AAAI 2026 Papers

Large and high-quality motion datasets are essential for advancing human motion modeling. However, limitations of existing motion datasets, such as insufficient scale or inadequate quality, significantly hinder the progress of this field. To address these limitations, we introduce Mimic-X, a large-scale (52 hours), physically plausible 3D human motion dataset. To construct Mimic-X, we develop an adaptive option framework that controls a physically simulated character to imitate low-quality motions extracted from a vast collection of online videos. Specifically, we first apply hierarchical clustering to group motions into clusters, and then train option policies to mimic motions sampled from these clusters. Considering the noisy nature of low-quality motions, we utilize a separate encoder for each cluster to map the noisy motions within the cluster into a compact latent space. This significantly enhances the quality of the imitated motions while accelerating the learning process. Subsequently, we employ dynamic programming as a meta-policy to efficiently organize the option policies to generate complete motion clips. Finally, we perform fine-tuning to each motion sequence to further refine motion quality. The proposed adaptive option framework outperforms state-of-the-art human motion recovery methods across various evaluation metrics, demonstrating that motions in Mimic-X exhibit higher quality and greater physical plausibility. Furthermore, experimental results show that Mimic-X enhances the performance of motion generation methods, verifying its effectiveness for motion modeling tasks.

3D Gaussian Splatting (3DGS) achieves impressive quality and rendering speed, but with millions of 3D Gaussians and significant storage and transmission costs. In this paper, we aim to develop a simple yet effective method called NeuralGS that compresses the original 3DGS into a compact representation. Our observation is that neural fields like NeRF can represent complex 3D scenes with Multi-Layer Perceptron (MLP) neural networks using only a few megabytes. Thus, NeuralGS effectively adopts the neural field representation to encode the attributes of 3D Gaussians with MLPs, only requiring a small storage size even for a large-scale scene. To achieve this, we adopt a clustering strategy and fit the Gaussians within each cluster using different tiny MLPs, based on importance scores of Gaussians as fitting weights. We experiment on multiple datasets, achieving a 91$\times$ average model size reduction without harming the visual quality.

While diffusion models show promise for intent-based grasp generation, their isotropic noise schedules struggle with joint-specific sensitivity and task-aware variability. This limitation leads to grasps with suboptimal semantic alignment or physical feasibility. To address this challenge, we propose Semantic-guided Noise Scaling for grasp generation (SNS-Grasp), a novel framework that integrates two key innovations. First, the Semantic-guided Noise Scaling Diffusion (SNS-Diff) module generates intent-aware grasps by replacing isotropic noise with anisotropic modulation, dynamically adapting to task semantics and joint-specific sensitivity. Specifically, SNS-Diff leverages a pretrained Intent Recognizer to extract task-aware confidence scores and joint-specific gradient sensitivities from the interaction context. These signals adjust the noise scaling during denoising, downweighting perturbations for semantically critical joints to ensure semantic alignment. Second, the Fine-grained Grasp Refinement (FGR) module establishes dynamic joint-vertex coupling through fine-grained hand-object spatial relationships, enabling iterative optimization of physically executable grasps. Extensive experiments on OakInk and GRAB demonstrate SNS-Grasp's superior performance in semantic accuracy and physical feasibility, with robust generalization to unseen objects.

The rapid progress of multi-modal large language models (MLLMs) has boosted the task of image quality assessment (IQA). However, a key challenge arises from the inherent mismatch between the discrete token outputs of MLLMs and the continuous nature of quality scores required by IQA tasks. This discrepancy significantly hinders the performance of MLLM-based IQA methods. Previous approaches that convert discrete token predictions into continuous scores often suffer from conversion errors. Moreover, the semantic confusion introduced by level tokens (e.g., “good”) further constrains the performance of MLLMs on IQA tasks and degrades their original capabilities to related tasks. To tackle these problems, we provide a theoretical analysis of the errors inherent in previous approaches and, motivated by this analysis, propose a simple yet effective framework, Q-Scorer. This framework incorporates a lightweight regression module and IQA-specific score tokens into the MLLM pipeline. Extensive experiments demonstrate that Q-Scorer achieves state-of-the-art performance across multiple IQA benchmarks, generalizes well to mixed datasets, and further improves combined with other methods.

