Search Results (21)

Surface reconstruction from 3D point clouds has a wide range of applications. In this paper, we focus on the reconstruction from raw, sparse point clouds. Although some existing methods work on this topic, the results often suffer from geometric defects. To solve this problem, we propose a novel method that optimizes a neural network (referred to as signed distance function) to fit the Signed Distance Field (SDF) from sparse point clouds. The signed distance function is optimized by projecting query points to its iso-surface accordingly. Our key idea is to encourage both the direction and distance of projection to be correct through the supervision provided by a nearest neighbor prior. In addition, we mitigate the error propagated from the prior function by augmenting the low-frequency components in the input. In our implementation, the nearest neighbor prior is trained with a large-scale local geometry dataset, and the positional encoding with a specified spectrum is used as a regularization for the optimization process. Experiments on the ShapeNetCore dataset demonstrate that our method achieves better accuracy than SDF-based methods while preserving smoothness. Full article

Reconstructing hand and object shapes from a single view during interaction remains challenging due to severe mutual occlusion and the need for high physical plausibility. To address this, we propose a novel framework for hand–object interaction reconstruction based on holistic, multi-stage collaborative optimization. Unlike methods that process hands and objects independently or apply constraints as late-stage post-processing, our model progressively enforces physical consistency and geometric accuracy throughout the entire reconstruction pipeline. Our network takes an RGB-D image as input. An adaptive feature fusion module first combines color and depth information to improve robustness against sensing uncertainties. We then introduce structural priors for 2D pose estimation and leverage texture cues to refine depth-based 3D pose initialization. Central to our approach is the iterative application of a dense mutual attention mechanism during sparse-to-dense mesh recovery, which dynamically captures interaction dependencies while refining geometry. Finally, we use a Signed Distance Function (SDF) representation explicitly designed for contact surfaces to prevent interpenetration and ensure physically plausible results. Through comprehensive experiments, our method demonstrates significant improvements on the challenging ObMan and DexYCB benchmarks, outperforming state-of-the-art techniques. Specifically, on the ObMan dataset, our approach achieves hand CD_h and object CD_o metrics of 0.077 cm² and 0.483 cm², respectively. Similarly, on the DexYCB dataset, it attains hand CD_h and object CD_o values of 0.251 cm² and 1.127 cm², respectively. Full article

Accurate and efficient simulation of airflows in human airways is critical for advancing the understanding of respiratory physiology, disease diagnostics, and inhalation drug delivery. Traditional computational fluid dynamics (CFD) provides detailed predictions but is often mesh-sensitive and computationally expensive for complex geometries. In this study, we explored the usage of physics-informed neural networks (PINNs) to simulate airflows in three geometries with increasing complexity: a duct, a simplified mouth–lung model, and a patient-specific upper airway. Key procedures to implement PINN training and testing were presented, including geometry preparation/scaling, boundary/constraint specification, training diagnostics, nondimensionalization, and inference mapping. Both the laminar PINN and SDF–mixing-length PINN were tested. PINN predictions were validated against high-fidelity CFD simulations to assess accuracy, efficiency, and generalization. The results demonstrated that nondimensionalization of the governing equations was essential to ensure training accuracy for respiratory flows at 1 m/s and above. Hessian-matrix-based diagnosis revealed a quick increase in training challenges with flow speed and geometrical complexity. Both the laminar and SDF–mixing-length PINNs achieved comparable accuracy to corresponding CFD predictions in the duct and simplified mouth–lung geometry. However, only the SDF–mixing-length PINN adequately captured flow details unique to respiratory morphology, such as obstruction-induced flow diversion, recirculating flows, and laryngeal jet decay. The results of this study highlight the potential of PINNs as a flexible alternative to conventional CFD for modeling respiratory airflows, with adaptability to patient-specific geometries and promising integration with static or real-time imaging (e.g., 4D CT/MRI). Full article

