3D Representation Research Articles

Text-to-building: experiments with AI-generated 3D geometry for building design and structure generation

The paper seeks to investigate novel potentials for building design and structure generation that arise at the intersection of computational design and AI-generated 3D geometries. Although the use of AI technologies is exponentially increasing inside the architectural discipline, the design of spatial building configurations using AI-generated 3D geometries is still limited in its applications and represents an ongoing field of investigation in advanced architectural research. In this regard, several questions still need to be answered: how can we design new building typologies from AI-generated 3D geometries? And how can we use these typologies to shape both the real and the virtual world?The paper proposes a new approach to architectural design where artificial intelligence is used as the starting point for design exploration, while computational design procedures are employed to convert AI-generated 3D geometries into building elements – such as columns, beams, horizontal and vertical surfaces. The paper starts with a general overview of the current use of artificial intelligence inside the architectural discipline, and then it moves towards the explanation of specific AI generative models for 3D geometry reconstruction and representation. Subsequently, the proposed working pipeline is analysed in more detail – from the creation of 3D geometries using generative AI models to the conversion of such geometries into building elements that can be further designed and optimised using computational design tools and methods. The results shown in the paper are achieved using Shap-E as the main AI model, though the proposed pipeline can be implemented with multiple AI models. The paper ends by showing some of the generated results, finally adding some considerations to the relationship between human and artificial creativity inside the architectural discipline.The work presented in the paper suggests that the use of computational design tools and methods combined with the tectonics of the latent space opens new opportunities for topological and typological explorations. In a time where traditional architectural typologies are moving towards stagnation due to their inability to satisfy new human needs and ways of living, exploring AI-based working pipelines related to architectural design allows the definition of new design solutions for the generation of new architectural spaces. In doing so, the serendipitous aspect of AI biases is used as an auxiliary force to inform design decisions, promoting the discovery of a new inbuilt dynamism between human and artificial creativity. In a time where AI is everywhere, understanding the measure of such dynamism represents a key aspect for the future of the architectural discipline.

Abundant Soliton Solutions to the Generalized Reaction Duffing Model and Their Applications

The main aim of this study is to obtain soliton solutions of the generalized reaction Duffing model, which is a generalization for a collection of prominent models describing various key phenomena in science and engineering. The equation models the motion of a damped oscillator with a more complex potential than in basic harmonic motion. Two effective techniques, the mapping method and Bernoulli sub-ODE technique, are used for the first time to obtain the soliton solutions of the proposed model. Initially, the traveling wave transform, which comes from Lie symmetry infinitesimals, is applied, and a nonlinear ordinary differential equation form is derived. These approaches effectively retrieve a hyperbolic, Jacobi function as well as trigonometric solutions while the appropriate conditions are applied to the parameters. Numerous innovative solutions, including the kink wave, anti-kink wave, bell shape, anti-bell shape, W-shape, bright, dark and singular shape soliton solutions, were produced via the mapping and Bernoulli sub-ODE approaches. The research includes comprehensive 2D and 3D graphical representations of the solutions, facilitating a better understanding of their physical attributes and proving the effectiveness of the proposed methods in solving complex nonlinear equations. It is important to note that the proposed methods are competent, credible and interesting analytical tools for solving nonlinear partial differential equations.

A methodological approach for enhancing visualization of country data representation in the presence of significant spatial disparity

In this article, we present a methodological approach to address spatial disparity in global data representation, introducing an algorithm called Flexible Mapping to Understand Spatial Analysis (FLEMUSA). We utilize world maps to depict various data points across countries, revealing substantial variation among them. However, conventional choropleth maps often fail to effectively represent regions with sparse data, obscuring valuable insights. To mitigate this issue, we propose interactive graphical methods in both two and three dimensions, implemented through open-source Python code accessible via Google Colab. Our approach includes several contributions such as excluding countries without data from the representation, scaling magnitudes within country borders, focusing on regional analysis, and using logarithmic scales for bubble maps proportional to country sizes. Additionally, we offer interactive 2D and 3D representations, rotatable 3D representations, and zoomable options, facilitating enhanced visualization of regional similarities amidst data heterogeneity. Through this algorithm, we aim to improve the clarity and interpretability of spatial data analysis, integrating solutions for extreme data overdispersion, all programmed with open-source code.-Utilization of world maps for visual representation of data across countries mitigating the overdispersion step by step.-Implementation of graphical methods, including interactive 2D and 3D maps, to address spatial disparity.-Provision of open-source code for customizable graphical representations, facilitating implementation in online journals as interactive code snippets.