Zero-shot Composed Image Retrieval (ZS-CIR) involves diverse tasks with varied visual manipulation intents across domains, scenes, objects, and attributes. A key challenge is that existing datasets contain limited intent-relevant annotations, making it hard for models to infer human intent from textual modifications. We introduce an intent-centric image–text dataset generated via reasoning by a Multimodal Large Language Model (MLLM) to better train ZS-CIR models for human manipulation intent understanding. Building on this dataset, we propose De-MINDS, a framework that distills the MLLM’s reasoning ability to capture manipulation intent and enhance models’ comprehension of modified text. A simple mapping network translates image information into language space and combines it with the manipulation text to form a query. De-MINDS then extracts intention-relevant information from this query and encodes it as pseudo-word tokens for accurate ZS-CIR. Across four ZS-CIR tasks, De-MINDS shows strong generalization and improves over existing methods by 2.15% to 4.05%, establishing new state-of-the-art results with comparable inference time.

Since high-fidelity reference images are difficult to obtain in real underwater scenes, most deep models trained by synthetic paired data cannot match real-world data exactly. In this paper, we propose an unsupervised training framework for underwater image enhancement (UIE) by leveraging an iterative training strategy and quantification of specific neural units. Specifically, to eliminate the heavy color cast and distortion in the underwater images, we decompose the unsupervised image enhancement as two targeted sub-tasks, namely colorization and color compensation. First, a diffusion model is introduced for colorization to correct the green and blue color casts. Then, to intensify the learning ability of balanced color information, we introduce an extra network branch and propose a quantification mechanism for color compensation. The extra branch encodes style information from normal images into the generative model, while the quantification mechanism identifies and adjusts neural units relevant to warm colors, improving the model’s ability to learn balanced color feature representations for robust generation. In the end, through iterative training, color cast and distortion are progressively reduced, leading to a gradual improvement in the quality of the generated images. Experimental results on various widely used underwater datasets demonstrate that our approach achieves excellent performance, even when compared to recent supervised methods.

Diffusion models have revealed powerful potential in all-in-one image restoration (AiOIR), which is talented in generating abundant texture details. The existing AiOIR methods either retrain a diffusion model or fine-tune the pretrained diffusion model with extra conditional guidance. However, they often suffer from high inference costs and limited adaptability to diverse degradation types. In this paper, we propose an efficient AiOIR method, Diffusion Once and Done (DOD), which aims to achieve superior restoration performance with only one-step sampling of Stable Diffusion (SD) models. Specifically, multi-degradation feature modulation is first introduced to capture different degradation prompts with a pretrained diffusion model. Then, parameter-efficient conditional low-rank adaptation integrates the prompts to enable the fine-tuning of the SD model for adapting to different degradation types. Besides, a high-fidelity detail enhancement module is integrated into the decoder of SD to improve structural and textural details. Experiments demonstrate that our method outperforms existing diffusion-based restoration approaches in both visual quality and inference efficiency.

Large vision-language models (LVLMs) have demonstrated impressive capabilities across diverse multimodal tasks, yet they remain highly susceptible to visual hallucinations (VH), often producing confident but inaccurate descriptions of visual content. Building on the insight that not all tokens and attention heads contribute equally to VH mitigation, we introduce VisFlow, a lightweight and training-free framework that alleviates hallucinations by directly modulating attention patterns during inference. To address two primary challenges of VH, namely insufficient visual attention and the dominance of language priors, we identify three problematic attention behaviors in LVLMs: (1) disproportionate allocation of attention to uninformative or trailing visual tokens, (2) over-dependence on the previously generated token, and (3) excessive fixation on system prompts that hinders multimodal integration. To overcome these issues, VisFlow introduces a dual-level Attention Intervention, consisting of Token-level Attention Intervention (TAI), which reinforces attention to salient visual regions, and Head-level Attention Intervention (HAI), which suppresses undue focus on system prompts and adjacent text tokens. Together, these interventions strengthen visual alignment while reducing linguistic bias. Extensive experiments across diverse models and benchmarks demonstrate that VisFlow effectively mitigates hallucinations with minimal computational overhead.

