MOTIVATION: Developing segmentation models that remain reliable across diverse medical imaging domains and accurately delineate complex anatomical boundaries remains a persistent challenge for clinical deployment. Variations in imaging modalities, scanners, and acquisition settings introduce significant domain shifts, while fuzzy or overlapping tissue boundaries further complicate precise segmentation. Despite extensive research, most approaches address these challenges separately, leading to limited generalization and reduced robustness in real-world clinical scenarios. RESULTS: To overcome these limitations, we propose HyperSeg-DG, a novel medical image segmentation approach that integrates the WMamba backbone with the Multi-Scale Hyper Feature Context Block (HFCB). The HFCB addresses foreground-background uncertainty and boundary ambiguities by capturing multi-scale feature relations and long-range contextual dependencies. This enables the model to focus on relevant pathological features while helping reduce the influence of irrelevant co-occurring ones, such as similarly sized polyps, especially in low-contrast or poorly lit environments. WMamba further improves domain generalization by processing images in localized windows and using its selective 2D scanning mechanism to learn robust, transferable features that reduce feature misalignment under domain shift. Extensive experiments across multiple medical segmentation benchmarks demonstrate that HyperSeg-DG achieves consistent 2%-3% improvements over strong baselines, confirming its effectiveness in enhancing segmentation performance and generalization across diverse, unseen domains. AVAILABILITY AND IMPLEMENTATION: The code and datasets of HyperSeg-DG are available at https://github.com/Pollob001/HyperSeg-DG.

While significant advancements have been made in appearance editing for 3D scenes, achieving intuitive and user-friendly editing remains a persistent challenge. Recently, 3D Gaussian Splatting (3D-GS) has ushered in a new era in novel view synthesis. Building upon this representation, we integrate palette-based recoloring to propose RecolorGaussian, a novel framework for 3D scene color editing. Our method decomposes the color of each Gaussian into a linear combination of multiple representative color bases shared across the scene, termed palette colors. To address view-dependent effects, we introduce neural offsets in the CIE LAB color space, enabling chromaticity-aligned adjustments to palette colors while capturing high-frequency details such as specular highlights. During the optimization stage, we design a smooth L0 norm approximation as a loss function to encourage sparser color blending weights, enabling higher-quality recoloring. RecolorGaussian enables users to intuitively recolor 3D scenes through direct palette manipulation, delivering an efficient and user-friendly workflow. Experimental results demonstrate that our method outperforms existing state-of-the-art methods in 3D scene recoloring both quantitatively and qualitatively.

Abstract Generating 3D objects with complex topologies from monocular images remains a challenge in computer graphics, due to the difficulty of modeling varying 3D shapes with disentangled, steerable geometry and visual attributes. While NeRF‐based methods suffer from slow volumetric rendering and limited structural controllability. Recent advances in 3D Gaussian Splatting provide a more efficient alternative and its generative modeling with separate control over structure and appearance remains underexplored. In this paper, we propose G‐SplatGAN , a novel 3D‐aware generation framework that combines the rendering efficiency of 3D Gaussian Splatting with disentangled latent modeling. Starting from a shared Gaussian template, our method uses dual modulation branches to modulate geometry and appearance from independent latent codes, enabling precise shape manipulation and controllable generation. We adopt a progressive adversarial training scheme with multi‐scale and patch‐based discriminators to capture both global structure and local detail. Our model requires no 3D supervision and is trained on monocular images with known camera poses, reducing data reliance while supporting real image inversion through a geometry‐aware encoder. Experiments show that G‐SplatGAN achieves superior performance in rendering speed, controllability and image fidelity, offering a compelling solution for controllable 3D generation using Gaussian representations.

Global illumination combines direct and indirect lighting to create realistic lighting effects, bringing virtual scenes closer to reality. Static global illumination is a crucial component of virtual scene rendering, leveraging precomputation and baking techniques to significantly reduce runtime computational costs. Unfortunately, many existing works prioritize visual quality by relying on extensive texture storage and massive pixel-level texture sampling, leading to large performance overhead. In this paper, we introduce an illumination reconstruction method that effectively reduces sampling in fragment shader and avoids additional render passes, making it well-suited for low-end platforms. To achieve high-quality global illumination with reduced memory usage, we adopt a spherical harmonics fitting approach for baking effective illumination information and propose an inverse probe distribution method that generates unique probe associations for each mesh. This association, which can be generated offline in the local space, ensures consistent lighting quality across all instances of the same mesh. As a consequence, our method delivers highly competitive lighting effects while using only approximately 5% of the memory required by mainstream industry techniques.