O-307 Decrease of human sperm hypercondensed chromatin volume after slow freezing revealed by three-dimensional (3D) nuclear morphometry

Abstract Study question Are the 3D morphometry parameters of the human sperm nucleus affected by slow freezing? Summary answer The slow freezing-thawing cycle induces significant alterations on hypercondensed nucleus zone by reducing its volume and splits it into several small zones. What is known already Sperm cryopreservation is an essential technique for infertility care, but it seems to highly affect the genetic and epigenetic integrity of sperm nuclei with poorly identified mechanisms. Moreover, it is known that sperm nuclear alterations may be involved in pregnancy failures. Therefore, sperm nuclear biomarkers are an essential issue to assess the outcome of ART procedures when frozen sperm is used. Thereby, 3D visualization of chromatin is a good way to understand modifications in the organization and composition of sperm chromatin after a slow freezing procedure. Study design, size, duration Ten semen of patients were analyzed (Germetheque biobank) after informed consent. Standard semen parameters, DNA fragmentation, chromatin decondensation, sperm telomere length (STL) and 3D nuclear morphometry were performed before and after slow freezing. Morphometric measurements were performed from a digital representation of the sperm nucleus in 3 dimensions that allow us to measure the volume, flatness and elongation of each sperm nucleus and characterize hypercondensed zones, with live/dead spermatozoa distinction. Participants/materials, setting, methods After addition of cryoprotective medium v/v (Cryosperm®), sperm was frozen with a programmable freezer (NanoDigitcool®). Standards sperm parameters were measured in accordance with WHO guidelines 2021, DNA fragmentation by flow cytometry TUNEL assay, chromatin decondensation by epifluorescent microscopy with Chromomycin A3 and STL using qPCR. 3D nuclear representation was performed by a z-axis image acquisition series of DAPI/PI stained slides with optigrid epifluorescent microscopy. Analyses were carried out using the NucleusJ2.0 software with NODeJ plugin. Main results and the role of chance Freezing induces an alteration of sperm vitality (p < 0.0001) and progressive and total motility (p < 0.001 and p < 0.0001 respectively) and an increase in the percentage of sperm with DNA fragmentation (18.30% fresh vs 31.6% frozen, p < 0.05). We did not observe a modification of the median volume of the sperm nucleus before and after freezing (37.15 [IC = 36.5;38.14] and 37.10 [33.27;37.78] µm3 respectively), as the same for flatness and elongation (2.48 vs. 2.50 a.u. and 2.04 vs. 2.04 a.u. respectively). Nevertheless, five samples showed significant alterations of nucleus volume (p < 0.0001), due to diminution of flatness and more or less elongated. A significant decrease of the total volume of the hypercondensed chromatin zone was observed after freezing (0.94 [0.84;0.99] before vs. 0.61 [0.53;0.69] µm3 after freezing-thawing, p < 0.0001), with an effect size of -1.48 [-2.10; -0.86]. Decrease of total volume of hypercondensed region was measured homogeny in all the samples tested, particularly on live (1.00 vs. 0.68 µm3 respectively, p < 0.0001) compared with dead spermatozoa (0.72 vs. 0.57 µm3, p < 0.05). We showed also a significant increase of the number of hypercondensed zones in each nucleus (p < 0.001), reflecting a split into several parts of the hypercondensed chromatin. Limitations, reasons for caution A 3D-FISH analysis would allow to characterize the hypercondensed chromatin portion, probably as heterochromatin and which we suspect to be the nucleus ring. Moreover, a chromosomes specific marking would allow for the identification of whether whole chromosomes or specific parts of chromosomes (centromeres or telomeres) are affected by this decompaction. Wider implications of the findings This study is the first to make exacts measurements of human sperm nucleus because of a non-decondensing method. We demonstrate that slow freezing affects hypercondensed chromatin probably through a decondensation and modification of the nucleus-specific organization. This new assessment will allow a better proficiency of men fertility preservation techniques. Trial registration number NCT04715828

