AAAI 2026 Papers

Depth super-resolution has achieved impressive performance, and the incorporation of multi-frame information further enhances reconstruction quality. Nevertheless, statistical analyses reveal that video depth super-resolution remains affected by pronounced long-tailed distributions, with the long-tailed effects primarily manifesting in spatial non-smooth regions and temporal variation zones. To address these challenges, we propose a novel SpatioTemporal Difference Network (STDNet) comprising two core branches: a spatial difference branch and a temporal difference branch. In the spatial difference branch, we introduce a spatial difference mechanism to mitigate the long-tailed issues in spatial non-smooth regions. This mechanism dynamically aligns RGB features with learned spatial difference representations, enabling intra-frame RGB-D aggregation for depth calibration. In the temporal difference branch, we further design a temporal difference strategy that preferentially propagates temporal variation information from adjacent RGB and depth frames to the current depth frame, leveraging temporal difference representations to achieve precise motion compensation in temporal long-tailed areas. Extensive experimental results across multiple datasets demonstrate the effectiveness of our STDNet, outperforming existing approaches.

Fine-Grained Domain Generalization (FGDG) presents greater challenges than conventional domain generalization due to the subtle inter-class differences and relatively pronounced intra-class variations inherent in fine-grained recognition tasks. Under domain shifts, the model becomes overly sensitive to fine-grained cues, leading to the suppression of critical features and a significant drop in performance. Cognitive studies suggest that humans classify objects by leveraging both common and specific attributes, enabling accurate differentiation between fine-grained categories. However, current deep learning models have yet to incorporate this mechanism effectively. Inspired by this mechanism, we propose Concept-Feature Structuralized Generalization (CFSG). This model explicitly disentangles both the concept and feature spaces into three structured components: common, specific, and confounding segments. To mitigate the adverse effects of varying degrees of distribution shift, we introduce an adaptive mechanism that dynamically adjusts the proportions of common, specific, and confounding components. In the final prediction, explicit weights are assigned to each pair of components. Extensive experiments on three single-source benchmark datasets demonstrate that CFSG achieves an average performance improvement of 9.87% over baseline models and outperforms existing state-of-the-art methods by an average of 3.08%. Additionally, explainability analysis validates that CFSG effectively integrates multi-granularity structured knowledge and confirms that feature structuralization facilitates the emergence of concept structuralization.

Image Fusion (IF) aims to integrate complementary features from multiple source images into a single image. However, a key challenge in this field is the lack of large-scale real-world training datasets. Existing models typically rely on either small datasets or synthetic, less realistic datasets. To address this, we propose SigFusion, a unified signal-level self-supervised learning paradigm for various IF tasks.The core idea is to use signal-level Pseudo-Label Generation Networks (PLGN) to automatically synthesize training sets and pseudo labels with real multi-source signal characteristics from vast unlabeled natural images.PLGN includes two critical components: learnable 1D Signal Modulators (SM) and SigFormer. SM learns implicit 1D signal patterns across various source images and embeds them into natural images, reducing the domain gap between synthetic and real datasets. SigFormer integrates Transformer with signal processing methods, establishing an appropriate signal representation space for SM. Its cascaded, multi-level design allows hierarchical feature learning from coarse to fine detail. Moreover, SigFormer can serve as a flexible backbone for IF, as its design adheres to the classic decomposition-reconstruction paradigm. Experimental results demonstrate that SigFusion achieves state-of-the-art performance across multiple IF tasks, including medical image fusion, infrared-visible image fusion, multi-focus image fusion, and multi-exposure image fusion.

3D medical image fusion (MIF) and segmentation (MIS) are critical and inherently synergistic tasks in medical image analysis. However, fundamentally integrating them remains highly challenging, since effective collaborative paradigms are still scarce and their optimization objectives fundamentally diverge. Moreover, existing continual learning techniques are unable to achieve truly advanced performance for both tasks using a shared weight. To address these challenges, we propose M²-CoFS, a unified model capable of jointly handling both tasks. Our core contribution is a “network-guided network learning” paradigm designed to break the task boundaries. We model the weight spaces of MIF and MIS as high-dimensional manifolds and innovatively use a lightweight neural network to implicitly construct a shared manifold. Interestingly, this network yields a unified weight for both tasks. To ensure the shared manifold retains the intrinsic geometry of both original manifolds, we embed manifold distances into the loss function of this network as a constraint. Additionally, we design a tailored three-stage training paradigm for our core contribution mentioned above. Stage I focuses on independent task optimization for high-quality weights; Stage II aims to reduce parameter-space distance between tasks via our cross-task weight adaptation strategy; Our core innovation serves as stage III. Experimental results show that M²-CoFS consistently outperforms state-of-the-art comparison models on both MlF and MIS.