Most adversarial attacks on point clouds perturb a large number of points, causing widespread geometric changes and limiting applicability in real-world scenarios. While recent works explore sparse attacks by modifying only a few points, such approaches often struggle to maintain effectiveness due to the limited influence of individual perturbations. In this paper, we propose SCP, a sparse and cooperative perturbation framework that selects and leverages a compact subset of points whose joint perturbations produce amplified adversarial effects. Specifically, SCP identifies the subset where the misclassification loss is locally convex with respect to their joint perturbations, determined by checking the positive-definiteness of the corresponding Hessian block. The selected subset is then optimized to generate high-impact adversarial examples with minimal modifications. Extensive experiments show that SCP achieves 100% attack success rates, surpassing state-of-the-art sparse attacks, and delivers superior imperceptibility to dense attacks with far fewer modifications.

Multimodal Large Language Models struggle to maintain reliable performance under extreme real-world visual degradations, which impede their practical robustness. Existing robust MLLMs predominantly rely on implicit training/adaptation that focuses solely on visual encoder generalization, suffering from limited interpretability and isolated optimization. To overcome these limitations, we propose Robust-R1, a novel framework that explicitly models visual degradations through structured reasoning chains. Our approach integrates: (i) supervised fine-tuning for degradation-aware reasoning foundations, (ii) reward-driven alignment for accurately perceiving degradation parameters, and (iii) dynamic reasoning depth scaling adapted to degradation intensity. To facilitate this approach, we introduce a specialized 11K dataset featuring realistic degradations synthesized across four critical real-world visual processing stages, each annotated with structured chains connecting degradation parameters, perceptual influence, pristine semantic reasoning chain, and conclusion. Comprehensive evaluations demonstrate state-of-theart robustness: Robust-R1 outperforms all general and robust baselines on the real-world degradation benchmark R-Bench, while maintaining superior anti-degradation performance under multi-intensity adversarial degradations on MMMB, MMStar, and RealWorldQA.

Modular design of planning-oriented autonomous driving has markedly advanced end-to-end systems. However, existing architectures remain constrained by an over-reliance on ego status, hindering generalization and robust scene understanding. We identify the root cause as an inherent design within these architectures that allows ego status to be easily leveraged as a shortcut. Specifically, the premature fusion of ego status in the upstream BEV encoder allows an information flow from this strong prior to dominate the downstream planning module. To address this challenge, we propose AdaptiveAD, an architectural-level solution based on a multi-context fusion strategy. Its core is a dual-branch structure that explicitly decouples scene perception and ego status. One branch performs scene-driven reasoning based on multi-task learning, but with ego status deliberately omitted from the BEV encoder, while the other conducts ego-driven reasoning based solely on the planning task. A scene-aware fusion module then adaptively integrates the complementary decisions from the two branches to form the final planning trajectory. To ensure this decoupling does not compromise multi-task learning, we introduce a path attention mechanism for ego-BEV interaction and add two targeted auxiliary tasks: BEV unidirectional distillation and autoregressive online mapping. Extensive evaluations on the nuScenes dataset demonstrate that AdaptiveAD achieves state-of-the-art open-loop planning performance. Crucially, it significantly mitigates the over-reliance on ego status and exhibits impressive generalization capabilities across diverse scenarios.

Current Zero-Shot Temporal Action Localization (ZSTAL) methods, whether training-based or training-free ones, still predominantly rely on a single, unified query to localize an entire action. This unified representation is fundamentally ill-suited for complex real-world activities, as it fails to capture their internal compositional structure and adapt to dynamic, multi-stage variations across videos. To address this, we regard ZSTAL as a compositional reasoning task and introduce CASCADE, a Context-Aware Staged Action DEcomposition framework. Inspired by the human cognitive process of perceiving context, decomposing events, and reconstructing instances, CASCADE follows a training-free pipeline. It first perceives the video's context by leveraging a Multimodal Large Language Model (MLLM) to both filter out irrelevant actions and then generate a rich, video-specific caption for each action present in the video. An LLM then decomposes this caption into multiple, temporally ordered stages, which serve as fine-grained queries to guide the MLLM in estimating frame-level confidence scores. Recognizing that this decomposition can fragment a single action, a novel hierarchical merging logic then reconstructs complete instances by intelligently fusing these preliminary temporal segments based on their semantic progression and coherence. Extensive experiments and ablation studies on THUMOS14 and ActivityNet-1.3 show that CASCADE not only sets a new state-of-the-art among training-free methods but, most notably, significantly outperforms all prior training-based approaches on ActivityNet-1.3.