Physics-Informed Neural Networks (PINNs) represent a transformative approach to solving partial differential equation (PDE)-based boundary value problems by embedding physical laws into the learning process, addressing challenges such as non-physical solutions and data scarcity, which are inherent in traditional neural networks. This review analyzes critical challenges in PINN development, focusing on loss function design, geometric information integration, and their application in engineering modeling. We explore advanced strategies for constructing loss functions—including adaptive weighting, energy-based, and variational formulations—that enhance optimization stability and ensure physical consistency across multiscale and multiphysics problems. We emphasize geometry-aware learning through analytical representations—signed distance functions (SDFs), phi-functions, and R-functions—with complementary strengths: SDFs enable precise local boundary enforcement, whereas phi/R capture global multi-body constraints in irregular domains; in practice, hybrid use is effective for engineering problems. We also examine adaptive collocation sampling, domain decomposition, and hard-constraint mechanisms for boundary conditions to improve convergence and accuracy and discuss integration with commercial CAE via hybrid schemes that couple PINNs with classical solvers (e.g., FEM) to boost efficiency and reliability. Finally, we consider emerging paradigms—Physics-Informed Kolmogorov–Arnold Networks (PIKANs) and operator-learning frameworks (DeepONet, Fourier Neural Operator)—and outline open directions in standardized benchmarks, computational scalability, and multiphysics/multi-fidelity modeling for digital twins and design optimization. Full article

This article presents a novel fully automatic indoor surveying (FAIS) framework for large-scale indoor environments using a Terrestrial Laser Scanning (TLS) hardware system. Traditional methods for indoor surveying are labor-intensive and time-consuming, as they rely on manually positioning scanners for data capture and placing markers for registration. What is more, positioning scanners manually may cause uneven scanning or rescanning, including unstructured areas specifically. To ensure full coverage of the scene, we precisely obtain the number and location of scan stations through the Signed Distance Function (SDF) based method. Meanwhile, we propose an efficient large-scale dense point cloud registration method without markers. The proposed framework is adapted to environments where the scanner operates on a flat surface, such as office spaces, theater stage spaces, urban areas, and some cultural heritage scenic areas. Experiments demonstrate that the proposed method decreases computation time and obtains a more complete point cloud. Full article

In this paper, we investigate implicit surface reconstruction methods based on deep learning, enhanced by multi-sensor data fusion, to improve the accuracy of 3D reconstruction in complex scenes. Existing single-sensor approaches often struggle with occlusions and incomplete observations. By fusing complementary information from multiple sensors (e.g., multiple cameras or a combination of cameras and depth sensors), our proposed framework alleviates the issue of missing or partial data and further increases reconstruction fidelity. We introduce a novel deep neural network that learns a continuous signed distance function (SDF) for scene geometry, conditioned on fused multi-sensor feature representations. The network seamlessly merges multi-modal data into a unified implicit representation, enabling precise and watertight surface reconstruction. We conduct extensive experiments on 3D datasets, demonstrating superior accuracy compared to single-sensor baselines and classical fusion methods. Quantitative and qualitative results reveal that multi-sensor fusion significantly improves reconstruction completeness and geometric detail, while our implicit approach provides smooth, high-resolution surfaces. Additionally, we analyze the influence of the number and diversity of sensors on reconstruction quality, the model’s ability to generalize to unseen data, and computational considerations. Our work highlights the potential of coupling deep implicit representations with multi-sensor fusion to achieve robust 3D reconstruction in challenging real-world conditions. Full article

Reconstructing neuronal morphology from microscopy image stacks is essential for understanding brain function and behavior. While existing methods are capable of tracking neuronal tree structures and creating membrane surface meshes, they often lack seamless processing pipelines and suffer from stitching artifacts and reconstruction inconsistencies. In this study, we propose a new approach utilizing implicit neural representation to directly extract neuronal isosurfaces from raw image stacks by modeling signed distance functions (SDFs) with multi-layer perceptrons (MLPs). Our method accurately reconstructs the tubular, tree-like topology of neurons in complex spatial configurations, yielding highly precise neuronal membrane surface meshes. Extensive quantitative and qualitative evaluations across multiple datasets demonstrate the superior reliability of our approach compared to existing methods. The proposed method achieves a volumetric reconstruction accuracy of up to 98.2% and a volumetric IoU of 0.90. Full article

