Vector graphics provide continuous and often even smooth geometric representations of images. While recent approaches to automatically vectorize images lead to relatively good results, they typically leave ample room for improvement: the geometry and color of the vector graphics primitives can be further (automatically) optimized. We propose a novel method that generates high-quality vectorizations based on optimizing input curved triangle meshes (optionally with mesh colors). To overcome the key challenge of establishing a differentiable mapping between the input parameters, i.e., geometry and (mesh) colors of the gradient mesh, and the difference between the vectorized and input image, we treat the input image as a continuous bilinear interpolatory spline and employ Monte Carlo integration. We test our algorithm on various images and show that it can effectively and efficiently improve the quality of an initial vectorization.
Chilling injury (CI) is a major postharvest physiological disorder in fruits, causing quality degradation and economic losses during low-temperature storage. While physically-based methods exist for simulating plant deformation, they are computationally intensive and not optimized for capturing the subtle, spatially distributed symptoms of CI, such as browning, pitting, and wrinkling. In this paper, we propose a biologically-informed, texture-based framework for dynamic CI simulation that links biological symptom progression to visual representation. Browning and pitting are modeled using a texture-based de-chilling technique driven by a Logistic model of the chilling injury index (CII), with a histogram-matching-based algorithm ensuring alignment between simulated symptoms and CII values. Wrinkling is simulated by combining kinetic models of water loss with bump maps generated using Worley noise, which approximate the quasi-random yet locally continuous surface deformations caused by epidermal shrinkage. The proposed framework efficiently integrates biologically-driven modeling, dynamic texture evolution, and water-loss-induced surface deformation, producing realistic CI simulations without high-resolution meshes. It applies to multiple fruit types — including tropical climacteric (banana), Solanaceous (tomato), Cucurbit (cucumber), and Citrus (orange, lemon) — offering an effective approach for visualizing CI progression.
Cartesian mesh-based fluid simulation methods are gaining popularity due to their fully automated mesh generation capabilities for geometries without repair. The performance and flexibility of Cartesian mesh generation significantly influence their application across various fields. This study introduces an efficient adaptive Cartesian mesh generation framework directly for arbitrary geometries. Initially, we propose a robust, high-quality build-in tessellation method and compute proximity. Subsequently, we design a hierarchical storage method combined with binary search for efficient intersection determination. To enhance flexibility, a fully unstructured data type and compressed data representation are established. Finally, we develop a four-step refinement mechanism to achieve geometric adaptation and smooth transitions effectively. The robustness and efficiency of the approach were validated through typical case studies, demonstrating that the mesh generation process for complex models can reach speeds of up to 105 cells per second, which presents significant potential to address the challenges of real-time simulations.
Applications in fields such as fluid mechanics, video games and image processing frequently involve the simulation of 3D objects with spherical topology, with a surface quantity that varies with the geometry and/or according to some surface partial differential equation. However, when the geometry undergoes continuous deformations, significant distortions in the surface point distribution may arise. This can lead to aliasing effects and numerical instability, reducing the overall accuracy of the simulation.To address this issue, we introduce a novel tool to improve the efficiency of a spectral reparametrization procedure able to ensure the optimal representation of such 3D objects and surface quantities, even when dealing with long time simulations. This new strategy makes the reparametrization technique fully adaptive, selecting only the frequencies that are needed to represent a given surface, improving the efficiency of the algorithm, preventing degradation in the quality of the simulation and enhancing the overall stability.
In this paper we present a lattice structure optimization approach by leveraging the capabilities of neural networks for implicit geometry representation. We employ the Deep Signed Distance Field (DeepSDF) method, where a continuous and low-dimensional latent space is introduced to encode the geometric information. In contrast to traditional topology optimization methods, this allows the restriction of the design space to specific geometries. In our case, the latent space is used to represent the geometry of different unit cells, that are stacked to form a lattice structure. Moreover, continuously varying the latent vector over the structure allows a functional grading and optimization. Unlike other lattice-structure optimization methods, we neither assume a large separation of scale nor periodicity. Instead, we perform a full-scale finite element analysis at each optimization step. The required mesh is obtained by a differentiable extension of the dual marching cubes algorithm, which enables gradient-based optimization.
Graph Convolutional Network (GCN) has achieved remarkable result in skeleton-based action recognition. In GCNs, multi-order information has shown notable improvement for recognition and the graph topology, which is the key to fusing and extracting representative features. However, the GCN-based methods still face the following problems: (1) Nodes will have over-smooth problems in deep and complex networks. (2) Lack of efficient methods to fuse data streams of different modalities. In this paper, we proposed a novel data-fusing method, Feedback Directed Graph Convolution (FD-GC), to dynamically construct diverse correlation matrices and effectively aggregate both joint and bone features in different hierarchical update state and utilize them as feedback loops to participate in aggregation respectively for both streams. Our methods significantly reduce the difficulty of modeling multi-streams features at a small parameter cost. Furthermore, the experimental results indicate FD-GC alleviates the over-smooth effect via the feedback mechanism, constructing stronger representation capabilities of fine-grained actions, and performs as well as most skeletal motion recognition algorithms on two large public datasets NTU RGB+D 60, NTU RGB+D 120 and Northwestern-UCLA.
