AAAI 2026 Papers

Causal inference has emerged as a promising approach for identifying decisive semantic factors and eliminating spurious correlations in visual representation learning. However, most existing methods rely on latent, data-driven confounder modeling, normally attributing the source of bias to background information while neglecting object-level semantic confusions that commonly occur in complex scenes. This limits their effectiveness in disentangling causal factors from confounding semantics. To address this challenge, we propose an explicit modeling approach for both causal factors and confounders, termed Explicit Modeling Causal Model (EMCM). The proposed framework consists of three key components. The Features Stability Estimation module explicitly models the relationship between visual semantics and class labels by leveraging clustering patterns to perform class-aware separation of causal and confounding factors. It produces class-specific causal factors and confounding factors linked to ambiguous categories. Subsequently, the Discriminative Features Enhancing module integrates causal factors into fused patch features via front-door intervention for stable semantics. In parallel, the Explicit Confounder Modeling and Debiasing Module learns confounders under clear label guidance and derives debiased context features by TDE modeling. This framework leverages two complementary causal perspectives to construct a unified semantic representation that facilitates improved generalization. Extensive experiments on two datasets demonstrate that EMCM effectively disentangles causal and confounding factors in complex scenarios, consistently outperforming state-of-the-art causal debiasing methods and text-guided methods in all metrics.

Video-based human pose estimation has vast applications such as action recognition, sports analytics, and crime detection. However, this task is challenging as it involves interpreting both spatial context and temporal dynamics to accurately localize human anatomical keypoints in video sequences. Current approaches, often based on attention mechanisms, perform well but struggle in challenging scenarios like rapid motion and pose occlusion. We attribute these failures to two fundamental limitations: spatial uniformity, where models indiscriminately assign attention to both joint-relevant features and background clutter, thereby introducing spatial noise; and temporal rigidity, an inability to adapt to large joint displacements, resulting in severe feature misalignment during rapid motion. To overcome these challenges, we introduce PSTPose, a novel progressive spatiotemporal refinement framework. Specifically, to address the spatial uniformity problem, we propose a Discriminative Feature Enhancement (DFE) module that emphasizes joint-relevant features and a Feature Cluster Grouping (FCG) module that forms compact, semantically meaningful regions. For the temporal rigidity problem, we introduce a Deformable Spatiotemporal Fusion (DSF) module that adaptively aligns features across consecutive frames via deformation-aware sampling. This design ensures robust keypoint localization, particularly in cluttered and dynamic scenes. Extensive experiments on three large-scale benchmarks, PoseTrack2017, PoseTrack2018, PoseTrack21, demonstrate that PSTPose establishes a new state-of-the-art.

Video-based human pose estimation aims to localize keypoints across frames, enabling robust analysis of human motion in applications such as sports, surveillance, and healthcare. However, existing methods rely solely on visual cues, limiting their robustness in complex scenes involving occlusion, motion blur, or poor lighting. In contrast, dual coding theory from psychology suggests that human cognition is inherently multimodal: we learn by integrating visual perception with linguistic context to form structured, semantic understandings of the world. Visual input provides concrete spatiotemporal grounding, while language offers symbolic abstraction that enhances reasoning and generalization. Motivated by this cognitive principle, we present the first framework that explicitly incorporates language as an auxiliary modality to enhance video-based pose estimation. To address the lack of paired video-text datasets, we first employ a Multimodal Large Language Model (MLLM) to generate textual descriptions of human interactions from videos. We then propose a novel coarse-to-fine multimodal alignment pipeline: a cross-modal semantic interaction module establishes initial grounding between spatiotemporal visual features and textual embeddings, while an optimal transport-based feature matching mechanism enforces fine-grained, geometry-aware alignment. This cognitively inspired design enables more accurate and robust pose estimation, especially in visually challenging scenes like occlusion and motion blur. Extensive experiments on three benchmarks confirm that our method consistently outperforms state-of-the-art approaches.

Unsupervised industrial anomaly detection requires accurately identifying defects without labeled data. Traditional autoencoder-based methods often struggle with incomplete anomaly suppression and loss of fine details, as their single-pass decoding fails to effectively handle anomalies with varying severity and scale. We propose a recursive architecture for autoencoder (RcAE), which performs reconstruction iteratively to progressively suppress anomalies while refining normal structures. Unlike traditional single-pass models, this recursive design naturally produces a sequence of reconstructions, progressively exposing suppressed abnormal patterns. To leverage this reconstruction dynamics, we introduce a Cross Recursion Detection (CRD) module that tracks inconsistencies across recursion steps, enhancing detection of both subtle and large-scale anomalies. Additionally, we incorporate a Detail Preservation Network (DPN) to recover high-frequency textures typically lost during reconstruction. Extensive experiments demonstrate that our method significantly outperforms existing non-diffusion methods, and achieves performance on par with recent diffusion models with only 10% of their parameters and offering substantially faster inference. These results highlight the practicality and efficiency of our approach for real-world applications.

