Multi-level -nearest neighbors algorithm for direct point cloud-based engineering analysis.

3D point clouds are widely utilized in critical vision tasks such as autonomous driving, augmented reality, and model reconstruction. Given the unstructured nature and large-scale characteristics of point cloud data, 3D point semantic segmentation often faces challenges in achieving both high accuracy and computational efficiency. To better balance accuracy and performance for domain-specific applications, this paper proposes a lightweight Rand Transformer Net (RTN), which constructs a more efficient multi-scale feature extraction module by using a random downsampling strategy and incorporates a specially designed Rand Transformer Block to capture local geometric features of point clouds. To address the issue of semantic ambiguity in boundary regions, A novel loss function, termed ABL loss, is introduced to impose additional constraints derived from labeled boundaries. Experimental results on a newly introduced Bridge Dataset, which is composed of large-scale point cloud-based bridge components, demonstrate that RTN achieves superior performance over the majority of existing point cloud semantic segmentation methods, with strong scalability and efficiency advantages in handling large-scale point cloud scenes.

Structural health monitoring of underground tunnels increasingly uses advanced sensing and data-driven methods. Laser-scanned 3D point clouds capture spatially rich measurements of segmental tunnel linings and require segmentation as a prerequisite for downstream analysis. Deep learning (DL) is effective for point-cloud segmentation, but scarce datasets and costly annotation limit practical use. This paper presents Tunnel Scanner , a high-fidelity simulator that synthesises realistic tunnel point clouds with automatic annotation. The plug-and-play module Hybrid Position–Normal Local Spatial Encoding embeds geometric priors into DL backbones and combines with transfer learning (TL) to exploit synthetic data for domain adaptation. Models trained only on synthetic data achieved 71.9% mean Intersection-over-Union (mIoU) and 86.6% Overall Accuracy (OA), and TL increased performance to at least 78.8% mIoU and 90.9% OA with limited real data. This paper highlights geometry-informed data synthesis as a viable augmentation approach for digital inspection and asset management of large-scale tunnels. • Develop a high-fidelity simulator to address the scarcity of tunnel point clouds. • Propose a plug-and-play module encoding geometric features into DL backbones. • Investigate transfer learning to enhance 3D Sim-to-Real domain adaptation. • The end-to-end framework yields ＋24% segmentation accuracy with limited real data.

Tooth3dNet: A preliminary exploration for automatic 3D morphology design of dental crowns with a deep generative network.

L2G-Net: Local-to-global feature enhancement via cluster tokens for 3D place recognition.

High-fidelity 3D tree mesh models have broad applications in virtual geographic environments, forest ecology, and digital entertainment. Although laser point clouds provide precise data, existing point cloud-based methods struggle with accuracy and completeness due to tree structural complexity, leaf occlusion, and data gaps. To address these challenges, we propose a novel framework that integrates data-driven and morphology-knowledge-driven techniques for generating high-fidelity tree mesh models with accurate trunk and overall crown morphology. The proposed framework consists of three main stages. First, in the trunk separation stage, non-scattering feature clustering is employed to extract the main trunk points. Second, in the robust trunk mesh generation and optimization stage, a coarse-to-fine skeleton extraction and refinement method is proposed, followed by weighted Levenberg–Marquardt cylindrical fitting to generate the trunk geometry. Finally, we propose a novel method that integrates improved space colonization with Alpha Shape constraints to overcome challenges of reconstructing fine branches within the crown, thereby generating complete tree mesh models with accurate trunk and crown morphology. The method was evaluated on laser-scanned, 3D Gaussian, and synthetic point clouds, and benchmarked against state-of-the-art techniques. Results demonstrate superior accuracy and robustness across diverse point cloud types, with strong resilience to noise and incomplete data.