Characterizing parameters and incorporating action potentials via the Hodgkin-Huxley model in a novel electric model for living cells

ABSTRACT To enhance our understanding of electroporation and optimize the pulses used within the frequency range of 1 kHz to 100 MHz, with the aim of minimizing side effects such as muscle contraction, we introduce a novel electrical model, structured as a 2D representation employing exclusively lumped elements. This model adeptly encapsulates the intricate dynamics of living cells’ impedance variation. A distinguishing attribute of the proposed model lies in its capacity to decipher the distribution of transmembrane potential across various orientations within living cells. This aspect bears critical importance, particularly in contexts such as electroporation and cellular stimulation, where precise knowledge of potential gradients is pivotal. Furthermore, the augmentation of the proposed electrical model with the Hodgkin-Huxley (HH) model introduces an additional dimension. This integration augments the model’s capabilities, specifically enabling the exploration of muscle cell stimulation and the generation of action potentials. This broader scope enhances the model’s utility, facilitating comprehensive investigations into intricate cellular behaviors under the influence of external electric fields.

Photogrammetry of airborne image set at high landslide disaster risk region using Meshroom and Blender

Abstract We report the application of the photogrammetry method to reconstruct three-dimensional topographic mesh at a high disaster risk region. This work is aimed to identify the landslide potencies of the area by means of slope analysis of the given area around the Bandung Basin rim. Our method proposes a low-cost aerial survey and an effective algorithm in order to map the local landslide tendencies. In order to perform image processing and 3D representation we utilized the photogrammetric pipeline in the Meshroom software. The post-processing of the 3D model was performed using Blender. In this research, we use a commercial camera drone as an image capturing device. As a test case we processed 183 images of an area of 51 m². The image-processing takes about 39 minutes on a 12 cores computer accompanied with a GPU card. From this set of images, we have yielded a three-dimensional slope structure comprising a total of 85,850 mesh nodes. As the result of this research, we propose the optimization of the image set and a list of post processing works of the nodes that can be used in order to characterize landslide disaster risk of the disaster-prone region in a fast and accurate manner.

Assisting Directional Drilling by Calculating a Safe Operating Envelope

Summary Nowadays, complex 3D trajectories are executed with a succession of circular arcs (CAs). Although they have constant curvature, their tool face is not constant. Consequently, directional drillers must adjust the tool face regularly to reach the target entry within its tolerances. This paper investigates the use of the constant curvature and constant tool face (CTC in short) curve as an alternative to the CA to assist the directional drilling work to reach the target entry within its boundaries. The problem is addressed by calculating a safe operating envelope (SOE) to reach the boundaries of the target entry and provide a tolerance window for the curvature and tool face to support directional drilling decisions. The target entry tolerance is discretized as a polygon. From the current bit position and its direction, the possible choices of curvatures and tool faces are obtained to reach the edges of the target entry shape. The SOE can be calculated with the CA or with the CTC curve. It is, therefore, possible to compare the advantages and disadvantages of both types of curves to attain the target entry and stay within its boundaries. The CA is shorter than the CTC curve. However, it requires adjusting the tool face during the navigation, which is not the case with the CTC curve. As a result, the directional driller can control the bottomhole assembly (BHA) direction such that the well lands within the target entry limits by using set points for tool face and curvature inside the calculated SOE. Furthermore, a new way to represent the SOE is introduced. It makes use of a 3D cylindrical representation where the curvature is mapped as the height of a cylinder, while the tool face corresponds to the azimuth in the cylindrical coordinate system, and the length is linked to the radial distance. This provides a visual aid to understand the SOE. Moreover, this visualization helps to appreciate the relationship between the initial bit location and direction in the construction of the SOE and how the margins increase in a particular manner as the bit approaches the target entry polygon. The CTC curve is the natural one followed by directional positive displacement motors (PDMs) or rotary steerable systems (RSS). Potentially, the CTC curve may be a more straightforward solution to automated directional drilling control because it is easier to be followed by both PDM and RSS.