Recent advances in Multi-modal Large Language Models (MLLMs) have showcased remarkable capabilities in vision-language understanding. However, enabling robust video spatial reasoning—the ability to comprehend object locations, orientations, and inter-object relationships in dynamic 3D scenes—remains a key unsolved challenge. Existing approaches primarily rely on spatially grounded supervised fine-tuning or reinforcement learning, yet we observe that such models often exhibit query-locked reasoning, focusing narrowly on objects explicitly mentioned in the prompt while ignoring critical contextual cues. To address this limitation, we propose Object-Centric 3D Rollout (OCR), a novel strategy that introduces structured perturbations to the 3D geometry of selected objects during training. By degrading object-specific visual cues and projecting the altered geometry into 2D space, OCR compels the model to reason holistically across the entire scene. We further design a rollout-based training pipeline that jointly leverages vanilla and region-noisy videos to optimize spatial reasoning trajectories. Experiments demonstrate state-of-the-art performance: our 3B-parameter model achieves 47.5% accuracy on VSI-Bench, outperforming several 7B baselines. Ablations confirm OCR’s superiority over prior rollout strategies (e.g., T-GRPO, NoisyRollout).

Composed Image Retrieval (CIR) enables image search by combining a reference image with modification text. Intrinsic noise in CIR triplets incurs intrinsic uncertainty and threatens model's robustness. Probabilistic learning approaches have shown promise in addressing such issues; however, they fall short for CIR due to their instance-level holistic modeling and homogeneous treatments for queries and targets. This paper introduces a Heterogeneous Uncertainty-Guided (HUG) paradigm to overcome these limitations. HUG utilizes a fine-grained probabilistic learning framework, where queries and targets are represented by Gaussian embeddings capturing detailed concepts and uncertainties. We customize heterogeneous uncertainty estimations for multi-modal queries and uni-modal targets. Given a query, we capture uncertainties not only regarding uni-modal content quality but also multi-modal coordination, followed by a provable dynamic weighting mechanism to derive the comprehensive query uncertainty. We further design uncertainty-guided objectives, including query-target holistic contrast and fine-grained contrasts with comprehensive negative sampling strategies, which effectively enhance discriminative learning. Experiments on benchmarks demonstrate HUG's effectiveness beyond state-of-the-art baselines, with faithful analysis justifying the technical contributions.

Efficient reference structures are essential in video compression, enabling the exploitation of temporal dependencies across frames to reduce redundancy. In this paper, we delve into the inter-frame reference management mechanism in neural video codecs (NVCs). Previous schemes have inherited the reference propagation mechanism with the guidance of predefined reference structure, but the reference modeling across diverse reference sources remains underexplored. Moreover, the mismatch between the reference structure used for motion estimation and motion compensation limits the effectiveness of inter-frame prediction. To address the above limitations, we propose the unified reference hierarchy that integrates a learned hierarchical reference structure into the existing inherent reference propagation mechanism. Specifically, we first propose the hierarchical reference structure (HRS) to manage the multiple temporal contexts in the propagated reference feature, where a hierarchy-aware reference modulation module is integrated to select the most relevant reference features across different quality levels under the guidance of the reference balance loss. In addition, we propose the HRS-guided feature-wise inter-frame prediction that learns the low-rank approximation of the selected reference feature for ensuring the consistency and improving the inter-frame prediction performance. We conduct experiments on a state-of-the-art NVC, DCVC-DC. Experimental results show that our codec achieves an average 26% bitrate saving over H.266/VVC, and a 28.2% bitrate reduction compared to DCVC-DC without increasing the decoding complexity.

