent studies highlight strong correlations between segmentation and matting, suggesting the feasibility of a unified model capable of both tasks. In this paper, we introduce Segment And Matte Anything (SAMA), a lightweight extension of SAM that delivers high-quality interactive image segmentation and matting with minimal extra parameters or computational cost. Our Multi-View Localization Encoder (MVLE) captures detailed features from local views, while the Localization Adapter (Local-Adapter) refines mask outputs by recovering subtle boundary details. We also incorporate two prediction heads for each task into the architecture to generate segmentation and matting tasks, simultaneously. Trained on a diverse dataset aggregated from publicly available sources, SAMA achieves state-of-the-art performance across multiple segmentation and matting benchmarks, showcasing its adaptability and effectiveness in a wide range of downstream tasks.

Deep learning models excel in visual recognition but suffer severe performance drops when training labels are corrupted by noise. Under label noise prior work cannot learn accurate similarities and thus misguide the learning process. In this paper, we uncover a complementary and novel phenomenon, Dissimilarity Invariance, whereby semantic dissimilarity between unrelated samples remains stable despite label noise. Leveraging this insight, we propose NegScale, a plug-and-play framework that shifts focus from fragile similarity to robust dissimilarity. NegScale integrates: (1) Structured Negative Orthogonality Penalty (SNOP), enforcing subspace orthogonality for unrelated samples; and (2) Dissimilarity-Calibrated Similarity Adjustment (DCSA), suppressing spurious similarity using dissimilarity anchors. We also give theoretical analysis that proves Dissimilarity Invariance and the effectiveness of NegScale. Empirical results demonstrate that NegScale consistently outperforms state-of-the-art baselines, establishing new benchmarks on CIFAR with synthetic noise and real-world datasets.

For long-tailed recognition (LTR) tasks, high intra-class compactness and inter-class separability in both head and tail classes, as well as balanced separability among all the classifier vectors, are preferred. The existing LTR methods based on cross-entropy (CE) loss not only struggle to learn features with desirable properties but also couple imbalanced classifier vectors in the denominator of its Softmax, amplifying the imbalance effects in LTR. In this paper, for the LTR, we propose a binary cross-entropy (BCE)-based tripartite synergistic learning, termed BCE3S, which consists of three components: (1) BCE-based joint learning optimizes both the classifier and sample features, which achieves better compactness and separability among features than the CE-based joint learning, by decoupling the metrics between feature and the imbalanced classifier vectors in multiple Sigmoid; (2) BCE-based contrastive learning further improves the intra-class compactness of features; (3) BCE-based uniform learning balances the separability among classifier vectors and interactively enhances the feature properties by combining with the joint learning. The extensive experiments show that the LTR model trained by BCE3S not only achieves higher compactness and separability among sample features, but also balances the classifier's separability, achieving SOTA performance on various long-tailed datasets such as CIFAR10-LT, CIFAR100-LT, ImageNet-LT, and iNaturalist2018.

Deep neural networks have demonstrated remarkable performance across various domains, yet their decision-making processes remain opaque. Although many explanation methods are dedicated to bringing the obscurity of DNNs to light, they exhibit significant limitations: post-hoc explanation methods often struggle to faithfully reflect model behaviors, while self-explaining neural networks sacrifice performance and compatibility due to their specialized architectural designs. To address these challenges, we propose a novel self-explaining framework that integrates Shapley value estimation as an auxiliary task during training, which achieves two key advancements: 1) a fair allocation of the model prediction scores to image patches, ensuring explanations inherently align with the model's decision logic, and 2) enhanced interpretability with minor structural modifications, preserving model performance and compatibility. Extensive experiments on multiple benchmarks demonstrate that our method achieves state-of-the-art interpretability.

Existing Human Motion Prediction (HMP) methods based on RGB(D) cameras are sensitive to lighting conditions and raise privacy concerns, limiting their real-world applications such as firefighting and elderly care. Motivated by the robustness and privacy-preserving nature of millimeter-wave (mmWave) radar, this work introduces radar as a novel sensing modality for HMP for the first time. Nevertheless, radar signals often suffer from specular reflections and multipath effects, resulting in noisy and temporally inconsistent measurements, such as body-part miss-detection. To address these radar-specific artifacts, we propose mmPred, the first diffusion-based framework tailored for radar-based HMP. mmPred introduces a dual-domain historical motion representation to guide the generation process, combining a Time-domain Pose Refinement (TPR) branch for fine-grained details and a Frequency-domain Dominant Motion (FDM) branch for capturing global motion trends and suppressing frame-level inconsistency. Furthermore, we design a Global Skeleton-relational Transformer (GST) as the diffusion backbone to model global inter-joint cooperation, enabling corrupted joints to dynamically aggregate information from others. Extensive experiments show that mmPred achieves state-of-the-art performance, outperforming existing methods by 8.6% on mmBody and 22% on mm-Fi.