Uni MS-PS: A multi-scale encoder-decoder transformer for universal photometric stereo

Photometric Stereo (PS) addresses the challenge of reconstructing a three-dimensional (3D) representation of an object by estimating the 3D normals at all points on the object’s surface. This is achieved through the analysis of at least three photographs, all taken from the same viewpoint but with distinct lighting conditions. This paper introduces a novel approach for Universal PS, i.e., when both the active lighting conditions and the ambient illumination are unknown. Our method employs a multi-scale encoder–decoder architecture based on Transformers that allows to accommodates images of any resolutions as well as varying number of input images. We are able to scale up to very high resolution images like 6000 pixels by 8000 pixels without losing performance and maintaining a decent memory footprint. Moreover, experiments on publicly available datasets establish that our proposed architecture improves the accuracy of the estimated normal field by a significant factor compared to state-of-the-art methods. Code and dataset available at: https://clement-hardy.github.io/Uni-MS-PS/index.html.

Improved source classification and performance analysis using Gaia DR3

The Discrete Source Classifier (DSC) provides probabilistic classification of sources in Gaia Data Release 3 (GDR3) using a Bayesian framework and a global prior. The DSC Combmod classifier in GDR3 achieved for the extragalactic classes (quasars and galaxies) a high completeness of 92%, but a low purity of 22% (all sky, all magnitudes) due to contamination from the far larger star class. However, these single metrics mask significant variation in performance with magnitude and sky position. Furthermore, a better combination of the individual classifiers that comprise Combmod is possible. Here we compute two-dimensional (2D) representations of the completeness and the purity as a function of Galactic latitude and source brightness, and also exclude the Magellanic Clouds where stellar contamination significantly reduces the purity. Reevaluated on a cleaner validation set and without introducing changes to the published GDR3 DSC probabilities themselves, we here achieve for Combmod average 2D completenesses of 92% and 95% and average 2D purities of 55% and 89% for the quasar and galaxy classes, respectively. Since the relative proportions of extragalactic objects to stars in Gaia is expected to vary significantly with brightness and latitude, we then introduce a new prior that is a continuous function of brightness and latitude, and compute new class probabilities from the GDR3 DSC component classifiers, Specmod and Allosmod. Contrary to expectations, this variable prior only improves the performance by a few percentage points, mostly at the faint end. Significant improvement, however, is obtained by a new additive combination of Specmod and Allosmod. This classifier, Combmod-α, achieves average 2D completenesses of 82% and 93% and average 2D purities of 79% and 93% for the quasar and galaxy classes, respectively, when using the global prior. Thus, we achieve a significant improvement in purity for a small loss of completeness. The improvement is most significant for faint quasars (G≥20) where the purity rises from 20% to 62%.

Efficient graph representation in graph neural networks for stress predictions in stiffened panels

Machine learning (ML) and deep learning (DL) techniques have gained significant attention as reduced order models (ROMs) to computationally expensive structural analysis methods, such as finite element analysis (FEA). Graph neural network (GNN) is a particular type of neural network which processes data that can be represented as graphs. This allows for efficient representation of complex geometries that can change during the conceptual design of a structure or a product. In this study, we propose a novel graph embedding for the efficient representation of 3D stiffened panels by considering separate plate domains as vertices. This approach is considered using Graph Sampling and Aggregation (GraphSAGE) to predict stress distributions in stiffened panels with varying geometries. A comparison between a finite element-vertex graph representation is conducted to demonstrate the effectiveness of the proposed approach. A comprehensive parametric study is performed to examine the effect of structural variables on stress predictions. A wide range of geometries is considered, material nonlinearity, a few boundary conditions, together with uniform and patch loading at various positions. The study involves straight and curved panels with uni- and bi-directional stiffeners. The proposed unit-vertex representation of the panel requires only about 2% of GPU memory and about 4% of training time in comparison to a finite element-vertex embedding. The GraphSAGE model with the proposed unit-vertex representation accurately captures stress distribution across all panels, achieving an average prediction accuracy of 92.3% for the maximum von Mises stress. Our results demonstrate the immense potential of graph neural networks with the proposed graph embedding as a robust reduced-order model for 3D structures.

