WACV 2026 论文列表

Image fusion aims to synthesize a single high-quality image from a pair of inputs captured under challenging conditions, such as differing exposure levels or focal depths. A core challenge lies in effectively handling disparities in dynamic range and focus depth between the inputs. With the advent of vision-language models, recent methods incorporate textual descriptions as auxiliary guidance to enhance fusion quality. However, simply incorporating coarse-grained descriptions hampers the understanding of fine-grained details and poses challenges for precise cross-modal alignment. To address these limitations, we propose Multi-grained Text-guided Image Fusion (MTIF), a novel fusion paradigm with three key designs. First, it introduces multi-grained textual descriptions that separately capture fine details, structural cues, and semantic content, guiding image fusion through a hierarchical cross-modal modulation module. Second, it involves supervision signals at each granularity to facilitate alignment between visual and textual features and enhance the utility of auxiliary text. Third, it adopts a saliency-driven enrichment module to augment training data with dense semantic content, further strengthening the cross-modal modulation and alignment. Extensive experiments show that MTIF consistently outperforms previous methods on both multi-exposure and multi-focus image fusion tasks.

Dance is a form of human motion characterized by emotional expression and communication, playing a role in various fields such as music, virtual reality, and content creation. Existing methods for dance generation often fail to adequately capture the inherently sequential, rhythmical, and music-synchronized characteristics of dance. In this paper, we propose MambaDance, a new dance generation approach that leverages a Mamba-based diffusion model. Mamba, well-suited to handling long and autoregressive sequences, is integrated into our two-stage diffusion architecture, substituting off-the-shelf Transformer. Additionally, considering the critical role of musical beats in dance choreography, we propose a Gaussian-based beat representation to explicitly guide the decoding of dance sequences. Experiments on AIST++ and FineDance datasets for each sequence length show that our proposed method effectively generates plausible dance movements while reflecting essential characteristics, consistently from short to long dances, compared to the previous methods. Additional qualitative results and demo videos are available at \small https://vision3d-lab.github.io/mambadance .

We propose SFMNet, a novel 3D sparse detector that combines the efficiency of sparse convolutions with the ability to model long-range dependencies. While traditional sparse convolution techniques efficiently capture local structures, they struggle with modeling long-range relationships. However, these relationships are essential for 3D object detection. In contrast, transformers are designed to capture these long-range dependencies through attention mechanisms. But, they come with high computational costs, due to their quadratic query-key-value interactions. Furthermore, directly applying attention to non-empty voxels is inefficient due to the sparse nature of 3D scenes. Our SFMNet is built on a novel Sparse Focal Modulation (SFM) module, which integrates short- and long-range contexts with linear complexity by leveraging a new hierarchical sparse convolution design. This approach enables SFMNet to achieve high detection performance with improved efficiency, making it well-suited for large-scale LiDAR scenes. We show that our detector achieves state-of-the-art performance on autonomous driving datasets.

Point cloud-based object/place recognition remains a problem of interest in applications such as autonomous driving, scene reconstruction, and localization. Extracting a meaningful global descriptor from a query point cloud that can be matched with the descriptors of the database point clouds is a challenging problem. Furthermore, when the query point cloud is noisy or has been transformed (e.g., rotated), it adds to the complexity. To this end, we propose a novel methodology, named TopoRec, which utilizes Topological Data Analysis (TDA) for extracting local descriptors from a point cloud, thereby eliminating the need for resource-intensive GPU-based machine learning training. More specifically, we used the ATOL vectorization method to generate vectors for point clouds. To test the quality of the proposed TopoRec technique, we have implemented it on multiple real-world (e.g., Oxford RobotCar, NCLT) and realistic (e.g., ShapeNet) point cloud datasets for large-scale place and object recognition, respectively. Unlike existing learning-based approaches such as PointNetVLAD and PCAN, our method does not require extensive training, making it easily adaptable to new environments. Despite this, it consistently outperforms both state-of-the-art learning-based and handcrafted baselines (e.g., M2DP, ScanContext) on standard benchmark datasets, demonstrating superior accuracy and strong generalization.