The 3D Gaussian Splatting (3D-GS) method has recently sparked a revolution in novel view synthesis with its remarkable visual effects and fast rendering speed. However, its reliance on simple spherical harmonics for color representation leads to subpar performance in complex scenes, particularly with effects like specular highlights and light refraction. Also, 3D-GS adopts a periodic split strategy, which significantly increases the model's disk space and hinders rendering efficiency. To tackle these challenges, we propose Gaussian Splatting with Neural Basis Extension (GSNB), a novel approach that substantially enhances the performance of 3D-GS in demanding scenes while reducing storage consumption. Drawing inspiration from basis function, GSNB utilizes a light-weight MLP to share feature coefficients with Spherical Harmonics (SH). This extends the color calculation of 3D Gaussians, resulting in more accurate visual effect modeling. This combination allows GSNB to achieve remarkable results even in scenes with challenging lighting and reflection conditions. Additionally, GSNB uses pre-computation to bake the MLP's output, thereby alleviating inference workload and subsequent speed loss. Furthermore, to leverage the capabilities of Neural Basis Extension and eliminate redundant Gaussians, we propose a new importance criterion to prune the converged Gaussian model and obtain a more compact representation through re-optimization. Our experimental results demonstrate that our method delivers high-quality rendering in most scenarios and effectively reduces redundant Gaussians without compromising rendering speed. The code is available at https://github.com/Dojizz/GSNB/.

Abstract Reconstructing the three‐dimensional (3D) shape and texture of the face from a single image is a significant and challenging task in computer vision and graphics. In recent years, learning‐based reconstruction methods have exhibited outstanding performance, but their effectiveness is severely constrained by the scarcity of available training data with 3D annotations. To address this issue, we present the PR3D (Precise and Realistic 3D face reconstruction) method, which consists of high‐precision shape reconstruction based on semi‐supervised learning and high‐fidelity texture reconstruction based on StyleGAN2. In shape reconstruction, we use in‐the‐wild face images and 3D annotated datasets to train the auxiliary encoder and the identity encoder, encoding the input image into parameters of FLAME (a parametric 3D face model). Simultaneously, a novel semi‐supervised hybrid landmark loss is designed to more effectively learn from in‐the‐wild face images and 3D annotated datasets. Furthermore, to meet the real‐time requirements in practical applications, a lightweight shape reconstruction model called fast‐PR3D is distilled through teacher–student learning. In texture reconstruction, we propose a texture extraction method based on face reenactment in StyleGAN2 style space, extracting texture from the source and reenacted face images to constitute a facial texture map. Extensive experiments have demonstrated the state‐of‐the‐art performance of our method.

Reconstruction of a continuous surface of two-dimensional manifold from its raw, discrete point cloud observation is a long-standing problem in computer vision and graphics research. The problem is technically ill-posed, and becomes more difficult considering that various sensing imperfections would appear in the point clouds obtained by practical depth scanning. In literature, a rich set of methods has been proposed, and reviews of existing methods are also provided. However, existing reviews are short of thorough investigations on a common benchmark. The present paper aims to review and benchmark existing methods in the new era of deep learning surface reconstruction. To this end, we contribute a large-scale benchmarking dataset consisting of both synthetic and real-scanned data; the benchmark includes object- and scene-level surfaces and takes into account various sensing imperfections that are commonly encountered in practical depth scanning. We conduct thorough empirical studies by comparing existing methods on the constructed benchmark, and pay special attention on robustness of existing methods against various scanning imperfections; we also study how different methods generalize in terms of reconstructing complex surface shapes. Our studies help identity the best conditions under which different methods work, and suggest some empirical findings. For example, while deep learning methods are increasingly popular in the research community, our systematic studies suggest that, surprisingly, a few classical methods perform even better in terms of both robustness and generalization; our studies also suggest that the practical challenges of misalignment of point sets from multi-view scanning, missing of surface points, and point outliers remain unsolved by all the existing surface reconstruction methods. We expect that the benchmark and our studies would be valuable both for practitioners and as a guidance for new innovations in future research.