Applications of augmented reality (AR) in chemical engineering education: Virtual laboratory work demonstration to digital twin development

With advances in information technologies, virtual technologies have become a viable option for chemical engineering educators. Although traditional teaching methods remain effective, they have many limitations when presenting complex phenomena. Virtual technologies like Augmented Reality (AR) have shown great potential as an addition to traditional teaching methods. This study explored the practicality and potential applications of AR technologies in chemical engineering education. In this work, traditional 2D representations of CFD simulation results were replaced by AR experiences to visualise the flow pattern and assist the understanding of the working mechanism behind different devices. Other than CFD result integration, animations were also included in AR experiences for personalised training and inclusive laboratory work experience. AR-assisted training has shown good potential in reducing financial loss and equipment downtime due to improper handling. Besides delivering teaching content, the development process of a digital twin with an AR interface was used to demonstrate how virtual technologies can be involved in digitalisation in Industry 4.0. Beyond the technical contents, important design philosophies were also taught as part of the courses.

MolLM: a unified language model for integrating biomedical text with 2D and 3D molecular representations.

The current paradigm of deep learning models for the joint representation of molecules and text primarily relies on 1D or 2D molecular formats, neglecting significant 3D structural information that offers valuable physical insight. This narrow focus inhibits the models' versatility and adaptability across a wide range of modalities. Conversely, the limited research focusing on explicit 3D representation tends to overlook textual data within the biomedical domain. We present a unified pre-trained language model, MolLM, that concurrently captures 2D and 3D molecular information alongside biomedical text. MolLM consists of a text Transformer encoder and a molecular Transformer encoder, designed to encode both 2D and 3D molecular structures. To support MolLM's self-supervised pre-training, we constructed 160K molecule-text pairings. Employing contrastive learning as a supervisory signal for learning, MolLM demonstrates robust molecular representation capabilities across four downstream tasks, including cross-modal molecule and text matching, property prediction, captioning, and text-prompted molecular editing. Through ablation, we demonstrate that the inclusion of explicit 3D representations improves performance in these downstream tasks. Our code, data, pre-trained model weights, and examples of using our model are all available at https://github.com/gersteinlab/MolLM. In particular, we provide Jupyter Notebooks offering step-by-step guidance on how to use MolLM to extract embeddings for both molecules and text.

Proposal of a domain model for 3D representation of buildings in Ecuador.

The accelerated urban sprawl of cities around the world presents major challenges for urban planning and land resource management. In this context, it is crucial to have a detailed 3D representation of buildings enriched with accurate alphanumeric information. A distinctive aspect of this proposal is its specific focus on the spatial unit corresponding to buildings. In order to propose a domain model for the 3D representation of buildings, the national standard of Ecuador and the international standard (ISO 19152:2012 LADM) were considered. The proposal includes a detailed specification of attributes, both for the general subclass of buildings and for their infrastructure. The application of the domain model proposal was crucial in a study area located in the Riobamba canton, due to the characteristics of the buildings in that area. For this purpose, a geodatabase was created in pgAdmin4 with official information, taking into account the structure of the proposed model and linking it with geospatial data for an adequate management and 3D representation of the buildings in an open-source Geographic Information System. This application improves cadastral management in the study region and has wider implications. This model is intended to serve as a benchmark for other countries facing similar challenges in cadastral management and 3D representation of buildings, promote efficient urban development and contribute to global sustainable development.

Utility of 3D modelling of the patient's living environment as perceived by occupational therapists.