We present a central-peripheral vision-inspired framework (CVP), a simple yet effective multimodal model for spatial reasoning that draws inspiration from the two types of human visual fields -- central vision and peripheral vision. Existing approaches primarily rely on unstructured representations, such as point clouds, voxels, or patch features, and inject scene context implicitly via coordinate embeddings. However, this often results in limited spatial reasoning capabilities due to the lack of explicit, high-level structural understanding. To address this limitation, we introduce two complementary components into a Large Multimodal Model-based architecture: target-affinity token, analogous to central vision, that guides the model's attention toward query-relevant objects; and allocentric grid, akin to peripheral vision, that captures global scene context and spatial arrangements. These components work in tandem to enable structured, context-aware understanding of complex 3D environments. Experiments show that CVP achieves state-of-the-art performance across a range of 3D scene understanding benchmarks.

Generating stable and controllable character motion in real-time is a key challenge in computer animation. Existing methods often fail to provide fine-grained control or suffer from motion degradation over long sequences, limiting their use in interactive applications. We propose COMET, an autoregressive framework that runs in real time, enabling versatile character control and robust long-horizon synthesis. Our efficient Transformer-based conditional VAE allows for precise, interactive control over arbitrary user-specified joints for tasks like goal-reaching and in-betweening from a single model. To ensure long-term temporal stability, we introduce a novel reference-guided feedback mechanism that prevents error accumulation. This mechanism also serves as a plug-and-play stylization module, enabling real-time style transfer. Extensive evaluations demonstrate that COMET robustly generates high-quality motion at real-time speeds, significantly outperforming state-of-the-art approaches in complex motion control tasks and confirming its readiness for demanding interactive applications. Video results and code are available at https://comet-proj.github.io/.

We introduce Forensim, an attention-based state-space framework for image forgery detection that jointly localizes both manipulated (target) and source regions. Unlike traditional approaches that rely solely on artifact cues to detect spliced or forged areas, Forensim is designed to capture duplication patterns crucial for understanding context. In scenarios like protest imagery, detecting only the forged region--e.g., a duplicated act of violence inserted into a peaceful crowd--can mislead interpretation, highlighting the need for joint source-target localization. Forensim outputs three-class masks (pristine, source, target) and supports detection of both splicing and copy-move forgeries within a unified architecture. We propose a visual state-space model that leverages normalized attention maps to identify internal similarities, paired with a region-based block-attention module to distinguish manipulated regions. This design enables end-to-end training and precise localization. Forensim achieves state-of-the-art performance on standard benchmarks. We also release CMFD_Anything, a new dataset addressing limitations of existing copy-move forgery datasets. Source code and dataset will be released.

Continuous normalizing flows (CNFs) and diffusion models (DMs) generate high-quality data from a noise distribution. However, their sampling process demands multiple iterations to solve an ordinary differential equation (ODE) with high computational complexity. State-of-the-art methods focus on reducing the number of discrete time steps during sampling to improve efficiency. In this work, we explore a complementary direction in which the quality-complexity tradeoff can also be controlled in terms of the neural network length. We achieve this by rewiring the blocks in the transformer-based architecture to solve an inner discretized ODE w.r.t. its depth. Then, we apply a length consistency term during flow matching training, and as a result, the sampling can be performed with an arbitrary number of time steps and transformer blocks. Unlike others, our ODE_t(ODE_l) approach is solver-agnostic in time dimension and reduces both latency and, importantly, memory usage. CelebA-HQ and ImageNet generation experiments show a latency reduction of up to 2x in the most efficient sampling mode, and FID improvement of up to 2.8 points for high-quality sampling when applied to prior methods. We open-source our code and checkpoints at github.com/gudovskiy/odelt.