This paper proposes a method for 3D reconstruction from Freehand Design Sketching (FDS) in architecture and industrial design. The implementation begins by extracting features from the FDS using the self-supervised learning model DINO, followed by the continuous Signed Distance Function (SDF) regression as an implicit representation through a Multi-Layer Perceptron network. Taking eyeglass frames as an example, the 2D contour and freehand sketch optimize the alignment by their geometrical similarity while exploiting symmetry to improve reconstruction accuracy. Experiments demonstrate that this method can effectively reconstruct high-quality 3D models of eyeglass frames from 2D freehand sketches, outperforming existing deep learning-based 3D reconstruction methods. This research offers practical information for understanding 3D modeling methodology for FDS, triggering multiple modes of design creativity and efficient scheme adjustments in industrial or architectural conceptual design. In conclusion, this novel approach integrates self-supervised learning and geometric optimization to achieve unprecedented fidelity in 3D reconstruction from FDS, setting a new benchmark for AI-driven design processes in industrial and architectural applications. Full article

The integration of robotics in the garment industry remains relatively limited, primarily due to the challenges in the highly deformable nature of garments. The objective of this study is thus to explore a vision-based garment recognition and environment reconstruction model to facilitate the application of robots in garment processing. Object SLAM (Simultaneous Localization and Mapping) was employed as the core methodology for real-time mapping and tracking. To enable garment detection and reconstruction, two datasets were created: a 2D garment image dataset for instance segmentation model training and a synthetic 3D mesh garment dataset to enhance the DeepSDF (Signed Distance Function) model for generative garment reconstruction. In addition to garment detection, the SLAM system was extended to identify and reconstruct environmental planes, using the CAPE (Cylinder and Plane Extraction) model. The implementation was tested using an Intel Realsense^® camera, demonstrating the feasibility of simultaneous garment and plane detection and reconstruction. This study shows improved performance in garment recognition with the 2D instance segmentation models and an enhanced understanding of garment shapes and structures with the DeepSDF model. The integration of CAPE plane detection with SLAM allows for more robust environment reconstruction that is capable of handling multiple objects. The implementation and evaluation of the system highlight its potential for enhancing automation and efficiency in the garment processing industry. Full article

Efficiently reconstructing complex and intricate surfaces at scale remains a significant challenge in 3D surface reconstruction. Recently, implicit neural representations have become a popular topic in 3D surface reconstruction. However, how to handle loop closure and bundle adjustment is a tricky problem for neural methods, because they learn the neural parameters globally. We present an algorithm that leverages the concept of surfels and expands relevant definitions to address such challenges. By integrating neural descriptors with surfels and framing surfel association as a deformation graph optimization problem, our method is able to effectively perform loop closure detection and loop correction in challenging scenarios. Furthermore, the surfel-level representation simplifies the complexity of 3D neural reconstruction. Meanwhile, the binding of neural descriptors to corresponding surfels produces a dense volumetric signed distance function (SDF), enabling the mesh reconstruction. Our approach demonstrates a significant improvement in reconstruction accuracy, reducing the average error by 16.9% compared to previous methods, while also generating modeling files that are up to 90% smaller than those produced by traditional methods. Full article

Reconstructing 3D indoor scenes from 2D images has always been an important task in computer vision and graphics applications. For indoor scenes, traditional 3D reconstruction methods have problems such as missing surface details, poor reconstruction of large plane textures and uneven illumination areas, and many wrongly reconstructed floating debris noises in the reconstructed models. This paper proposes a 3D reconstruction method for indoor scenes that combines neural radiation field (NeRFs) and signed distance function (SDF) implicit expressions. The volume density of the NeRF is used to provide geometric information for the SDF field, and the learning of geometric shapes and surfaces is strengthened by adding an adaptive normal prior optimization learning process. It not only preserves the high-quality geometric information of the NeRF, but also uses the SDF to generate an explicit mesh with a smooth surface, significantly improving the reconstruction quality of large plane textures and uneven illumination areas in indoor scenes. At the same time, a new regularization term is designed to constrain the weight distribution, making it an ideal unimodal compact distribution, thereby alleviating the problem of uneven density distribution and achieving the effect of floating debris removal in the final model. Experiments show that the 3D reconstruction effect of this paper on ScanNet, Hypersim, and Replica datasets outperforms the state-of-the-art methods. Full article