Generative models have become a powerful tool for synthesizing training data in computer vision tasks. Current approaches solely focus on aligning generated images with the target dataset distribution. As a result, they capture only the common features in the real dataset and mostly generate "easy samples", which are already well learned by models trained on real data. In contrast, those rare "hard samples", with atypical features but crucial for enhancing performance, cannot be effectively generated. Consequently, these approaches must synthesize large volumes of data to yield appreciable performance gains, yet the improvement remains limited. To overcome this limitation, we present a novel method that can learn to control the learning difficulty of samples during generation while also achieving domain alignment. Thus, it can efficiently generate valuable "hard samples" that yield significant performance improvements for target tasks. This is achieved by incorporating learning difficulty as an additional conditioning signal in generative models, together with a designed encoder structure and training–generation strategy. Experimental results across multiple datasets show that our method can achieve higher performance with lower generation cost. Specifically, we obtain the best performance with only 10% additional synthetic data, saving 63.4 GPU hours of generation time compared to the previous SOTA on ImageNet. Moreover, our method provides insightful visualizations of category-specific hard factors, serving as a tool for analyzing datasets.

Diffusion Transformer (DiT)-based video diffusion models generate high-quality videos at scale but incur prohibitive processing latency and memory costs for long videos. To address this, we propose a novel distributed inference strategy, termed DualParal. The core idea is that, instead of generating an entire video on a single GPU, we parallelize computation by partitioning both video frames and model layers across multiple GPUs. However, a naive parallel implementation is not feasible. Because all frames need to share the same noise level, they can't be processed independently. Instead, every step must wait for all others to finish, which cancels out the speed benefits of parallel processing. We overcome this obstacle with a block-wise denoising scheme. Namely, we segment the video into sequential blocks, each with a different noise level. As a result, we process them in a pipeline across the GPUs. Each GPU, holding a subset of the model layers, processes a specific block of frames and passes the results to the next GPU, enabling asynchronous computation and communication. To further optimize performance, we incorporate two key enhancements. Firstly, each GPU uses a feature cache technique to reduce the overhead of smooth transitions by reusing only features involved in cross-frame computation from the prior block, minimizing inter-GPU communication and redundant computation. Secondly, we employ a coordinated noise initialization strategy, ensuring globally consistent temporal dynamics by sharing initial noise patterns across GPUs. Together, these enable fast, artifact-free, and infinitely long video generation. Applied to the latest diffusion transformer video generator, our method efficiently produces 1,025-frame videos with up to 6.54x lower latency and 1.48x lower memory cost on 8xRTX 4090 GPUs.

Multimodal Large Language Models (MLLMs) have garnered significant attention recently and demonstrate outstanding capabilities in various tasks such as OCR, VQA, captioning, etc. However, hallucination remains a persistent issue. While numerous methods have been proposed to mitigate hallucinations, achieving notable improvements, these methods primarily focus on mitigating hallucinations related to object/noun concepts. Verb concepts, which are crucial for understanding human actions, have been largely overlooked. In this paper, to the best of our knowledge, we are the first to investigate the verb hallucination phenomenon of MLLMs from various perspectives. Our findings reveal that most state-of-the-art MLLMs suffer from severe verb hallucination. To assess the effectiveness of existing mitigation methods for object concept hallucination in relation to verb hallucination, we evaluated these methods and found that they do not effectively address verb hallucination. To address this issue, we propose a baseline method based on fine-tuning with rich verb knowledge, achieving decent superiority. The experiment results demonstrate that our method significantly reduces hallucinations related to verbs.