In this work, we introduce the 3D Gaussian Point Encoder, an explicit per-point embedding built on mixtures of learned 3D Gaussians. This explicit geometric representation for 3D recognition tasks is a departure from widely used implicit representations such as PointNet. However, it is difficult to learn 3D Gaussian encoders in end-to-end fashion with standard optimizers. We develop optimization techniques based on natural gradients and distillation from PointNets to find a Gaussian Basis that can reconstruct PointNet activations. The resulting 3D Gaussian Point Encoders are faster and more parameter efficient than traditional PointNets. As in the 3D reconstruction literature where there has been considerable interest in the move from implicit (e.g., NeRF) to explicit (e.g., Gaussian Splatting) representations, we can take advantage of computational geometry heuristics to accelerate 3D Gaussian Point Encoders further. We extend filtering techniques from 3D Gaussian Splatting to construct encoders that run 2.7x faster as a comparable accuracy PointNet while using 46% less memory and 88% fewer FLOPs. Furthermore, we demonstrate the effectiveness of 3D Gaussian Point Encoders as a component in Mamba3D, running 1.27x faster and achieving a reduction in memory and FLOPs by 42% and 54% respectively. 3D Gaussian Point Encoders are lightweight enough to achieve high framerates on CPU-only devices.

Recent multimodal deepfake detection methods typically rely on single-stage training, which can cause the model to focus on dataset-specific multimodal cues while missing important features that are helpful to detect unseen manipulations, thereby limiting generalization. While some approaches attempt to address this using self-supervised audio-visual pretraining, they may not fully exploit cross-modal temporal information. Also, they often assume that manipulations affect the entire video, ignoring more realistic cases where only short segments are altered. To overcome these limitations, we propose a two-stage training framework that first learns audio-visual temporal alignment in real videos and then uses this information to detect and localize potential deepfakes by identifying temporal inconsistencies. We propose a self-supervised shift-prediction pretraining objective to fully understand cross-modal temporal alignment across multiple temporal shifts applied to the audio input. The pretrained features enable the model to identify manipulations across entire videos as well as accurately localize deepfake segments in partially tampered content. Moreover, the pretrained components do not require task-specific fine-tuning, improving the model's flexibility for both classification and localization. Experiments on benchmark datasets demonstrate strong within-dataset performance, robust generalization to new manipulations and datasets, and accurate temporal localization.

OCR and document layout analysis algorithms are essential components of AI-based document systems, yet they are typically trained on clean, degradation-free images. When applied to degraded documents such as blurred scans or pages spoiled by undesired handwritten text their performance drops significantly. To address this issue, we propose DocWaveDiff, a novel document restoration method based on a predict-and-refine diffusion framework incorporating wavelet U-Nets. Given a degraded image patch and optionally its prior features, our Early Predictor generates an initial restoration, which is then refined by a Denoiser Refiner that estimates the residual image. The combination of these outputs yields the final restored result. We evaluate DocWaveDiff on multiple public benchmarks and demonstrate its strong performance across various document degradation scenarios, including deblurring and handwriting removal. Our results confirm that integrating wavelet transforms into the predict-and-refine framework enhances restoration quality and supports more robust document understanding systems.

Vector quantization (VQ) is a prevalent and fundamental technique that discretizes continuous feature vectors by approximating them using a codebook.As the diversity and complexity of data and models continue to increase, there is an urgent need for high-capacity, yet more compact VQ methods.This paper aims to reconcile this conflict by presenting a new approach called LooC, which utilizes an effective Low-dimensional codebook for Compositional vector quantization.Firstly, LooC introduces a parameter-efficient codebook by reframing the relationship between codevectors and feature vectors, significantly expanding its solution space.Instead of individually matching codevectors with feature vectors, LooC treats them as lower-dimensional compositional units within feature vectors and combines them, resulting in a more compact codebook with improved performance.Secondly, LooC incorporates a parameter-free extrapolation-by-interpolation mechanism to enhance and smooth features during the VQ process, which allows for better preservation of details and fidelity in feature approximation. The design of LooC leads to full codebook usage, effectively utilizing the compact codebook while avoiding the problem of collapse. Thirdly, LooC can serve as a plug-and-play module for existing methods for different downstream tasks based on VQ. Finally, extensive evaluations on different tasks, datasets, and architectures demonstrate that LooC outperforms existing VQ methods, achieving state-of-the-art performance with a significantly smaller codebook.