The expression of robot arm morphology is a critical foundation for achieving effective motion planning and collision avoidance in robotic systems. Traditional geometry-based approaches usually suffer from the contradiction between the high demand for computing resources for fine expression and the insufficient detail expression caused by the pursuit of efficiency. The signed distance function addresses these drawbacks due to its ability to handle complex and arbitrary shapes and lower computational requirements. However, conventional robotic morphology methods based on the signed distance function often face challenges when the robot moves dynamically, since robots with different postures are modeled as independent individuals but the postures of robots are infinite. In this paper, we introduce RobotSDF, an implicit morphology modeling approach that can express the robot shape of arbitrary posture precisely. Instead of depicting a whole model of the robot arm, RobotSDF models the robot morphology as integrated implicit joint models driven by joint configurations. In this approach, the dynamic shape change process of the robot is converted into the coordinate transformations of query points within each joint’s coordinate system. Experimental results with the Elfin robot demonstrate that RobotSDF can accurately depict robot shapes across different postures up to the millimeter level, which exhibits $38.65\%$ and $66.24\%$ improvement over the Neural-JSDF and configuration space distance field algorithms, respectively, in representing robot morphology. We further verified the efficiency of RobotSDF through collision avoidance in both simulation and actual human–robot collaboration experiments. Full article

The 3D reconstruction of ancient buildings through inclined photogrammetry finds a wide range of applications in surveying, visualization and heritage conservation. Unlike indoor objects, reconstructing ancient buildings presents unique challenges, including the slow speed of 3D reconstruction using traditional methods, the complex textures of ancient structures and geometric issues caused by repeated textures. Additionally, there is a hash conflict problem when rendering outdoor scenes using neural radiation fields. To address these challenges, this paper proposes a 3D reconstruction method based on depth-supervised neural radiation fields. To enhance the representation of the geometric neural network, the addition of a truncated signed distance function (TSDF) supplements the existing signed distance function (SDF). Furthermore, the neural network’s training is supervised using depth information, leading to improved geometric accuracy in the reconstruction model through depth data obtained from sparse point clouds. This study also introduces a progressive training strategy to mitigate hash conflicts, allowing the hash table to express important details more effectively while reducing feature overlap. The experimental results demonstrate that our method, under the same number of iterations, produces images with clearer structural details, resulting in an average 15% increase in the Peak Signal-to-Noise Ratio (PSNR) value and a 10% increase in the Structural Similarity Index Measure (SSIM) value. Moreover, our reconstruction model produces higher-quality surface models, enabling the fast and highly geometrically accurate 3D reconstruction of ancient buildings. Full article

The reconstruction of 3D shapes from a single view has been a longstanding challenge. Previous methods have primarily focused on learning either geometric features that depict overall shape contours but are insufficient for occluded regions, local features that capture details but cannot represent the complete structure, or structural features that encode part relationships but require predefined semantics. However, the fusion of geometric, local, and structural features has been lacking, leading to inaccurate reconstruction of shapes with occlusions or novel compositions. To address this issue, we propose a two-stage approach for achieving 3D shape reconstruction. In the first stage, we encode the hierarchical structure features of the 3D shape using an encoder-decoder network. In the second stage, we enhance the hierarchical structure features by fusing them with global and point features and feed the enhanced features into a signed distance function (SDF) prediction network to obtain rough SDF values. Using the camera pose, we project arbitrary 3D points in space onto different depth feature maps of the CNN and obtain their corresponding positions. Then, we concatenate the features of these corresponding positions together to form local features. These local features are also fed into the SDF prediction network to obtain fine-grained SDF values. By fusing the two sets of SDF values, we improve the accuracy of the model and enable it to reconstruct other object types with higher quality. Comparative experiments demonstrate that the proposed method outperforms state-of-the-art approaches in terms of accuracy. Full article

Automatic reconstruction of surfaces from satellite imagery is a hot topic in computer vision and photogrammetry. State-of-the-art reconstruction methods typically produce 2.5D elevation data. In contrast, we propose a one-stage method directly generating a 3D mesh model from multi-view satellite imagery. We introduce a novel Sat-Mesh approach for satellite implicit surface reconstruction: We represent the scene as a continuous signed distance function (SDF) and leverage a volume rendering framework to learn the SDF values. To address the challenges posed by lighting variations and inconsistent appearances in satellite imagery, we incorporate a latent vector in the network architecture to encode image appearances. Furthermore, we introduce a multi-view stereo constraint to enhance surface quality. This constraint minimizes the similarity between image patches to optimize the position and orientation of the SDF surface. Experimental results demonstrate that our method achieves superior visual quality and quantitative accuracy in generating mesh models. Moreover, our approach can learn seasonal variations in satellite imagery, resulting in texture mesh models with different and consistent seasonal appearances. Full article