This paper presents a critical analysis of image-based 3D reconstruction using neural radiance fields (NeRFs), with a focus on quantitative comparisons with respect to traditional photogrammetry. The aim is, therefore, to objectively evaluate the strengths and weaknesses of NeRFs and provide insights into their applicability to different real-life scenarios, from small objects to heritage and industrial scenes. After a comprehensive overview of photogrammetry and NeRF methods, highlighting their respective advantages and disadvantages, various NeRF methods are compared using diverse objects with varying sizes and surface characteristics, including texture-less, metallic, translucent, and transparent surfaces. We evaluated the quality of the resulting 3D reconstructions using multiple criteria, such as noise level, geometric accuracy, and the number of required images (i.e., image baselines). The results show that NeRFs exhibit superior performance over photogrammetry in terms of non-collaborative objects with texture-less, reflective, and refractive surfaces. Conversely, photogrammetry outperforms NeRFs in cases where the object’s surface possesses cooperative texture. Such complementarity should be further exploited in future works.

Neural Radiance Field methods are innovative solutions to derive 3D data from a set of oriented images. This paper introduces new real and synthetic image datasets - called NeRFBK - specifically designed for testing and comparing NeRF-based 3D reconstruction algorithms. More and more reconstruction algorithms and techniques are available nowadays, raising the need to evaluate and compare the quality of derived 3D products currently used in various domains and applications. However, gathering diverse data with precise ground truth is challenging and may not encompass all relevant applications. The NeRFBK dataset addresses this issue by providing multi-scale, indoor and outdoor datasets with high-resolution images and videos and camera parameters for testing and comparing NeRF-based algorithms. This paper presents the design and creation of the NeRFBK set of data, various examples and application scenarios, and highlights its potential for advancing the field of 3D reconstruction.

Recent advancements in Virtual Reality (VR) technology have extended its applications beyond entertainment, offering promising tools for professional fields such as 3D data annotation. This paper explores the use of VR for labelling 3D point clouds in forestry and cultural heritage datasets. We employ Labelling Flora, an open-source VR annotation tool, to re-label three existing cultural heritage and one forestry datasets and assess the effectiveness of VR-based annotations in training machine learning models. By comparing the accuracy, precision, and F1-score of inference models trained with VR-generated labels to those trained with traditional desktop labelling methods, we evaluate the potential of VR to streamline labour-intensive annotation tasks. Our results indicate that VR enables intuitive 3D segmentation, even for individuals without technical expertise, particularly for very complex scenes, improving labelling efficiency and contributing to the overall automation of complex datasets. This study highlights therefore the potential of VR to enhance other workflows and make complex tasks more accessible to domain experts who may not be familiar with 3D data thus refining data accuracy and reliability.

In many countries, recent boosts in the construction and renovation sectors and energy efficiency directives are driving a growing interest in the built environment among designers and maintainers. In this context, customized software solutions tailored for Building Information Modelling (BIM) and Building Energy Modelling (BEM) are proving to be indispensable for optimizing operational efficiency within the Architecture, Engineering, Construction, Owner, and Operator (AECOO) sector and for facilitating the generation of buildings' Digital Twins (DTs). These DTs rely on accurate geometry and ancillary information (semantics, sensors, etc.) to define properties of single elements, enabling crucial simulations in structural conditions or energy needs. However, BIM and BEM model creation and their enrichment with semantic information are highly time-consuming and prone to manual errors. Hence, there is an increasing demand for automatic methods featuring a high level of geometric accuracy to reconstruct building elements, such as walls, floors, and openings captured via 3D reality-based surveying. This paper introduces an automated method for creating Boundary Representation (B-Rep) models from 3D surveying data for the generation of digital building replicas. The method is based on the detection and computation of topological elements from 3D reality-based point clouds. It proves valuable for architectural or design workflows and for conducting energy or quality system simulations.

3D point clouds feature valuable geometric and, often, radiometric and semantic information to support studies, analyses and understanding of the surveyed scene. Due to their generally large size, the use and interpretation of point clouds could be problematic. User-friendly and quick approaches for querying these valuable datasets and retrieving information could surely support end-users, in particular in the heritage sector. This work presents an ontology-based approach to facilitate the query and use of 3D heritage point clouds by means of sets of rules in order to infer properties and characteristics of the surveyed scene. Our approach is focused on linking together 3D spatial data and expert knowledge, in a way that the ontology can elaborate, represent, enrich and query a given point cloud. Results show how different queries can be set-up and how the procedure can be replicated to various queries and datasets.

This study proposes an innovative methodology for documenting and semantically analysing cultural heritage by integrating artificial intelligence (AI) with a photogrammetric 3D model. The case study is the Trajan’s Column in Rome, a monumental structure adorned with a continuous helical relief depicting Emperor Trajan’s Dacian campaigns. AI-driven semantic segmentation is used to identify key elements (such as human figures, battle scenes and natural motifs) within the digitised sculptural narrative. Starting from a high-resolution photogrammetric 3D model, the column’s texture is divided into multiple segments and a multimodal large language model (MLLM) is applied to produce context-aware segmentation masks via natural language prompts. Results are then projected onto the 3D geometry and visualised through a web-based 3D viewer.