We analyze Layer Normalization (LN) from a domain (batch) perspective and explain why BatchNorm-style test-time fixes often fail on Transformer backbones. As feature dimension and batch size grow, the per-feature batch marginals after LN's pre-affine step concentrate at mean ~ 0 and variance ~ 1, making cross-batch re-standardization unnecessary and often harmful. This yields a simple rule: keep the pre-affine LN intact and adjust only the post-affine mean and gain. We instantiate this with LN-TTA, a backpropagation-free and source-free, test-time adaptation that performs a single forward pass and uniformly reparameterizes each LN layer. On three corrupted 3D point-cloud suites (ScanObjectNN-C, ModelNet40-C, ShapeNet-C), LN-TTA improves over Source-Only by +12.35, +15.58, and +3.03 points, surpasses backpropagation baselines (e.g., TENT), and sustains up to 93 samples/s, on average 39xfaster and 5xmore memory-efficient than the next-best backprop-free method. Code is available at: github.com/MosyMosy/LN_TTA

Volumetric medical image segmentation remains a challenging and critical task in both clinical and research settings due to the inherent complexity of anatomical structures, modality-specific variability, and the need to capture both fine-grained local details and long-range spatial dependencies across 3D volumes. To address these challenges, we propose Hymavi, a novel hybrid architecture that combines Mamba-based sequence modeling with attention mechanisms in a parallel design. This dual-branch structure enables Hymavi to simultaneously leverage the high-resolution spatial reasoning capabilities of attention and the efficient global context modeling afforded by Mamba's recurrent-style architecture. In addition to its architectural innovations, Hymavi incorporates a multi-view learning strategy that leverages sagittal and coronal perspectives alongside the conventional axial view. This multi-view fusion enriches volumetric representation by integrating complementary anatomical information from different orientations, allowing the network to better capture inter-slice continuity and organ-specific variations. Extensive experiments on three widely used benchmark datasets including ACDC, BraTS2023, and AMOS22 demonstrate the effectiveness and strong generalization ability of our method across diverse segmentation tasks and imaging modalities. These results underscore the potential of Hymavi as a powerful tool for advancing automated medical image analysis. The code is publicly available at https://github.com/sydat2701/Hymavi_segmentation.

Training robust deep video representations has proven to be computationally challenging due to substantial decoding overheads, the enormous size of raw video streams, and their inherent high temporal redundancy. Different from existing schemes, operating exclusively in the compressed video domain and exploiting all freely available modalities, i.e., I-frames, and P-frames (motion vectors and residuals) offers a compute-efficient alternative. Existing methods approach this task as a naive multi-modality problem, ignoring the temporal correlation and implicit sparsity across P-frames for modeling stronger shared representations for videos of the same action, making training and generalization easier. By revisiting the high-level design of dominant video understanding backbones, we increase inference speed by a factor of 56 while retaining similar performance. For this, we propose a hybrid end-to-end framework that factorizes learning across three key concepts to reduce inference cost by 330xversus prior art: First, a specially designed dual-encoder scheme with efficient Spiking Temporal Modulators to minimize latency while retaining cross-domain feature aggregation. Second, a unified transformer model to capture inter-modal dependencies using global self-attention to enhance I-frame -- P-frame contextual interactions. Third, a Multi-Modal Mixer Block to model rich representations from the joint spatiotemporal token embeddings. Experiments show that our method results in a lightweight architecture achieving state-of-the-art video recognition performance on UCF-101, HMDB-51, K-400, K-600 and SS-v2 datasets with favorable costs (0.73J/V) and fast inference (16V/s). Further, our observations bring new insights into practical design choices for efficient next-generation spatiotemporal learners.