Despite significant progress in point cloud upsampling, most existing methods rely heavily on supervised training with paired data, which are often difficult to acquire. Moreover, the inherent lack of explicit connectivity in point clouds makes it difficult to achieve both continuous and uniform densification while accurately recovering fine geometric structures. To address these problems, we propose an upsampling model that treats this task as saliency-aware implicit surface sampling, enabling self-supervised and fine-grained point densification. Central to our idea is correlating implicit surface reconstruction with salient point identification, and carrying out sampling on the saliency-aware surface representation. Motivated by this, we introduce a salient point detector along with a corresponding implicit surface-based interpolator, and a geometry filter, upon which we develop a three-stage architecture consisting of a pre-trained saliency guidance block, a saliency-aware enhancer, and an upsampler. The guidance block captures meaningful shape patterns to prevent detail loss and incomplete recovery, while the enhancer facilitates detail enhancement in complex salient regions. These components are integrated with the upsampler to generate dense results that accurately retain both global shape and meticulous structures. In comparison with state-of-the-art methods, our model significantly improves the upsampling quality. Extensive experiments conducted on various datasets, comprising both synthetic and real-world captured shapes, demonstrate the flexibility and availability of our method in processing the point clouds across different scales and distributions.

Achieving rapid and high-precision 3D reconstructions of urban scenes is crucial for building the digital Earth. The key challenge lies in accurately recovering fine structural edges, but sparse point clouds provide limited geometric constraints. Although dense point clouds increase reconstruction accuracy, their high computational cost restricts their practical use. Moreover, most existing methods overlook line features as essential geometric primitives for surface modeling, reducing both accuracy and efficiency. To address these challenges, a surface reconstruction method that integrates 3D line features and sparse point clouds based on multiview images is proposed. First, a line feature sampling method based on point cloud density, which combines line features with sparse point clouds in 3D space, is proposed. Second, accurate edge visual information of the surface model is obtained by constructing the average ray of the line features. Third, an energy function with a line feature constraint is proposed to ensure the integrity of the surface edge. Quantitative evaluations indicate that the median surface-to-point distance decreases by 15.3% in small-scale scenes and 14.9% in large-scale scenes. The reconstruction time is reduced by 98.3% compared with that of dense matching, indicating the high accuracy and efficiency of the proposed approach.

The task of surface reconstruction from point clouds is to produce high-quality meshes using sampled 3D points (no images available). Traditional methods primarily focus on geometric accuracy but often produce meshes without texture colors. In this paper, we present a brand-new perspective in point cloud reconstruction task-Imagining points as more informative Gaussian splats and obtaining colored surfaces through free-form Gaussian-rendering reconstruction. We train a universal Point-to-Gaussian model to infer the attributes of Gaussian splats for any given pointcloud with merely point coordinates (and color optionally) as input, without requiring any image. Significant technical designs are applied on initialization, regularization and loss functions, making the whole learning process stable. The inferred Gaussian splats can faithfully recover the original appearance of objects or scenes (capable of quick rendering from any viewpoint, like human's imagination ability), meanwhile closely adhering to the input shape. After obtaining a sufficient number of virtually rendered images and depth maps, we employ the truncated signed distance function (TSDF) fusion to get the reconstruction results, producing high-quality and colored meshes. Extensive experiments demonstrate that our approach surpasses state-of-the-art methods in surface reconstruction metrics while maintaining high efficiency and simplicity.

Surface normal estimation is a fundamental task in point cloud processing and plays a crucial role in downstream applications. Existing methods typically extract features from local neighborhoods or patches, followed by surface fitting or direct regression to predict normals. However, the scale ambiguity in determining the optimal neighborhood hinders effective extraction of geometric information, making normal estimation for unstructured point clouds with significant density variations particularly challenging. To address this challenge, we propose APN-Net, an adaptive perception network for point cloud normal estimation. Specifically, we design the Graphical Information Self-perception (GIS) module, which provides an implicit manner for region partitioning and expands the receptive field, enabling automatic extraction of both local geometric details and global structural information, while alleviating the scale ambiguity in determining the optimal neighborhood. Moreover, to capture complex geometric details, we introduce the Adaptive Graph Convolution (AGC) module, which employs adaptive kernels to model relationships among points across different semantic regions, thereby enabling richer feature representation. Extensive experiments on both synthetic and real-world scanned datasets demonstrate that APN-Net achieves superior performance in unoriented normal estimation, particularly for point clouds with significant density variations.

NeuralBoneReg: An instance-specific label-free point cloud-based method for multi-modal bone surface registration.