Category-level 6D object pose estimation aims to predict the position and orientation of unseen object instances, which is a fundamental problem in robotic applications . Previous works mainly focused on exploiting visual cues from RGB images , while depth images received less attention. However, depth images contain rich geometric attributes about the object’s shape, which are crucial for inferring the object’s pose. This work achieves category-level 6D object pose estimation by performing sufficient geometric learning from depth images represented by point clouds. Specifically, we present a novel geometric consistency and geometric discrepancy learning framework called CD-Pose to resolve the intra-category variation, inter-category similarity, and objects with complex structures. Our network consists of a Pose-Consistent Module and a Pose-Discrepant Module. First, a simple MLP-based Pose-Consistent Module is utilized to extract geometrically consistent pose features of objects from the pre-computed object shape priors for each category. Then, the Pose-Discrepant Module, designed as a multi-scale region-guided transformer network, is dedicated to exploring each instance’s geometrically discrepant features. Next, the NOCS model of the object is reconstructed according to the integration of consistent and discrepant geometric representations . Finally, 6D object poses are obtained by solving the similarity transformation between the reconstruction and the observed point cloud. Experiments on the benchmark datasets show that our CD-Pose produces superior results to state-of-the-art competitors.

Category-level object pose estimation aims to predict the 6D pose and 3D metric size of objects from given categories. Due to significant intra-class shape variations among different instances, existing methods have mainly focused on estimating dense correspondences between observed point clouds and their canonical representations, i.e., normalized object coordinate space (NOCS). Subsequently, a similarity transformation is applied to recover the object pose and size. Despite these efforts, current approaches still cannot fully exploit the intrinsic geometric features to individual instances, thus limiting their ability to handle objects with complex structures (i.e., cameras). To overcome this issue, this paper introduces GPT-COPE, which leverages a graph-guided point transformer to explore distinctive geometric features from the observed point cloud. Specifically, our GPT-COPE employs a Graph-Guided Attention Encoder to extract multiscale geometric features in a local-to-global manner and utilizes an Iterative Non-Parametric Decoder to aggregate the multiscale geometric features from finer scales to coarser scales without learnable parameters. After obtaining the aggregated geometric features, the object NOCS coordinates and shape are regressed through the shape prior adaptation mechanism, and the object pose and size are obtained using the Umeyama algorithm. The multiscale network design enables perceiving the overall shape and structural information of the object, which is beneficial to handle objects with complex structures. Experimental results on the NOCS-REAL and NOCS-CAMERA datasets demonstrate that our GPT-COPE achieves state-of-the-art performance and significantly outperforms existing methods. Furthermore, our GPT-COPE shows superior generalization ability compared to existing methods on the large-scale in-the-wild dataset Wild6D and achieves better performance on the REDWOOD75 dataset, which involves objects with unconstrained orientations.

Recovery of an underlying scene geometry from multi-view images stands as a long-time challenge in computer vision research. The recent promise leverages neural implicit surface learning and differentiable volume rendering, and achieves both the recovery of scene geometry and synthesis of novel views, where deep priors of neural models are used as an inductive smoothness bias. While promising for object-level surfaces, these methods suffer when coping with complex scene surfaces. In the meanwhile, traditional multi-view stereo can recover the geometry of scenes with rich textures, by globally optimizing the local, pixel-wise correspondences across multiple views. We are thus motivated to make use of the complementary benefits from the two strategies, and propose a method termed Helix-shaped neural implicit Surface learning or HelixSurf; HelixSurf uses the intermediate prediction from one strategy as the guidance to regularize the learning of the other one, and conducts such intertwined regularization iteratively during the learning process. We also propose an efficient scheme for differentiable volume rendering in HelixSurf Experiments on surface reconstruction of indoor scenes show that our method compares favorably with existing methods and is orders of magnitude faster, even when some of existing methods are assisted with auxiliary training data. The source code is available at hups://guhub.com/Gorilla-Lab-SCUT/HelixSurf.

The pseudocode-to-code task has two main stages: code translation and search synthesis. It faces two main challenges: First, the generated candidate code pieces need to be more accurate. Second, there are problems with search efficiency and accuracy. To address the above challenges, this work proposes a novel approach: For the encoder of code translation, a new multi-scale pyramid feature extractor is proposed to obtain multi-scale local information, which is combined with the global information to improve the accuracy of code translation. For the search synthesis stage, this paper designs the intra-line attention, the inter-line attention, and the code-errMsg attention, which are adaptively integrated with the graph attention to effectively fuse global and local information. Under a budget of 100 program compilations, our final model, AGL-Code, outperforms the previous state-of-the-art models, achieving 46.1%/63.5% synthesis success rate on the TestP/TestW of the SPoC dataset, respectively.