Visiting a patient's living environment is important for occupational therapists, albeit costly and time consuming. MapIt is a mobile app producing a 3D representation of a home with the possibility of taking measurements. The purpose of this study was to explore the utility of a 3D representation of a patient's home for the clinical practice of occupational therapists. Case study in which the unit of analysis was the utility of MapIt as defined by ISO 9241-11:2018 and as perceived by occupational therapists in four different occupational therapy clinical settings (Canada). Onsite observations with 10 occupational therapists (and their patients) were triangulated with data from interviews, diaries, and logbooks. Inductive thematic condensation led to emerging conclusions for each clinical setting, fuelling the next case data collection and analysis. Inter-case analysis was corroborated by additional occupational therapists, through crowdsourcing and expert review. Occupational therapists' clinical reasoning was supported by the MapIt app, enhancing and streamlining their work and inducing adjustments to treatment plans. Occupational therapists saw and measured the patient's environment remotely, to better match person-environment-occupation and promote occupational engagement. MapIt's 3D representations were judged useful to communicate between occupational therapists and stakeholders, to educate, allow continuity, optimise resources, minimise the patient's time on a waitlist for homecare, and save time for everyone. MapIt allowed occupational therapists who performed home visits to bring a little of the patients' home to their office, whereas occupational therapists without access to the home could see it and take measurements. MapIt's utility was confirmed for practice in clinical settings and for better continuity of care between settings. MapIt makes it possible for occupational therapists to 'walk around' the patient's home remotely, but the possibility of measuring environmental elements is a 3D model's true added value over currently used photos or short videos.

STAFFormer: Spatio-temporal adaptive fusion transformer for efficient 3D human pose estimation

Existing two-stage methods for 3D Human Pose Estimation often use 2D poses as input, which are then lifted to obtain 3D representations. This typically involves frame-by-frame estimation, whether in 2D or 3D, resulting in high computational demands unsuitable for edge devices. Due to the continuity of human movement, the differences between adjacent frames can be minimal, a question arises: is frame-by-frame estimation necessary? Previous works demonstrated the feasibility of using Transformer-based models to estimate poses with sparse frames, focusing on either temporal or spatial dependencies but neglecting holistic spatio-temporal correlations. To address this, we introduce the Spatio-Temporal Adaptive Fusion Transformer (STAFFormer). First, STAFFormer recovers dense temporal frames from sparsely sampled ones obtained from a 2D pose estimator through the Temporal Dense Frame Recovery (TDFR) module. This significantly reduces the computational complexity. Second, STAFFormer employs an adaptive fusion attention mechanism, enhancing accuracy by attentively navigating both spatial and temporal dimensions through the Spatio-Temporal Adaptive Fusion (STAF) module. Furthermore, We introduce a kinematic coherence loss to adapt to subtle joint movements, improving pose estimation fidelity. Finally, we explore the possibility of integrating different pre-training strategies using extensive marker-based datasets. Experimental results on challenging datasets show our network achieves state-of-the-art performance with low computational complexity.

Numerical investigation of two-dimensional fuzzy fractional heat problem with an external source variable.

This study suggests a strategy for calculating the fuzzy analytical solutions to a two-dimensional fuzzy fractional-order heat problem including a diffusion variable connected externally. We propose Sawi residual power series scheme (SRPSS) which is the amalgamation of Sawi transform and residual power series scheme under the Caputo fractional differential operator. We demonstrate three different examples to derive the fuzzy fractional series solution which is characterized by its rapid convergence and easy finding of the unknown coefficients using the concept of limit at infinity. The most significant aspect of this scheme is that it derives the results without time effort compared with the traditional residual power series approach. Our findings confirm that the SRPSS is a robust and valuable method for approximating the solution of fuzzy fractional problems. Furthermore, we provide 2D and 3D symbolic representations to present the physical behavior of fuzzy fractional problems under the lower and upper bounded solutions.

Multiphase Reconstruction of Heterogeneous Materials Using Machine Learning and Quality of Connection Function.