We present MCGS (Markov Chain Gaussian Splatting), a novel approach for high-fidelity dynamic scene reconstruction via combining Markov chain and 3D Gaussian splatting. Our method addresses the critical challenge of artifact-free temporal consistency in dynamic neural rendering. By integrating a Markov chain-based deformation network with multi-head temporal attention, MCGS effectively captures motion patterns and temporal dependencies, producing more accurate and stable 3D representations over time. The key innovations include: (1) a Markov Deform Network that models state transitions while preserving temporal coherence, (2) a temporal attention mechanism that adaptively weights historical states within a sliding window, and (3) strategic noise injection during training to enhance model robustness and generalization. Experiments on representative dynamic scene datasets demonstrate that MCGS outperforms previous methods in both visual quality and temporal coherence, while maintaining competitive rendering speed and efficiency. These results suggest the practical applicability of our approach to real-world dynamic scene understanding and synthesis.

Personalizing 3D scenes from a single reference image enables intuitive user-guided editing, which requires achieving both multi-view consistency across perspectives and referential consistency with the input image. However, these goals are particularly challenging due to the viewpoint bias caused by the limited perspective provided in a single image. Lacking the mechanisms to effectively expand reference information beyond the original view, existing methods of image-conditioned 3DGS personalization often suffer from this viewpoint bias and struggle to produce consistent results. Therefore, in this paper, we present Consistent Personalization for 3D Gaussian Splatting (CP-GS), a framework that progressively propagates the single-view reference appearance to novel perspectives. In particular, CP-GS integrates pre-trained image-to-3D generation and iterative LoRA fine-tuning to extract and extend the reference appearance, and finally produces faithful multi-view guidance images and the personalized 3DGS outputs through a view-consistent generation process guided by geometric cues. Extensive experiments on real-world scenes show that our CP-GS effectively mitigates the viewpoint bias, achieving high-quality image-conditioned 3DGS personalization that significantly outperforms existing methods.

Video large language models have achieved remarkable performance in tasks such as video question answering, however, their temporal understanding remains suboptimal. To address this limitation, we curate a dedicated instruction fine-tuning dataset that focuses on enhancing temporal comprehension across five key dimensions. In order to reduce reliance on costly temporal annotations, we introduce a multi-task prompt fine-tuning approach that seamlessly integrates temporal-sensitive tasks into existing instruction datasets without requiring additional annotations. Furthermore, we develop a novel benchmark for temporal-sensitive video understanding that not only fills the gaps in dimension coverage left by existing benchmarks but also rigorously filters out potential shortcuts, ensuring a more accurate evaluation. Extensive experimental results demonstrate that our approach significantly enhances the temporal understanding of video-LLMs while avoiding reliance on shortcuts.

Scene recognition (SR) is a fundamental task in computer vision (CV). In recent years, Transformer-based methods have achieved remarkable success in scene recognition tasks. Most existing approaches primarily rely on visual features, while failing to effectively model the structural relationships within scenes, which are crucial for accurate scene recognition. To this end, we propose Topology Attention Network for Scene Recognition (TANSR), an innovative method that leverages topological relationships from graphs to guide scene recognition. Specifically, Graph Attention Mask Generation Network (GAMGN) generates topology-aware masks from graph representations constructed by Graph Generation Module (GGM) and integrates them with patch embeddings by Topology Attention Guidance (TAG), enabling the transformer's attention mechanism to incorporate topological information. Furthermore, we introduce an innovative attention-driven multimodal fusion strategy that integrates graph-derived topological cues with visual patch embeddings, substantially enhancing the transformer’s capability to capture topological information and improving performance in complex scene recognition tasks. We evaluate TANSR on the benchmarks MIT-67, Scene-15 and SUN397, where it achieves consistent state-of-the-art (SOTA) performance, including 98.58% accuracy on MIT-67.

Multimodal large language models (MLLMs) frequently hallucinate by over-committing to spurious visual cues. Prior remedies–Visual and Instruction Contrastive Decoding (VCD, ICD)–mitigate this issue, yet the mechanism remains opaque. We first empirically show that their improvements systematically coincide with redistributions of cross-modal attention. Building on this insight, we propose Attention-Steerable Contrastive Decoding (ASCD), which directly steers the attention scores during decoding. ASCD combines (i) positive steering, which amplifies automatically mined text-centric heads–stable within a model and robust across domains–with (ii) negative steering, which dampens on-the-fly identified critical visual tokens. The method incurs negligible runtime/memory overhead and requires no additional training. Across five MLLM backbones and three decoding schemes, ASCD reduces hallucination on POPE, CHAIR, and MMHal-Bench by up to 38.2% while improving accuracy on standard VQA benchmarks, including MMMU, MM-VET, ScienceQA, TextVQA, and GQA. These results position attention steering as a simple, model-agnostic, and principled route to safer, more faithful multimodal generation.