Recent 4D reconstruction methods have yielded impressive results but rely on sharp videos as supervision. However, motion blur often occurs in videos due to camera shake and object movement, while existing methods render blurry results when using such videos for reconstructing 4D models. Although a few approaches attempted to address the problem, they struggled to produce high-quality results, due to the inaccuracy in estimating continuous dynamic representations within the exposure time. Encouraged by recent works in 3D motion trajectory modeling using 3D Gaussian Splatting (3DGS), we take 3DGS as the scene representation manner, and propose Deblur4DGS to obtain a high-quality 4D model from blurry monocular video. Specifically, we transform continuous dynamic representations estimation within an exposure time into the exposure time estimation. Moreover, we introduce the exposure regularization term, multi-frame, and multi-resolution consistency regularization term to avoid trivial solutions. Furthermore, to better represent objects with large motion, we suggest blur-aware variable canonical Gaussians. Beyond novel-view synthesis, Deblur4DGS can be applied to improve blurry video from multiple perspectives, including deblurring, frame interpolation, and video stabilization. Extensive experiments in both synthetic and real-world data on the above four tasks show that Deblur4DGS outperforms state-of-the-art 4D reconstruction methods.

Incremental Object Detection (IOD) aims to continuously learn new object classes without forgetting previously learned ones. A persistent challenge is catastrophic forgetting, primarily attributed to background shift in conventional detectors. While pseudo-labeling mitigates this in dense detectors, we identify a novel, distinct source of forgetting specific to DETR-like architectures: background foregrounding. This arises from the exhaustiveness constraint of the Hungarian matcher, which forcibly assigns every ground truth target to one prediction, even when predictions primarily cover background regions (i.e., low IoU). This erroneous supervision compels the model to misclassify background features as specific foreground classes, disrupting learned representations and accelerating forgetting. To address this, we propose a Quality-guided Min-Cost Max-Flow (Q-MCMF) matcher. To avoid forced assignments, Q-MCMF builds a flow graph and prunes implausible matches based on geometric quality. It then optimizes for the final matching that minimizes cost and maximizes valid assignments. This strategy eliminates harmful supervision from background foregrounding while maximizing foreground learning signals. Extensive experiments on the COCO dataset under various incremental settings demonstrate that our method consistently outperforms existing state-of-the-art approaches.

Ring artifacts are prevalent in 3D cone-beam computed tomography (CBCT) due to non-ideal responses of X-ray detectors, substantially affecting image quality and diagnostic reliability. Existing state-of-the-art (SOTA) ring artifact reduction (RAR) methods rely on supervised learning with large-scale paired CT datasets. While effective in-domain, supervised methods tend to struggle to fully capture the physical characteristics of ring artifacts, leading to pronounced performance drops in complex real-world acquisitions. Moreover, their scalability to 3D CBCT is limited by high memory demands. In this work, we propose Riner, a new unsupervised RAR method. Based on a theoretical analysis of ring artifact formation, we reformulate RAR as a multi-parameter inverse problem, where the non-ideal responses of X-ray detectors are parameterized as solvable physical variables. Using a new differentiable forward model, Riner can jointly learn the implicit neural representation of artifact-free images and estimate the physical parameters directly from CT measurements, without external training data. Additionally, Riner is memory-friendly due to its ray-based optimization, enhancing its usability in large-scale 3D CBCT. Experiments on both simulated and real-world datasets show Riner outperforms existing SOTA supervised methods.

