One of the primary challenges inherent in utilizing deep learning models is the scarcity and accessibility hurdles associated with acquiring datasets of sufficient size to facilitate effective training of these networks. This is particularly significant in object detection, shape completion, and fracture assembly. Instead of scanning a large number of real-world fragments, it is possible to generate massive datasets with synthetic pieces. However, realistic fragmentation is computationally intensive in the preparation (e.g., pre-factured models) and generation. Otherwise, simpler algorithms such as Voronoi diagrams provide faster processing speeds at the expense of compromising realism. In this context, it is required to balance computational efficiency and realism. This paper introduces a GPU-based framework for the massive generation of voxelized fragments derived from high-resolution 3D models, specifically prepared for their utilization as training sets for machine learning models. This rapid pipeline enables controlling how many pieces are produced, their dispersion and the appearance of subtle effects such as erosion. We have tested our pipeline with an archaeological dataset, producing more than 1M fragmented pieces from 1,052 Iberian vessels ( Github ). Although this work primarily intends to provide pieces as implicit data represented by voxels, triangle meshes and point clouds can also be inferred from the initial implicit representation. To underscore the unparalleled benefits of CPU and GPU acceleration in generating vast datasets, we compared against a realistic fragment generator that highlights the potential of our approach, both in terms of applicability and processing time. We also demonstrate the synergies between our pipeline and realistic simulators, which frequently cannot select the number and size of resulting pieces. To this end, a deep learning model was trained over realistic fragments and our dataset, showing similar results. • GPU-based fragmentation of voxelized artefacts for simulating brittle fractures. • Generation of large datasets for training machine learning models. • Definition of how many fragments are produced, their dispersion and appearance. • Publication of a dataset comprising more than 1M fragments.

Accurate completion of archaeological artifacts is a critical aspect in several archaeological studies, including documentation of variations in style, inference of chronological and ethnic groups, and trading routes trends, among many others. However, most available pottery is fragmented, leading to missing textural and morphological cues. Currently, the reassembly and completion of fragmented ceramics is a daunting and time-consuming task, done almost exclusively by hand, which requires the physical manipulation of the fragments. To overcome the challenges of manual reconstruction, reduce the materials' exposure and deterioration, and improve the quality of reconstructed samples, we present IberianVoxel, a novel 3D Autoencoder Generative Adversarial Network (3D AE-GAN) framework tested on an extensive database with complete and fragmented references. We generated a collection of 1001 3D voxelized samples and their fragmented references from Iberian wheel-made pottery profiles. The fragments generated are stratified into different size groups and across multiple pottery classes. Lastly, we provide quantitative and qualitative assessments to measure the quality of the reconstructed voxelized samples by our proposed method and archaeologists' evaluation.

Tree modeling has been extensively studied in computer graphics. Recent advances in the development of high-resolution sensors and data processing techniques are extremely useful for collecting 3D datasets of real-world trees and generating increasingly plausible branching structures. The wide availability of versatile acquisition platforms allows us to capture multi-view images and scanned data that can be used for guided 3D tree modeling. In this paper, we carry out a comprehensive review of the state-of-the-art methods for the 3D modeling of botanical tree geometry by taking input data from real scenarios. A wide range of studies has been proposed following different approaches. The most relevant contributions are summarized and classified into three categories: (1) procedural reconstruction, (2) geometry-based extraction, and (3) image-based modeling. In addition, we describe other approaches focused on the reconstruction process by adding additional features to achieve a realistic appearance of the tree models. Thus, we provide an overview of the most effective procedures to assist researchers in the photorealistic modeling of trees in geometry and appearance. The article concludes with remarks and trends for promising research opportunities in 3D tree modeling using real-world data.

This work describes an efficient approach for generating large 3D thermal point clouds considering the occlusion of camera viewpoints. For that purpose, RGB and thermal imagery are first corrected and fused with an intensity correlation-based algorithm. Then, absolute temperature values are obtained from the normalized data. Finally, thermal imagery is mapped on the point cloud using the Graphics Processing Unit (GPU) hardware. The proposed occlusion-aware mapping algorithm is massively parallelized using OpenGL's compute shaders. Our solution allows generating dense thermal point clouds in a lower response time compared with other notable soft-ware solutions (e.g., Agisoft Metashape or Pix4Dmapper) that yield results with a significantly lower point density.

The widespread use of LiDAR technology in a multitude of domains has produced a growing availability of massive high-resolution point datasets that demand new approaches for efficient organization and storage, filtering using different spatio-temporal criteria, selective/progressive visualization, processing and analysis, and collaborative editing. Ideally, LiDAR data coming from multiple sources and organized in different datasets should be accessible in a simple, uniform, and ubiquitous way to comply with the FAIR principle proposed by the Open Geospatial Consortium: Findable, Accessible, Interoperable, and Reusable. With this goal in mind, we present SPSLiDAR, a conceptual model with a simple interface for repositories of LiDAR data that can be adapted to the needs of different applications. SPSLiDAR includes aspects, such as the arrangement of related datasets into workspaces on a world scale, support for overlapping datasets with different resolutions or acquired at different times, and hierarchical organization of point data, enabling levels of detail and selective download. We also describe in detail an implementation of this model aimed at visualization and downloading of large datasets using the MongoDB database. Finally, we show some experimental results of this implementation using real data, such as its space requirements, upload latency, access latency, and throughput.

In this work, we present an efficient graphics processing unit (GPU)-based light detection and ranging (LiDAR) scanner simulator. Laser-based scanning is a useful tool for applications ranging from reverse engineering or quality control at an object scale to large-scale environmental monitoring or topographic mapping. Beyond that, other specific applications require a large amount of LiDAR data during development, such as autonomous driving. Unfortunately, it is not easy to get a sufficient amount of ground-truth data due to time constraints and available resources. However, LiDAR simulation can generate classified data at a reduced cost. We propose a parameterized LiDAR to emulate a wide range of sensor models from airborne to terrestrial scanning. OpenGL’s compute shaders are used to massively generate beams emitted by the virtual LiDAR sensors and solve their collision with the surrounding environment even with multiple returns. Our work is mainly intended for the rapid generation of datasets for neural networks, consisting of hundreds of millions of points. The conducted tests show that the proposed approach outperforms a sequential LiDAR scanning. Its capabilities for generating huge labeled datasets have also been shown to improve previous studies.

Abstract Collision detection (CD) among complex objects in motion is an open question because of its algorithmic complexity. In this paper, we present a CD algorithm between a particle and a complex rigid solid. In order to represent solids, we use a simplicial covering scheme by means of 3‐simplices. With this representation system, complex polyhedra and closed triangle meshes can be represented and used in CD with a particle. A particle can be represented by a point, where the real dimensions of the particle are not important. This CD algorithm has been extended for a particle represented by a sphere which in turn represents an approximation to the particle dimensions. In order to efficiently classify the particles and the simplices of the polyhedra covering, we use a new hierarchical data structure named tetra‐tree that decomposes the space into tetra‐cones. These algorithms also use bounding volumes and temporal and geometric coherence, as well as incremental calculations in order to determine the collision in an efficient and exact way. Because of the use of sign operations with signed volumes and barycentric coordinates, we conclude that operations carried out using this method are more robust and efficient than those used in classic algorithms.