3D Gaussian Splatting (3DGS) has emerged as a mainstream solution for real-time rendering and high-fidelity novel view synthesis. Building on this foundation, methods based on Textured Gaussians further improve the expression ability by incorporating explicit texture mapping into Gaussians. However, their reliance on fixed texture resolution often results in noticeable visual incoherence, triggering artifacts such as aliasing or inconsistent sharpness under different viewpoints. To address these issues, we propose PATexGS, a perceptual-adaptive texture scheduling framework designed to improve visual coherence for Textured Gaussians. Specifically, we introduce an entropy-guided texture allocation strategy that dynamically adjusts texture resolution based on each Gaussian’s spatial gradient and rendering contribution, constantly preserving details while being memory efficiency. Furthermore, we incorporate a mipmap-inspired hierarchical scheduling mechanism that adaptively schedule texture levels according to view-dependent projection scale, effectively suppressing aliasing and further enhancing perceptual consistency. Extensive experiments on diverse real-world scenes demonstrate that PATexGS significantly improves visual coherence while maintaining high rendering quality, outperforming existing TexturedGS variants in both perceptual fidelity and storage efficiency.

Magnetic Resonance Imaging (MRI) and its automatic segmentation are pivotal in assisting physicians with clinical diagnosis. In recent years, with the scarcity of labeled data, significant advancements have been made in semi-supervised segmentation. However, the prediction of many current methods is affected by the presence of false positive regions, which limits their reliability in clinical applications. To tackle this issue, we propose a pseudo-label optimization method based on polar coordinate modeling and prior constraints (PMPC), which refines false positive regions in pseudo-labels by leveraging prior knowledge within the polar coordinate system. Firstly, to improve the efficiency and rationality during polar coordinate modeling, the Adaptive Pole Selection (APS) algorithm is presented to ensure that the pole is located within the foreground region. Secondly, to mitigate false positive regions in pseudo-labels that violate medical anatomical priors, we propose the Prior Knowledge Constraint in Polar Coordinate System (KCP) module to reassign pixel categories in these regions. Finally, the Shape-aware Weighting (SaW) strategy is presented to evaluate the quality of the optimized pseudo-labels based on their shape and then determine their weight in guiding network parameter updates. Experiments on three MRI datasets demonstrate that the proposed method can be effectively integrated with existing pelvic MRI segmentation approaches, significantly reducing false positive rates and further improving segmentation quality.

3D Gaussian Splatting (3DGS) has become a powerful technique for real-time novel view synthesis, using explicit, end-to-end optimized 3D Gaussians to represent scenes. However, its training objective is primarily based on pixel-wise photometric loss, and its densification strategy fails to account for structural consistency and localized perceptual priorities. As a result, 3DGS struggles to capture fine textures and boundary details in underconstrained areas, leading to inefficient use of representational capacity and degraded rendering quality in critical regions. To overcome this limitation, we introduce TileGS, a tile-wise, perceptually guided framework designed to refine scene representation based on local rendering quality. Our method features a tile-guided densification approach that performs per-tile perceptual analysis between rendered and ground-truth tiles to identify areas and Gaussians requiring refinement. Additionally, we incorporate a tile-level structural loss to enforce localized consistency during training. TileGS is designed to be a plug-and-play framework, seamlessly integrating into existing 3DGS pipelines with minimal adjustments. Experiments across multiple datasets demonstrate that TileGS improves rendering quality while maintaining an efficient representation, showcasing its versatility and effectiveness in diverse rendering scenarios.