Video generation models have achieved remarkable progress, particularly excelling in realistic scenarios; however, their performance degrades notably in imaginative scenarios. These prompts often involve rarely co-occurring concepts with long-distance semantic relationships, falling outside training distributions. Existing methods typically apply test-time scaling for improving video quality, but their fixed search spaces and static reward designs limit adaptability to imaginative scenarios. To fill this gap, we propose ImagerySearch, a dynamic test-time scaling law strategy inspired by imagery that adaptively adjusts the inference search space and reward guided by prompts, effectively enhancing generation quality in imaginative scenarios. Furthermore, we introduce LDT-Bench, the first benchmark targeting long-distance semantic prompts, designed to evaluate the creativity of video generation models. It comprises 2,839 challenging concept pairs from diverse recognition datasets and incorporates an automatic evaluation protocol to assess creative capacity. Extensive experiments on LDT-Bench demonstrate that our approach consistently outperforms general generation models and test-time scaling approaches. Additionally, ImagerySearch achieves strong performance on VBench, confirming its effectiveness in improving video generation quality under diverse conditions.

Cross-view geo-localization (CVGL) aims to accurately localize street-view images through retrieval of corresponding geo-tagged satellite images. While prior works have achieved nearly perfect performance on certain standard datasets, their robustness in real-world corrupted environments remains under-explored. This oversight causes severe performance degradation or failure when images are affected by corruption such as blur or weather, significantly limiting practical deployment. To address this critical gap, we introduce MRGeo, the first systematic method designed for robust CVGL under corruption. MRGeo employs a hierarchical defense strategy that enhances the intrinsic quality of features and then enforces a robust geometric prior. Its core is the Spatial-Channel Enhancement Block, which contains: (1) a Spatial Adaptive Representation Module that models global and local features in parallel and uses a dynamic gating mechanism to arbitrate their fusion based on feature reliability; and (2) a Channel Calibration Module that performs compensatory adjustments by modeling multi-granularity channel dependencies to counteract information loss. To prevent spatial misalignment under severe corruption, a Region-level Geometric Alignment Module imposes a geometric structure on the final descriptors, ensuring coarse-grained consistency. Comprehensive experiments on both robustness benchmark and standard datasets demonstrate that MRGeo not only achieves an average R@1 improvement of 2.92% across three comprehensive robustness benchmarks (CVUSA-C-ALL, CVACT_val-C-ALL, and CVACT_test-C-ALL) but also establishes superior performance in cross-area evaluation, thereby demonstrating its robustness and generalization capability.

All-in-one image restoration (AIR) aims to address diverse degradations within a unified model by leveraging informative degradation conditions to guide the restoration process. However, existing methods often rely on implicitly learned priors, which may entangle feature representations and hinder performance in complex or unseen scenarios. Histogram of Oriented Gradients (HOG) as a classical gradient representation, we observe that it has strong discriminative capability across diverse degradations, making it a powerful and interpretable prior for AIR. Based on this insight, we propose HOGformer, a Transformer-based model that integrates learnable HOG features for degradation-aware restoration. The core of HOGformer is a Dynamic HOG-aware Self-Attention (DHOGSA) mechanism, which adaptively models long-range spatial dependencies conditioned on degradation-specific cues encoded by HOG descriptors. To further adapt the heterogeneity of degradations in AIR, we propose a Dynamic Interaction Feed-Forward (DIFF) module that facilitates channel–spatial interactions, enabling robust feature transformation under diverse degradations. Besides, we propose a HOG loss to explicitly enhance structural fidelity and edge sharpness. Extensive experiments on a variety of benchmarks, including adverse weather and natural degradations, demonstrate that HOGformer achieves state-of-the-art performance and generalizes well to complex real-world scenarios.

While existing underwater image compression (UIC) methods optimize for human perception or basic redundancies, they neglect inter-image correlations and fail to prioritize machine-friendly features essential for automated analysis. This paper introduces a novel -quantized (VQ) codebook-driven framework for machine-centric UIC. We leverage VQ codebooks -- pre-trained as external priors on diverse underwater data -- to unify three critical stages: (1) Machine-friendly feature extraction via contrastive learning with high/low-quality codebooks, enhancing degradation robustness; (2) Compact compression using variable-size codebooks to map discriminative features to entropy-coded indices, enabling ultra-low bitrates (less than 0.04bpp); and (3) Feature refinement at the decoder, restoring semantic fidelity for downstream tasks. In addition, we contribute the first Underwater Visual Question Answering (UVQA) benchmark to holistically evaluate machine perception across object presence, counting, and localization. Extensive experiments demonstrate that our framework significantly outperforms state-of-the-art codecs in machine vision task performance at ultra-low bitrates. The VQ-codebook effectively harnesses inter-image redundancy, combats joint degradation, and delivers compact, analysis-friendly representations, establishing a new paradigm for machine-centric UIC.