Establishing accurate structure-property linkages and precise phase volume accuracy in 3D microstructure reconstruction of materials remains challenging, particularly with limited samples. This paper presents an optimized method for reconstructing 3D microstructures of various materials, including isotropic and anisotropic types with two and three phases, using convolutional occupancy networks and point clouds from inner layers of the microstructure. The method emphasizes precise phase representation and compatibility with point cloud data. A stage within the Quality of Connection Function (QCF) repetition loop optimizes the weights of the convolutional occupancy networks model to minimize error between the microstructure's statistical properties and the reconstructive model. This model successfully reconstructs 3D representations from initial 2D serial images. Comparisons with screened Poisson surface reconstruction and local implicit grid methods demonstrate the model's efficacy. The developed model proves suitable for high-quality 3D microstructure reconstruction, aiding in structure-property linkages and finite element analysis.

Automatic Classification of Submerged Macrophytes at Lake Constance Using Laser Bathymetry Point Clouds

Submerged aquatic vegetation, also referred to as submerged macrophytes, provides important habitats and serves as a significant ecological indicator for assessing the condition of water bodies and for gaining insights into the impacts of climate change. In this study, we introduce a novel approach for the classification of submerged vegetation captured with bathymetric LiDAR (Light Detection And Ranging) as a basis for monitoring their state and change, and we validated the results against established monitoring techniques. Employing full-waveform airborne laser scanning, which is routinely used for topographic mapping and forestry applications on dry land, we extended its application to the detection of underwater vegetation in Lake Constance. The primary focus of this research lies in the automatic classification of bathymetric 3D LiDAR point clouds using a decision-based approach, distinguishing the three vegetation classes, (i) Low Vegetation, (ii) High Vegetation, and (iii) Vegetation Canopy, based on their height and other properties like local point density. The results reveal detailed 3D representations of submerged vegetation, enabling the identification of vegetation structures and the inference of vegetation types with reference to pre-existing knowledge. While the results within the training areas demonstrate high precision and alignment with the comparison data, the findings in independent test areas exhibit certain deficiencies that are likely addressable through corrective measures in the future.

PointNet-based modeling of systematic distance deviations for improved TLS accuracy

Abstract Terrestrial laser scanners (TLSs) have become indispensable for acquiring highly detailed and accurate 3D representations of the physical world. However, the acquired data is subject to systematic deviations in distance measurements due to external influences, such as distance and incidence angle. This research introduces a calibration approach by applying a deep learning model based on PointNet to predict and correct these systematic distance deviations, incorporating not only the XYZ coordinates but also additional features like intensity, incidence angle, and distances within a local neighbourhood radius of 5 cm. By predicting and subsequently correcting systematic distance deviations, the quality of TLS point clouds can be improved. Hence, our model is designed to complement and build upon the foundation of prior internal TLS calibration. A data set collected under controlled environmental conditions, containing various objects of different materials, served as the basis for training and validation the PointNet based model. In addition our analysis showcase the model’s capability to accurately model systematic distance deviations, outperforming existing methods like gradient boosting trees by capturing the spatial relationships and dependencies within the data more effectively. By defining test data sets, excluded from the training process, we underscore the ongoing effectiveness of our model’s distance measurement calibration, showcasing its ability to improve the accuracy of the TLS point cloud.

MDF-DTA: A Multi-Dimensional Fusion Approach for Drug-Target Binding Affinity Prediction.

Drug-target affinity (DTA) prediction is an important task in the early stages of drug discovery. Traditional biological approaches are time-consuming, effort-consuming, and resource-consuming due to the large size of genomic and chemical spaces. Computational approaches using machine learning have emerged to narrow down the drug candidate search space. However, most of these prediction models focus on single feature encoding of drugs and targets, ignoring the importance of integrating different dimensions of these features. We propose a deep learning-based approach called Multi-Dimensional Fusion for Drug Target Affinity Prediction (MDF-DTA) incorporating different dimensional features. Our model fuses 1D, 2D, and 3D representations obtained from different pretrained models for both drugs and targets. We evaluated MDF-DTA on two standard benchmark data sets: DAVIS and KIBA. Experimental results show that MDF-DTA outperforms many state-of-the-art techniques in the DTA task across both data sets. Through ablation studies and performance evaluation metrics, we evaluate the importance of individual representations and the impact of each representation on MDF-DTA.