Frequency Modulated Continuous Wave (FMCW) radars can measure subtle chest wall oscillations to enable non-contact heartbeat sensing. However, traditional radar-based heartbeat sensing methods face performance degradation due to noise. Learning-based radar methods achieve better noise robustness but require costly labeled signals for supervised training. To overcome these limitations, we propose the first unsupervised framework for radar-based heartbeat sensing via Augmented Pseudo-Label and Noise Contrast (Radar-APLANC). We propose to use both the heartbeat range and noise range within the radar range matrix to construct the positive and negative samples, respectively, for improved noise robustness. Our Noise-Contrastive Triplet (NCT) loss only utilizes positive samples, negative samples, and pseudo-label signals generated by the traditional radar method, thereby avoiding dependence on expensive ground-truth physiological signals. We further design a pseudo-label augmentation approach featuring adaptive noise-aware label selection to improve pseudo-label signal quality. Extensive experiments on the Equipleth dataset and our collected radar dataset demonstrate that our unsupervised method achieves performance comparable to state-of-the-art supervised methods.

High-resolution Earth Observation technologies present unprecedented opportunities for geospatial analysis, yet traditional 2D aerial-view semantic segmentation remains limited by its inability to model spatial relationships and handle object occlusions. While 3D Aerial-view Segmentation (3DAS) has emerged to address these limitations, existing methods predominantly rely on 2D discriminative models pre-trained on natural scenes. These models struggle to accurately recognize aerial-view imagery, resulting in suboptimal performance due to significant domain discrepancies. This paper introduces ID-Splat, a novel object-centric framework that directly leverages multi-view object identities without discriminative information to enhance 3D semantic understanding. ID-Splat implements a two-stage process: first, Mask-object Tracking combines SAM and Point Tracking to establish robust and consistent object identities across multi-view aerial images; second, Object Integration & Propagation assigns these identities to 3D Gaussian Splatting (3DGS) points, enabling complete 3D segmentation through semantic propagation. Experimental results on the 3D-AS dataset demonstrate that ID-Splat significantly outperforms existing methods, particularly under sparse supervision conditions. ID-Splat also achieves state-of-the-art performance while reducing the need for extensive labeled data by effectively leveraging the inherent 3D structure.

The sharp, intricate contours of lunar regolith particles hold critical clues to the Moon's geological evolution and inform engineering applications from habitat construction to spacecraft design, making their precise segmentation a task of significant scientific and engineering value. However, this task exposes a weakness in deep learning models known as spectral bias, an inherent tendency to learn smooth, low-frequency functions which causes them to systematically erase the very high-frequency boundary details that are of primary interest. To resolve this conflict, we propose a framework to deeply seek object boundaries. First, we propose High-Frequency Initialized LoRA (HiFi-LoRA) to counteract spectral bias. By initializing the LoRA adaptation matrices as the optimal low-rank approximation of a high-pass filter, it fundamentally enhances the model's high-frequency perception and injects a strong preference for edges. Second, we propose the Wavelet Energy Modulation (WEM) regularizer. It guides the model to learn the intrinsic correlation between contour complexity and mask area, forcing the model to build a geometric understanding of contour morphology upon its high-frequency perception, thereby enabling the generation of boundary details commensurate with the object's scale. Experimentally, we constructed the Lunar Regolith Segmentation Dataset (LRSD), the first large-scale benchmark with expert-annotated contours. Extensive experiments demonstrate that our method sets a new state of the art on this challenging benchmark, not only achieving top performance on regional metrics like mIoU and DSC but, more critically, drastically outperforming existing models on boundary accuracy. This work not only provides a powerful computational tool for lunar science but also offers a robust and synergistic design pattern for other fine-grained segmentation challenges.

This paper aims to interactively generate and edit disentangled 3D characters based on precise user instructions. Existing methods generate and edit 3D characters via rough and simple editing guidance and entangled representations, making it difficult to achieve precise and comprehensive control over fine-grained local editing and free clothing transfer for characters. To enable accurate and intuitive control over the generation and editing of high-quality 3D characters with freely interchangeable clothing, we propose a novel user-interactive approach for disentangled 3D character creation. Specifically, to achieve precise control over 3D character generation and editing, we introduce two user-friendly interaction approaches: a sketch-based layered character generation/editing method, which supports clothing transfer; and a 3D-proxy-based part-level editing method, enabling fine-grained disentangled editing. To enhance 3D character quality, we propose a 3D Gaussian reconstruction strategy guided by geometric priors, ensuring that 3D characters exhibit detailed local geometry and smooth global surfaces. Extensive experiments on both public datasets and in-the-wild data demonstrate that our approach not only generates high-quality disentangled 3D characters but also supports precise and fine-grained editing through user interaction.