AI-aided floorplan design is a longstanding task in computer graphics. However, most of the existing methods focus on generating floorplans by limited architecture-level elements (e.g., room sizes, positions, and adjacencies), which ignore environmental factors and do not support customized designs. In this paper, we propose FlexPlan, an interactive approach for high-flexibility floorplan design. In FlexPlan, we propose a novel graph structure, named ArchiGraph, which enables flexible editing more comprehensive layout elements (e.g., architectures, environments, human needs) in a floorplan. First, we match similar floorplans according to the input architecture and environment features. Then, leveraging ArchiGraph, we interactively produce rooms’ attributes and quickly output the vectorized floorplans. For ArchiGraph, we design a RelationNet to predict room adjacencies, and propose a BoxNet to generate high-quality room boxes. Subjective and objective experiments show that our method is compatible with generating diverse complex floorplans (e.g., floorplans with irregular layout boundaries and room shapes). Compared with the state-of-the-art methods, our method can produce higher quality floorplans, and increase the speed of layout generation by nearly 20 times at most.
3D facial animation of digital avatars driven by RGB videos has extensive applications. However, the practical implementation encounters a significant challenge due to the various identities and environments of in-the-wild videos and varying rigging designs. The traditional industry pipeline necessitates a labor-intensive alignment process to ensure compatibility, while the recent novel methods are constrained to a specific rigging standard or require additional labor on actor videos, making them difficult to apply to customized riggings and in-the-wild videos. To make the task easy and convenient, we introduce EasyAnim, which utilizes abundant 2D videos to learn an aligned implicit motion flow unsupervisedly and maps it to various rigging parameters in a generalized manner. A novel framework with self-and cross-reconstruction constraints is proposed to ensure the alignment of avatar and human actor domains. Extensive experiments demonstrate that EasyAnim generates comparable or even better results with no additional constraints and labor.
In medical imaging detection of oral Cone Beam Computed Tomography (CBCT), there exist tiny lesions that are challenging to detect with low accuracy. The existing detection models are relatively complex. To address this, this paper presents a dual-stage YOLO detection method improved based on YOLOv8. Specifically, we first reconstruct the backbone network based on MobileNetV3 to enhance computational speed and efficiency. Second, we improve detection accuracy from three aspects: we design a composite feature fusion network to enhance the model's feature extraction capability, addressing the issue of decreased detection accuracy for small lesions due to the loss of shallow information during the fusion process; we further combine spatial and channel information to design the C2f-SCSA module, which delves deeper into the lesion information. To tackle the problem of limited types and insufficient samples of lesions in existing CBCT images, our team collaborated with a professional dental hospital to establish a high-quality dataset, which includes 15 types of lesions and over 2000 accurately labeled oral CBCT images, providing solid data support for model training. Experimental results indicate that the improved method enhances the accuracy of the original algorithm by 3.5 percentage points, increases the recall rate by 4.7 percentage points, and raises the mean Average Precision (mAP) by 3.3 percentage points, a computational load of only 7.6 GFLOPs. This demonstrates a significant advantage in intelligent diagnosis of full-mouth lesions while improving accuracy and reducing computational load.
The widespread application of metal additive manufacturing (AM) technologies has enabled exploration of complex design spaces to achieve optimally performing components. Current optimization techniques make use of several advanced methods, such as adjoint shape optimization, to provide designs that are superior to existing versions. However, they seldom discuss the manufacturability of the optimal designs. This research introduces novel restrictive design for AM (DfAM) constraints through computer-aided design (CAD) file modification which were used to guide the adjoint shape optimization process. The baseline design, using an application of a gas turbine fuel injector, was parameterized using non-uniform rational B-splines (NURBS) surface information stored in standard initial graphics exchange specification (IGES) file format. Gradient information computed using a commercial computational fluid dynamics (CFD) solver was used for NURBS shape modification in Python while focusing on imposing overhang angle and thin wall constraints for metal-AM. A method was developed to selectively replace information in the IGES file to accommodate modified design of surfaces of interest while preserving the overall geometry and maintain file integrity. The proposed framework accounts for varying levels of design complexity, accepting gradient information from commercial simulation software while imposing user-defined metal-AM constraints to obtain an optimal, additively manufacturable design. Findings from this study can be readily implemented in DfAM of any surface fluidic devices produced via metal laser AM, specifically Laser-Powder Bed Fusion.
