Point cloud classification is a fundamental task in 3D applications. However, it challenging to achieve effective feature learning due the irregularity and unordered nature of point clouds. Lately, Transformers have been adopted improve processing. Nevertheless, massive Transformer layers tend incur huge computational memory costs. This paper presented novel hierarchical framework that incorporated convolutions with for classification, named Convolution-Transformer Network (3DCTN). It...

Hough voting, as has been demonstrated in VoteNet, is effective for 3D object detection, where voting a key step. In this paper, we propose novel VoteNet-based detector with vote enhancement to improve the detection accuracy cluttered indoor scenes. It addresses limitations of current schemes, i.e., votes from neighboring objects and background have significant negative impacts. Before replace classic MLP proposed Attentive (AMLP) backbone network get better feature description seed points....

Transformers have been at the heart of Natural Language Processing (NLP) and Computer Vision (CV) revolutions. The significant success in NLP CV inspired exploring use point cloud processing. However, how do cope with irregularity unordered nature clouds? How suitable are for different 3D representations (e.g., point- or voxel-based)? competent various processing tasks? As now, there is still no systematic survey research on these issues. For first time, we provided a comprehensive overview...

In the aircraft manufacturing industry, rivet inspection is a vital task for structure stability and aerodynamic performance. this article, we propose novel framework fully automated flush measurement, which key step in task. To efficiently perform first develop mobile 3-D scanning system to automatically capture point cloud of skin surface. Subsequently, regions are extracted through processing techniques. Instead relying on handcrafted features, data-driven approach extraction via...

Urban digital twins are virtual replicas of cities that use multi-source data and analytics to optimize urban planning, infrastructure management, decision-making. Towards this, we propose a framework focused on the single-building scale. By connecting cloud mapping platforms such as Google Map Platforms APIs, by leveraging state-of-the-art multi-agent Large Language Models analysis using ChatGPT(4o) Deepseek-V3/R1, our Gaussian Splatting-based mesh extraction pipeline, Digital Twin...

Global inspection of large-scale tunnels is a fundamental yet challenging task to ensure the structural stability and driving safety. Advanced LiDAR scanners, which sample into 3D point clouds, are making their debut in Tunnel Deformation Inspection (TDI). However, acquired raw clouds inevitably possess noticeable occlusions, missing areas, noise/outliers. Considering tunnel as geometrical sweeping feature, we propose an effective deformation algorithm by extracting global spatial axis from...

Point cloud segmentation is one of the most important tasks in LiDAR remote sensing with widespread scientific, industrial, and commercial applications. The research thereof has resulted many breakthroughs 3D object scene understanding. Existing methods typically utilize hierarchical architectures for feature representation. However, commonly used sampling grouping networks are not only time-consuming but also limited to point-wise coordinates, ignoring local semantic homogeneity point...

Abstract The intricacy of 3D surfaces often results cutting‐edge point cloud denoising (PCD) models in surface degradation including remnant noise, wrongly‐removed geometric details. Although using multi‐scale patches to encode the geometry a has become common wisdom PCD, we find that simple aggregation extracted features can not adaptively utilize appropriate scale information according around noisy points. It leads degradation, especially for points close edges and on complex curved...

In the semiconductor manufacturing industry, gold wires are commonly used to connect integrated circuits for transmitting signal and power. Hence, wire bonding quality inspection is a vital task functional stability of circuit chip. The shape each one most important factors influencing chip performance. However, as size quite small (around 0.02 mm), it difficult quantitatively measure shape, i.e., size-related indices. this article, we propose novel framework fully automated measurement,...

3D Transformers have achieved great success in point cloud understanding and representation. However, there is still considerable scope for further development effective efficient large-scale LiDAR scene segmentation. This paper proposes a novel Transformer framework, named Learnable Supertoken (3DLST). The key contributions are summarized as follows. Firstly, we introduce the first Dynamic Optimization (DSO) block token clustering aggregating, where learnable supertoken definition avoids...

Aircraft classification via remote sensing images has many commercial and military applications. The Swin-Transformer shown great promise, recently dominating general purpose image benchmarks such as ImageNet. In this manuscript, we test whether the performance of on translate to domain specific aircraft using Multi-Type from Remote Sensing Images dataset. We also investigate effect training procedure vs. model selection validation score. Our carefully trained achieved an impressive 99.4 %...

This paper applied a transformer based deep learning model 3D Point Cloud Transformer (3DPCT) to conduct tree species classification of Airborne LiDAR data. There are total 1291 single point clouds 11 different from coniferous and deciduous used in this paper. The integrated the local global feature modules both pointwise channel-wise, which provide promising results classification. We also investigate by adding more channels can be improved. Different number points per each sample as input...

3D urban scene reconstruction and modelling is a crucial research area in remote sensing with numerous applications academia, commerce, industry, administration. Recent advancements view synthesis models have facilitated photorealistic solely from 2D images. Leveraging Google Earth imagery, we construct Gaussian Splatting model of the Waterloo region centered on University are able to achieve view-synthesis results far exceeding previous based neural radiance fields which demonstrate our...

Recently, point cloud processing and analysis have made great progress due to the development of 3D Transformers. However, existing Transformer methods usually are computationally expensive inefficient their huge redundant attention maps. They also tend be slow requiring time-consuming sampling grouping processes. To address these issues, we propose an efficient TransFormer with Dynamic Token Aggregating (DTA-Former) for representation processing. Firstly, Learnable Sparsification (LTS)...