Product posters blend striking visuals with informative text to highlight the product and capture customer attention. However, crafting appealing posters and manually optimizing them based on online performance is laborious and resource-consuming. To address this, we introduce AutoPP, an automated pipeline for product poster generation and optimization that eliminates the need for human intervention. Specifically, the generator, relying solely on basic product information, first uses a unified design module to integrate the three key elements of a poster (background, text, and layout) into a cohesive output. Then, an element rendering module encodes these elements into condition tokens, efficiently and controllably generating the product poster. Based on the generated poster, the optimizer enhances its Click-Through Rate (CTR) by leveraging online feedback. It systematically replaces elements to gather fine-grained CTR comparisons and utilizes Isolated Direct Preference Optimization (IDPO) to attribute CTR gains to isolated elements. Our work is supported by AutoPP1M, the largest dataset specifically designed for product poster generation and optimization, which contains one million high-quality posters and feedback collected from over one million users. Experiments demonstrate that AutoPP achieves state-of-the-art results in both offline and online settings.

In recent years, learning-based underwater image enhancement (UIE) techniques have rapidly evolved. However, distribution shifts between high-quality enhanced outputs and natural images can hinder semantic cue extraction for downstream vision tasks, thereby limiting the adaptability of existing enhancement models. To address this challenge, this work proposes a new learning mechanism that leverages Vision-Language Models (VLMs) to empower UIE models with semantic-sensitive capabilities. To be concrete, our strategy first generates textual descriptions of key objects from a degraded image via a VLM. Subsequently, a text-image alignment model remaps these relevant descriptions back onto the image to produce a spatial semantic guidance map. This map then steers the UIE network through a dual-guidance mechanism, which combines cross-attention and an explicit alignment loss. This forces the network to focus its restorative power on semantic-sensitive regions during image reconstruction, rather than pursuing a globally uniform improvement, thereby ensuring the faithful restoration of key object features. Experiments confirm that when our strategy is applied to different UIE baselines, significantly boosts their performance on perceptual quality metrics as well as enhances their performance on detection and segmentation tasks, validating its effectiveness and adaptability.

Saliency maps have become a cornerstone of visual explanation in deep learning, yet there remains no consensus on their intended purpose and their alignment with specific user queries. This fundamental ambiguity undermines both the evaluation and practical utility of explanation methods. In this paper, we introduce the Reference-Frame x Granularity (RFxG) taxonomy—a principled framework that addresses this ambiguity by conceptualizing saliency explanations along two essential axes: the reference-frame axis (distinguishing between pointwise "Why Husky?" and contrastive "Why Husky and not Shih-tzu?" explanations) and the granularity axis (ranging from fine-grained class-level to coarse-grained group-level interpretations, e.g., “Why Husky?” vs. “Why Dog?”). Through this lens, we identify critical limitations in existing evaluation metrics, which predominantly focus on pointwise faithfulness while neglecting contrastive reasoning and semantic granularity. To address these gaps, we propose four novel faithfulness metrics that systematically assess explanation quality across both RFxG dimensions. Our comprehensive evaluation framework spans ten state-of-the-art methods, 4 model architectures, and 3 datasets. By suggesting a shift from model-centric to user-intent-driven evaluation, our work provides both the conceptual foundation and practical tools necessary for developing explanations that are not only faithful to model behavior but also meaningfully aligned with human understanding.

The Segment Anything Model 2 (SAM2) has established a new benchmark for high-precision image and video segmentation, offering significant potential for a wide range of computer vision tasks. Despite its impressive performance, the model's substantial computational and memory requirements present a significant obstacle to its practical deployment on resource-constrained devices. In this paper, we introduce a novel framework for optimizing SAM2 through two synergistic, importance-driven strategies: quantization and memory management. Specifically, an Importance-driven Mixed-Precision Quantization scheme, which analyzes the sensitivity of each layer using a Weight-Activation Importance Score, is employed to enable a targeted bit-width assignment, preserving model accuracy by keeping critical layers at higher precision. Then, the Selective Importance-driven Synthesis (SIS) mechanism is proposed to address the inefficient accumulation of redundant data in the memory bank. SIS intelligently compresses the memory by identifying the most contextually similar historical frames and synthesizing them into a single, representative feature, thereby preserving informational diversity while enhancing temporal context understanding. Extensive experiments on the COCO and SA-V benchmarks validate our approach, showing that our optimized model consistently outperforms state-of-the-art quantization methods. Our work provides a principled framework for the co-design of quantization and dynamic memory management, offering a practical path toward deploying powerful video segmentation models in real-world applications.