Abstract In this paper, we propose a novel surface subdivision scheme called non‐box subdivision, which is generalized from four‐directional S13 non‐box splines. The resulting subdivision surfaces achieve C1 continuity with the convex hull property. This scheme can be regarded as either a four‐directional subdivision or a special quadrilateral subdivision. When used as a quadrilateral subdivision, the proposed scheme can control the shape of the limit surface more flexibly than traditional schemes due to the natural introduction of auxiliary face control vertices.

This paper presents 6D vision transformer (6D-ViT), a transformer-based instance representation learning network suitable for highly accurate category-level object pose estimation based on RGB-D images. Specifically, a novel two-stream encoder-decoder framework is dedicated to exploring complex and powerful instance representations from RGB images, point clouds, and categorical shape priors. The whole framework consists of two main branches, named Pixelformer and Pointformer. Pixelformer contains a pyramid transformer encoder with an all-multilayer perceptron (MLP) decoder to extract pixelwise appearance representations from RGB images, while Pointformer relies on a cascaded transformer encoder and an all-MLP decoder to acquire the pointwise geometric characteristics from point clouds. Then, dense instance representations (i.e., correspondence matrix and deformation field) for NOCS model reconstruction are obtained from a multisource aggregation (MSA) network with shape prior, appearance and geometric information as inputs. Finally, the instance 6D pose is computed by solving the similarity transformation between the observed point clouds and the reconstructed NOCS representations. Extensive experiments with synthetic and real-world datasets demonstrate that the proposed framework achieves state-of-the-art performance for both datasets. Code is available at https://github.com/luzzou/6D-ViT.

Recent efforts have been made on domestic activities classification from audio recordings, especially the works submitted to the challenge of DCASE (Detection and Classification of Acoustic Scenes and Events) since 2018. In contrast, few studies were done on domestic activities clustering, which is a newly emerging problem. Domestic activities clustering from audio recordings aims at merging audio clips which belong to the same class of domestic activity into a single cluster. Domestic activities clustering is an effective way for unsupervised estimation of daily activities performed in home environment. In this study, we propose a method for domestic activities clustering using a convolutional capsule autoencoder network (CCAN). In the method, the deep embeddings are learned by the autoencoder in the CCAN, while the deep embeddings which belong to the same class of domestic activities are merged into a single cluster by a clustering layer in the CCAN. Evaluated on a public dataset adopted in DCASE- 2018 Task 5, the results show that the proposed method outperforms state-of-the-art methods in terms of the metrics of clustering accuracy and normalized mutual information.

Face views are particularly important in person-to-person communication. Differenes between the camera location and the face orientation can result in undesirable facial appearances of the participants during video conferencing. This phenomenon is particularly noticeable when using devices where the front-facing camera is placed in unconventional locations such as below the display or within the keyboard. In this paper, we take a video stream from a single RGB camera as input, and generate a video stream that emulates the view from a virtual camera at a designated location. The most challenging issue in this problem is that the corrected view often needs out-of-plane head rotations. To address this challenge, we reconstruct the 3D face shape and re-render it into synthesized frames according to the virtual camera location. To output the corrected video stream with natural appearance in real time, we propose several novel techniques including accurate eyebrow reconstruction, high-quality blending between the corrected face image and background, and template-based 3D reconstruction of glasses. Our system works well for different lighting conditions and skin tones, and can handle users wearing glasses. Extensive experiments and user studies demonstrate that our method provides high-quality results.

Summary form only given. Based on an optimal estimate of the convergence rate of the second order norm, an improved error estimate for extraordinary Catmull-Clark subdivision surface (CCSS) patches is proposed. If the valence of the extraordinary vertex of an extraordinary CCSS patch is even, a tighter error bound and, consequently, a more precise subdivision depth for a given error tolerance, can be obtained. Furthermore, examples of adaptive subdivision illustrate the practicability of the error estimation approach.

Real-time sky rendering plays a great role on visualization in many 3D applications. It is difficult to simulate a realistic sky without considering the effect of clouds and the scattering effect of the atmosphere. In this paper, we use a physical scattering model to simulate the sky, and a Perlin noise texture to create highly realistic clouds. Different effects of the sky are also generated by combining the lighting model with various atmospheric constituents.

In geometry processing a refined control mesh is often used to approximate a Catmull-Clark subdivision surface (CCSS). By pushing the control points to their limit positions, a limit mesh of the subdivision surface is obtained. We present a bound on the distance between a CCSS patch and its limit face in terms of the maximum norm of the second order differences of the control points. The bound shows that the limit mesh may approximate the limit surface better than the corresponding control mesh in general. Consequently, for a given error tolerance, fewer subdivision steps are needed if the refined control mesh is replaced with the corresponding limit mesh.