Multimodal Large Language Models (MLLMs) have demonstrated significant progress in vision-language tasks, yet they still face challenges when processing long-duration video inputs. The limitation arises from MLLMs' context limit and training costs, necessitating sparse frame sampling before feeding videos into MLLMs. However, building a trainable sampling method remains challenging due to the unsupervised and non-differentiable nature of sparse frame sampling in Video-MLLMs. To address these problems, we propose Temporal Sampling Policy Optimization (**TSPO**), advancing MLLMs' long-form video-language understanding via reinforcement learning. Specifically, we first propose a trainable event-aware temporal agent, which captures event-query correlation for performing probabilistic keyframe selection. Then, we propose the TSPO reinforcement learning paradigm, which models keyframe selection and language generation as a joint decision-making process, enabling end-to-end group relative optimization for the temporal sampling policy. Furthermore, we propose a dual-style long video training data construction pipeline, balancing comprehensive temporal understanding and key segment localization. Finally, we incorporate rule-based answering accuracy and temporal locating reward mechanisms to optimize the temporal sampling policy. Comprehensive experiments show that our TSPO achieves state-of-the-art performance across multiple long video understanding benchmarks, and shows transferable ability across different cutting-edge Video-MLLMs.

Underwater image enhancement (UIE) aims to address image degradation caused by water absorption and scattering effects. Despite significant progress in deep learning-based UIE methods, existing approaches still face key challenges due to the neglect of physical imaging principle. Moreover, while current Mamba models achieve global modeling via multi-directional scanning, their local sequential strategy lacks sufficient global context. To this end, we propose a novel Physical Model-Guided Global Mamba (PGMamba) that combines the efficient sequential modeling capability of Mamba with underwater imaging physical model. Specifically, we first design a Spatial-Aware Global Mamba (SAGMamba) that achieves efficient long-range dependency modeling through a spatial-aware ranking strategy with global context information. Second, we develop a Physical Model-Guided Feed-Forward Network (PMGFFN) that explicitly incorporates underwater optical imaging principles into the network architecture. Extensive experimental results and comprehensive ablation studies demonstrate the outstanding performance and importance of our proposed method.

Cross-modal hashing (CMH) is an effective tool for large-scale retrieval due to its low storage cost and high efficiency. However, real-world multi-modal datasets often contain noisy annotations, which can significantly impair model performance. Many existing methods address this issue by using the small-loss criterion to select a likely clean subset of data to guide model training. Nonetheless, this clean subset is typically dominated by easy samples, and treating all samples within it equally can undermine the model’s generalization ability. In this paper, we propose a novel meta-learning-based framework, named Meta-Guided Sample Reweighting for Cross-Modal Hashing Retrieval (MGSH), which integrates meta-learning into robust cross-modal hashing. To address the above issues, we design a Meta-Similarity Weighting Network (MSWN) that dynamically assigns importance weights to samples during training. By employing a bi-level optimization strategy, the meta-importance weights are used to scale the loss of training samples during the main network update, encouraging the model to focus on more challenging examples. Additionally, to further distinguish between noisy and clean samples, we incorporate adaptive-margin and meta-guided center aggregation into a robust hashing loss, both guided by the learned meta-importance weights. Extensive experiments on three widely used benchmark datasets demonstrate that MGSH consistently outperforms state-of-the-art methods, validating its effectiveness.

Diffusion models have emerged as state-of-the-art generative methods, particularly excelling in conditional tasks such as prompt-driven image synthesis. While recent research emphasizes the pivotal role of noise seeds in enhancing text-image alignment and generating human-preferred outputs,these works predominantly rely on random Gaussian noise or heuristic local adjustments, , overlooking the potential of global optimization trategies to systematically improve generation quality. To bridge this gap, we propose Seed Optimization based on Evolution (SOE), a hybrid framework that integrates global evolutionary search with local semantic refinement. The global evolutionary stage conducts seed selection by jointly optimizing text-image alignment (via CLIP-Score) and human preference estimation (via ImageReward), while the local stage employs diffusion inversion to inject conditional semantics into the noise seed. Together, these components constitute a model-agnostic, training-free optimization framework for conditional diffusion models. Extensive experiments across various diffusion models demonstrate that SOE consistently improves semantic fidelity and visual quality, highlighting its generalizability and potential as a plug-and-play enhancement for generative diffusion pipelines.