Point cloud registration is a fundamental task in 3D vision. Most existing methods only use geometric information for registration. Recently proposed RGB-D registration methods primarily focus on feature fusion or improving feature learning, which limits their ability to exploit image information and hinders their practical applicability. In this paper, we propose ViGG, a robust RGB-D registration method using mutual guidance. First, we solve clique alignment in a visual-geometric combination form, employing a geometric guidance design to suppress ambiguous cliques. Second, to mitigate accuracy degradation caused by noise in visual matches, we propose a visual-guided geometric matching method that utilizes visual priors to determine the search space, enabling the extraction of high-quality, noise-insensitive correspondences. This mutual guidance strategy brings our method superior robustness, making it applicable for various RGB-D registration tasks. The experiments on 3DMatch, ScanNet and KITTI datasets show that our method outperforms recent state-of-the-art methods in both learning-free and learning-based settings. Code is available at https://github.com/ccjccjccj/ViGG.

Atmospheric Turbulence (AT) degrades the clarity and accuracy of surveillance imagery, posing challenges not only for visualization quality but also for object classification and scene tracking. Deep learning-based methods have been proposed to improve visual quality, but spatio-temporal distortions remain a significant issue. Although deep learning-based object detection performs well under normal conditions, it struggles to operate effectively on sequences distorted by atmospheric turbulence. In this paper, we propose a novel framework that learns to compensate for distorted features while simultaneously improving visualization and object detection. This end-to-end training strategy leverages and exchanges knowledge of low-level distorted features in the AT mitigator with semantic features extracted in the object detector. Specifically, in the AT mitigator a 3D Mamba-based structure is used to handle the spatio-temporal displacements and blurring caused by turbulence. Optimization is achieved through back-propagation in both the AT mitigator and object detector. Our proposed DMAT outperforms state-of-the-art AT mitigation and object detection systems up to a 15% improvement on datasets corrupted by generated turbulence.

The proliferation of harmful internet memes poses a significant societal threat, yet their automated classification remains a formidable algorithmic challenge due to the nuanced, multimodal nature of their content. To address this, we introduce MemeTAG, a novel dual objective framework that pioneers a keyword-aware approach to meme classification. Our core innovation is a two-part semantic guid-ance mechanism: first, we leverage a pretrained Vision-Language Model to generate a set of descriptive keywords, that capture the high-level semantics. Second, we introduce the Aggregated Tag Inference Network (ATIN), an attention-based module that distills these keywords into a single, rich semantic embedding. This embedding servesas a target for a novel auxiliary reconstruction loss, which compels the model to learn deeply aligned visual and textual features. This approach, combined with an efficient three-stage training strategy, establishes a new state-of-the-art on the HarMeme, Hateful Memes Challenge (HMC) and PrideMM datasets, decisively outperforming existing state-of-the-art methods.

Few-shot learning aims to generalize deep models to novel categories with only a handful of labeled examples, but existing methods remain vulnerable to task-irrelevant noise, unstable prototype estimation, and limited adaptability under domain shift. To address these issues, we propose the Noise-Adaptive Prototype Perturbation Network (NAPP), a framework that enhances robustness and generalization for few-shot learning. NAPP introduces three key innovations: (1) a Noise Cancellation Mechanism embedded in Vision Transformer self-attention layers that dynamically suppresses spurious, task-irrelevant features. (2) a MixPerturbation Module that perturbs class prototypes through augmented feature combinations, producing more stable and transferable prototype representations. (3) an Adaptive Noise-Conditioned Meta-Learning scheme that finetunes less than 0.02% of noise-related parameters at metatest time, enabling efficient and rapid adaptation to unseen classes without eroding pretrained knowledge. Extensive experiments demonstrate that NAPP achieves competitive and superior performance compared to state-of-the-art fewshot classification methods across both in-domain and challenging cross-domain benchmarks.