With the exponential growth of video traffic, traditional video streaming systems are approaching their limits in communication capacity. To further reduce bitrate while maintaining quality, we propose Promptus, a disruptive semantic communication system that streams prompts instead of videos. Promptus represents the real-world video with a series of "prompts" for delivery and employs Stable Diffusion to generate the same video at the receiver. To ensure that the generated video is pixel-aligned with the original video, a gradient descent-based prompt fitting framework is proposed. Further, a low-rank decomposition-based bitrate control algorithm is introduced to achieve adaptive bitrate. For inter-frame compression, an interpolation-aware fitting algorithm is proposed. Evaluations across various video genres demonstrate that, compared to H.265, Promptus can achieve more than a 4x bandwidth reduction while preserving the same perceptual quality. On the other hand, at extremely low bitrates, Promptus can enhance the perceptual quality by 0.139 and 0.118 (in LPIPS) compared to VAE and H.265, respectively, and decreases the ratio of severely distorted frames by 89.3% and 91.7%. Our work opens up a new paradigm for efficient video communication.

Reconstructing human avatars using generative priors is essential for achieving versatile and realistic avatar models. Traditional approaches often rely on volumetric representations guided by generative models, but these methods require extensive volumetric rendering queries, leading to slow training. Alternatively, surface-based representations offer faster optimization through differentiable rasterization, yet they are typically limited by vertex count, restricting mesh resolution and scalability when combined with generative priors. Moreover, integrating generative priors into physically based human avatar modeling remains largely unexplored. To address these challenges, we introduce DIS (Deep Inverse Shading), a unified framework for high-fidelity, relightable avatar reconstruction that incorporates generative priors into a coherent surface representation. DIS centers on a mesh-based model that serves as the target for optimizing both surface and material details. The framework fuses multi-view 2D generative surface normal predictions, rich in detail but often inconsistent, into the central mesh using a normal conversion module. This module converts generative normal outputs into per-triangle surface offsets via differentiable rasterization, enabling the capture of fine geometric details beyond sparse vertex limitations. Additionally, DIS integrates a de-shading module, informed by generative priors, to recover accurate material properties such as albedo. This module refines albedo predictions by removing baked-in shading and back-propagates reconstruction errors to further optimize the mesh geometry. Through this joint optimization of geometry and material appearance, DIS achieves physically consistent, high-quality reconstructions suitable for accurate relighting. Our experiments show that DIS delivers SOTA relighting quality, enhanced rendering efficiency, lower memory consumption, and detailed surface reconstruction.

This paper presents an investigation of vision transformer learning for multi-view geometry tasks, such as optical flow estimation, by fine-tuning video foundation models. Unlike previous methods that involve custom architectural designs and task-specific pretraining, our research finds that general-purpose models pretrained on videos can be readily transferred to multi-view problems with minimal adaptation. The core insight is that general-purpose attention between patches learns temporal and spatial information for geometric reasoning. We demonstrate that appending a linear decoder to the Transformer backbone produces satisfactory results, and iterative refinement can further elevate performance to state-of-the-art levels. This conceptually simple approach achieves top cross-dataset generalization results for optical flow estimation with end-point error (EPE) of 0.69, 1.78, and 3.15 on the Sintel clean, Sintel final, and KITTI datasets, respectively. Our method additionally establishes a new record on the online test benchmark with EPE values of 0.79, 1.88, and F1 value of 3.79. Applications to 3D depth estimation and stereo matching also show strong performance, illustrating the versatility of video-pretrained models in addressing geometric vision tasks.

Due to large pixel movement and high computational cost, estimating the motion of high-resolution frames is challenging. Thus, most flow-based Video Frame Interpolation (VFI) methods first predict bidirectional flows at low resolution and then use high-magnification upsampling (e.g., bilinear) to obtain the high-resolution ones. However, this kind of upsampling strategy may cause blur or mosaic at the flows' edges. Additionally, the motion of fine pixels at high resolution cannot be adequately captured in motion estimation at low resolution, which leads to the misalignment of task-oriented flows. With such inaccurate flows, input frames are warped and combined pixel-by-pixel, resulting in ghosting and discontinuities in the interpolated frame. In this study, we propose a novel VFI pipeline, VTinker, which consists of two core components: guided flow upsampling (GFU) and Texture Mapping. After motion estimation at low resolution, GFU introduces input frames as guidance to alleviate the blurring details in bilinear upsampling flows, which makes flows' edges clearer. Subsequently, to avoid pixel-level ghosting and discontinuities, Texture Mapping generates an initial interpolated frame, referred to as the intermediate proxy. The proxy serves as a cue for selecting clear texture blocks from the input frames, which are then mapped onto the proxy to facilitate producing the final interpolated frame via a reconstruction module. Extensive experiments demonstrate that VTinker achieves state-of-the-art performance in VFI.