Soft shadows play a crucial role in enhancing visual realism in real-time rendering. Although traditional shadow mapping techniques offer high efficiency, they often suffer from artifacts and limited quality. In contrast, ray tracing can produce high-fidelity soft shadows but incurs substantial computational cost. In this paper, we propose a general-purpose, real-time soft shadow generation method based on neural networks. To encode shadow geometry, we employ the hard shadows via shadow mapping as input to our network, which effectively captures the spatial layout of shadow positions and contours. A lightweight U-Net architecture then refines this input to synthesize high-quality soft shadows in real time. The generated shadows closely approximate ray-traced references in visual fidelity. Compared to existing learning-based methods, our approach produces higher-quality soft shadows and offers improved generalization across diverse scenes. Furthermore, it requires no scene-specific precomputation, making it directly applicable to practical real-time rendering scenarios.
The rapid evolution of digital orthodontics has highlighted a critical need for automated treatment planning systems that balance computational efficiency with clinical reliability. However, existing methods still suffer from several limitations, including excessive clinician involvement (accounting for over 35% of treatment planning time), reliance on empirically defined key frames, and limited biomechanical plausibility, particularly in cases of severe dental crowding. This paper proposes a novel collision-free optimization framework to address these issues simultaneously. Our method defines a total movement energy function evaluated over each tooth’s pose at intermediate time frames. This energy is minimized iteratively using a steepest descent strategy. A rollback mechanism is employed: if inter-tooth penetration is detected during an update, the step size is halved repeatedly until collisions are eliminated. The framework allows flexible control over the number of intermediate frames to enforce a strict constraint on per-tooth displacement, limiting it to 0.2 mm translation or 2° rotation every 10 to 14 days. Clinical evaluations show that the proposed algorithm can generate desirable and clinically valid tooth movement plans, even in complex cases, while significantly reducing the need for manual intervention.
With the proliferation of mobile hardware-accelerated ray tracing, visual quality at low sampling rates (1spp) significantly deteriorates due to high-frequency noise and temporal artifacts introduced by Monte Carlo path tracing. Traditional spatiotemporal denoising methods, such as Spatiotemporal Variance-Guided Filtering (SVGF), effectively suppress noise by fusing multi-frame information and using geometry buffer (G-buffer) guided filters. However, their reliance on per-frame variance computation and global filtering imposes prohibitive overhead for mobile devices. This paper proposes an edge-aware, data-driven real-time denoising architecture within the SVGF framework, tailored explicitly for mobile computational constraints. Our method introduces two key innovations that eliminate variance estimation overhead: (1) an adaptive filtering kernel sizing mechanism, which dynamically adjusts filtering scope based on local complexity analysis of the G-buffer; and (2) a data-driven weight table construction strategy, converting traditional computational processes into efficient real-time lookup operations. These innovations significantly enhance processing efficiency while preserving edge accuracy. Experimental results on the Qualcomm Snapdragon 768G platform demonstrate that our method achieves 55 FPS with 1spp input. This frame rate is 67.42% higher than mobile-optimized SVGF, provides better visual quality, and reduces power consumption by 16.80%. Our solution offers a practical and efficient denoising framework suitable for real-time ray tracing in mobile gaming and AR/VR applications.
We consider 2D scalar fields sampled on a regular grid. When the gradient is low relative to the resolution of the dataset’s range, the signal may contain flat spots: connected areas where all points share the same value. Flat spots hinder certain analyses, such as topological characterization or drainage network computations. We present an algorithm to determine a symbolic slope inside flat spots and consistently place a minimal set of critical points, in a way that is less biased than state-of-the-art methods. We present experimental results on both synthetic and real data, demonstrating how our method provides a more plausible positioning of critical points and a better recovery of the Morse–Smale complex.
Fusing a sequence of static images captured at various exposures, has shown great potential to approach High Dynamic Range (HDR) imaging. However, in the presence of large motion, mis-alignment is almost inevitable and leads to the ghost artifacts. Besides, factors such as noise in the dark region or color saturation in the over-bright region may also fail to fill local image details to the HDR image. This paper provides a motion-robust multi-exposure fusion model based on Swin Transformer. Particularly, we design feature selection gates, which are integrated with the feature warping layers to detect and block mis-alignments. To reconstruct the missing local details, we exploit the long distance contextual dependency in the exposure-space pyramid by the self-attention mechanism. Extensive numerical and visual evaluation has been conducted on a variety of benchmark datasets. The experiments show that our model achieves superior accuracy and efficacy to existing methods.