• We developed a diffusion-based generative framework called SEN, designed for sketch-guided local shape editing. This framework ensures consistency in geometric coherence and preservation. • We introduced an attention fusion module that effectively connects the cross-modal conditions to produce the final shape editing results. • We developed a sketch generator to directly generate sketch images from point clouds to facilitate our SEN training. Modeling complex 3D shapes from 2D sketch images has gained significant attention due to recent advancements in deep generative models. While many efforts have focused on producing geometries that faithfully reflect sketches, there has been limited exploration of sketch-guided local editing. Existing shape generation techniques often modify not only the intended edited areas but also the unedited parts, compromising geometric coherence and fairness among different parts of the shape. This paper introduces a shape editing network (SEN) under the diffusion-based generative framework to specifically enhance consistency in sketch-guided local shape editing. Our core insight is capturing partial geometric structure correlations to achieve consistent editing results. To accomplish this, the SEN is designed to conditionally learn local edits of a partial shape while considering the geometries of the unedited parts and the sketch image. We also devise an attention fusion module that progressively blends and aligns cross-modal conditioning streams between the latent variables of the sketch and point cloud representations. Additionally, by incorporating a guided sampling strategy for diffusion modeling, our SEN allows for more flexible control, even with complex or limited conditioning patterns. Our method significantly improves the geometric coherence of the synthesized parts while preserving the shapes of unedited areas. We demonstrate the advantages of our approach qualitatively and quantitatively across various shape categories, outperforming state-of-the-art methodologies.

Dense matching algorithms face challenges in disparity discontinuity regions and weak/repetitive texture areas, leading to surface irregularities, geometric detail loss, and void artifacts in oblique photogrammetric 3D models. To address these limitations, this study proposes a novel geometry optimization framework guided by image-space semantic edge features through geometric mapping between 2D imagery and 3D models. Our methodology initiates with extracting structural point clouds from building models through geometric-topological analysis and multi-view texture coherence assessment. A hybrid feature extraction strategy combines Line Segment Detector (LSD)-based structural line detection with deep learning-enhanced edge characterization across multi-perspective imagery. The framework implements a multi-stage refinement process: 1) Depth-First Search (DFS)-based edge component labeling with multi-view geometric constraints, 2) Non-structural component rejection through attribute-driven filtering and projective geometry analysis, 3) Adaptive dynamic buffer zones with Markov Random Field (MRF) optimization for optimal structural line association, and 4) Epipolar-constrained position refinement using multi-homography geometric verification. Experimental validation demonstrates our method's superiority in architectural point cloud localization accuracy, achieving 20.43% average reduction in point-to-mesh distance compared to conventional approaches, and showing significant improvements in surface continuity and void reduction. The framework mitigates local deformations while preserving architectural details, establishing a robust solution for photogrammetric 3D model optimization.

Generalizable task-oriented object grasping through LLM-guided ontology and similarity-based planning

Automated as-built rebar reconstruction and quality assessment from point clouds via Elastic Cylindrical Growth

Leaf area extraction framework: Transformer-based segmentation and geometric-based completion approaches for accurate extraction of field maize leaf area from point clouds

3D Gaussian Splatting (3DGS) has been explored for surface reconstruction, however unstructured and discontinuous Gaussian point clouds lead to uneven surface reconstruction accuracy as well as frequent loss of Novel View Synthesis (NVS) quality. To address this problem, we propose a scale-aligned monocular depth-guided 3DGS, a promising novel framework that combines geometric prior regularization and consistency supervision to achieve high-quality rendering and surface reconstruction. Specifically, monocular depth, estimated by some recently proposed monocular depth estimation models, contain implicitly abundant valuable geometric cues, but scale ambiguity limits its application. Therefore we first propose a $K$K-Nearest Neighbor (KNN)-based depth alignment framework that utilizes the full-domain gradient at monocular depth map to align to the sparse point cloud obtained during the Structure from Motion (SfM), which is employed for regularization to enhance geometric representation. Then a pseudo-mesh-based multi-view consistency module is introduced to fine-tune and guide the model to recover the accurate surface. Finally, a pixel-level isotropic gradient aware method guides the appropriate growth of the Gaussians to further improve the surface and rendering quality. Experiments on dozens of indoor, outdoor, and object-centered/non-object-centered datasets demonstrate that our method achieves accurate surface reconstruction with excellent NVS performance.

Finite element model reconstruction of welded tubular joints from incomplete point cloud scans

Serialized point cloud segmentation with dilated patchwise attention for generating geometric twins of shield metro tunnels

HyPCV-Former: Hyperbolic spatio-temporal transformer for 3D point cloud video anomaly detection