Spatial transcriptomics (ST) bridges gene expression and tissue morphology but faces clinical adoption barriers due to technical complexity and prohibitive costs. While computational methods predict gene expression from H&E-stained whole-slide images (WSIs), existing approaches often fail to capture the intricate biological heterogeneity within spots and are susceptible to morphological noise when integrating contextual information from surrounding tissue. To overcome these limitations, we propose HiFusion, a novel deep learning framework that integrates two complementary components. First, we introduce the Hierarchical Intra-Spot Modeling module that extracts fine-grained morphological representations through multi-resolution sub-patch decomposition, guided by a feature alignment loss to ensure semantic consistency across scales. Concurrently, we present the Context-aware Cross-scale Fusion module, which employs cross-attention to selectively incorporate biologically relevant regional context, thereby enhancing representational capacity. This architecture enables comprehensive modeling of both cellular-level features and tissue microenvironmental cues, which are essential for accurate gene expression prediction. Extensive experiments on two benchmark ST datasets demonstrate that HiFusion achieves state-of-the-art performance across both 2D slide-wise cross-validation and more challenging 3D sample-specific scenarios. These results underscore HiFusion’s potential as a robust, accurate, and scalable solution for ST inference from routine histopathology.

Text-to-motion generation, which synthesizes 3D human motions from text inputs, holds immense potential for applications in gaming, film, and robotics. Recently, diffusion-based methods have been shown to generate more diversity and realistic motion. However, there exists a misalignment between text and motion distributions in diffusion models, which leads to semantically inconsistent or low-quality motions. To address this limitation, we propose Reward-guided sampling Alignment (ReAlign), comprising a step-aware reward model to assess alignment quality during the denoising sampling and a reward-guided strategy that directs the diffusion process toward an optimally aligned distribution. This reward model integrates step-aware tokens and combines a text-aligned module for semantic consistency and a motion-aligned module for realism, refining noisy motions at each timestep to balance probability density and alignment. Extensive experiments of both motion generation and retrieval tasks demonstrate that our approach significantly improves text-motion alignment and motion quality compared to existing state-of-the-art methods.

Online test-time adaptation addresses the train-test domain gap by adapting the model on unlabeled streaming test inputs before making the final prediction. However, online adaptation for 3D human pose estimation suffers from error accumulation when relying on self-supervision with imperfect predictions, leading to degraded performance over time. To mitigate this fundamental challenge, we propose a novel solution that highlights the use of motion discretization. Specifically, we employ unsupervised clustering in the latent motion representation space to derive a set of anchor motions, whose regularity aids in supervising the human pose estimator and enables efficient self-replay. Additionally, we introduce an effective and efficient soft-reset mechanism by reverting the pose estimator to its exponential moving average during continuous adaptation. We examine long-term online adaptation by continuously adapting to out-of-domain streaming test videos of the same individual, which allows for the capture of consistent personal shape and motion traits throughout the streaming observation. By mitigating error accumulation, our solution enables robust exploitation of these personal traits for enhanced accuracy. Experiments demonstrate that our solution outperforms previous online test-time adaptation methods and validate our design choices.

Modality-agnostic Semantic Segmentation (MaSS) aims to achieve robust scene understanding across arbitrary combinations of input modality. Existing methods typically rely on explicit feature alignment to achieve modal homogenization, which dilutes the distinctive strengths of each modality and destroys their inherent complementarity. To achieve cooperative harmonization rather than homogenization, we propose CHARM, a novel complementary learning framework designed to implicitly align content while preserving modality-specific advantages through two components: (1) Mutual Perception Unit (MPU), enabling implicit alignment through window-based cross-modal interaction, where modalities serve as both queries and contexts for each other to discover modality-interactive correspondences; (2) A dual-path optimization strategy that decouples training into Collaborative Learning Strategy (CoL) for complementary fusion learning and Individual Enhancement Strategy (InE) for protected modality-specific optimization. Experiments across multiple datasets and backbones indicate that CHARM consistently outperform the baselines, with significant increment on the fragile modalities. This work shifts the focus from model homogenization to harmonization, enabling cross-modal complementarity for true harmony in diversity.