Virtual human avatars are essential for applications such as gaming, augmented reality, and virtual production. However, existing methods struggle to achieve high fidelity reconstruction from monocular input while keeping hardware costs low. Many approaches rely on the SMPL body prior and apply vertex offsets to represent clothed avatars. Unfortunately, excessive offsets often cause misalignment and blurred contours, particularly around clothing wrinkles, silhouette boundaries, and facial regions. To address these limitations, we propose a dual branch framework for human avatar reconstruction from monocular video. A lightweight Vertex Align Net (VAN) predicts per-vertex normal direction offsets on the SMPL mesh to achieve coarse geometric alignment and guide Gaussian-based human avatar modeling. In parallel, we construct a high resolution facial Gaussian branch based on FLAME estimated parameters, with facial regions localized via pretrained detectors. The facial and body renderings are fused using a semantic mask to enhance facial clarity and ensure globally consistent avatar appearance. Experiments demonstrate that our method surpasses state of the art approaches in modeling animatable human avatars with fine grained fidelity.
3D registration methods based on point-level information struggle in situations with noise, density variation, large-scale points, and small overlaps, while existing primitive-based methods are usually sensitive to tiny errors in the primitive extraction process. In this paper, we present a reliable and efficient global registration algorithm exploiting the RANdom SAmple Consensus (RANSAC) in the plane space instead of the point space. To improve the inlier ratio in the putative correspondences, we design an inner plane-based descriptor, termed Convex Hull Descriptor (CHD), and an inter plane-based descriptor, termed PLane Feature Histograms (PLFH), which take full advantage of plane contour shape and plane-wise relationship, respectively. Based on those new descriptors, we randomly select corresponding plane pairs to compute candidate transformations, followed by a hypotheses verification step to identify the optimal registration. Extensive tests on large-scale point sets demonstrate the effectiveness of our method, and that it notably improves registration performance compared to state-of-the-art methods in terms of efficiency and accuracy.
Recently, indoor scene synthesis has gathered significant attention, leading to the development of numerous indoor datasets. However, existing datasets only address static furniture and scenes, ignoring the need for dynamic interior design scenarios that emphasize flexible functionalities. Addressing this gap, we present DIFF (Dataset for Indoor Flexible Furniture), featuring expertly crafted and labeled furniture modules capable of inter-transforming between different states, e.g., a cabinet can be inter-transformed to a desk. Each module exhibits flexibility in shifting to multiple shapes and functionalities. Additionally, we propose a method that adapts our dataset to generate flexible layouts. By matching our flexible objects to objects from existing datasets, we use a graph-based approach to migrate the spatial relation priors for optimizing a layout; subsequent layouts are then generated by minimizing a transition-cost function. Analyses and user studies validate the quality of our modules and demonstrate the plausibility of the proposed method.
The integration of embodied intelligence in indoor scene synthesis holds significant potential for future interior design applications. Nevertheless, prevailing methodologies for indoor scene synthesis predominantly adhere to data-driven learning paradigms. Despite achieving photorealistic 3D renderings through such approaches, current frameworks systematically neglect to incorporate agent-centric functional metrics essential for optimizing navigational topology and task-oriented interactivity in embodied AI systems like service robotics platforms or autonomous domestic assistants. For example, poorly arranged furniture may prevent robots from effectively interacting with the environment, and this issue cannot be fully resolved by merely introducing prior constraints. To fill this gap, we propose Nav2Scene, a novel plug-and-play fine-tuning mechanism that can be deployed on existing scene generators to enhance the suitability of generated scenes for efficient robot navigation. Specifically, we first introduce path planning score (PPS), which is defined based on the results of the path planning algorithm and can be used to evaluate the robot navigation suitability of a given scene. Then, we pre-compute the PPS of 3D scenes from existing datasets and train a ScoreNet to efficiently predict the PPS of the generated scenes. Finally, the predicted PPS is used to guide the fine-tuning of existing scene generators and produce indoor scenes with higher PPS, indicating improved suitability for robot navigation. We conduct experiments on the 3D-FRONT dataset for different tasks including scene generation, completion and re-arrangement. The results demonstrate that by incorporating our Nav2Scene mechanism, the fine-tuned scene generators can produce scenes with improved navigation compatibility for home robots, while maintaining superior or comparable performance in terms of scene quality and diversity.
Vector graphics are widely used in areas such as logo design and digital painting, including both stroked and filled paths as primitives. GPU-based rendering for filled paths already has well-established solutions. Due to the complexity of stroked paths, existing methods often render them by approximating strokes with filled shapes. However, the performance of existing methods still leaves room for improvement. This paper designs a GPU-accelerated rendering algorithm along with a curvature-guided parallel adaptive subdivision method to accurately and efficiently render stroke areas. Additionally, we propose an efficient Newton iteration-based method for arc-length parameterization of quadratic curves, along with an error estimation technique. This enables a parallel rendering approach for dashed stroke styles and arc-length guided texture filling. Experimental results show that our method achieves average speedups of 3.4× for rendering quadratic stroked paths and 2.5× for rendering quadratic dashed strokes, compared to the best existing approaches.
