This paper describes a method for estimating the room impulse response (RIR) for a microphone and a sound source located at arbitrary positions from the 3d mesh data of the room. Simulating realistic RIRs with pure ph...
详细信息
This paper describes a method for estimating the room impulse response (RIR) for a microphone and a sound source located at arbitrary positions from the 3d mesh data of the room. Simulating realistic RIRs with pure physics-driven methods often fails the balance between physical consistency and computational efficiency, hindering application to real-time speech processing. Alternatively, one can use mesh2IR, a fast black-box estimator that consists of an encoder extracting latent code from meshdata with a graph convolutional network (GCN) and a decoder generating the RIR from the latent code. Combining these two approaches, we propose a fast yet physically coherent estimator with interpretable latent code based on differentiable digital signal processing (ddSP). Specifically, the encoder estimates a virtual shoebox room scene that acoustically approximates the real scene, accelerating physical simulation with the differentiable image-source model in the decoder. Our experiments showed that our method outperformedmesh2IR for real meshdata obtained with the depth scanner of Microsoft HoloLens 2, and can provide correct spatial consistency for binaural RIRs.
We propose a robust face identification method based on high saliency extraction. The adopted algorithm is performed on enhanced tri-dimensional datamesh. The enhancement stage aims to upgrade the quality of the scan...
详细信息
ISBN:
(纸本)9781509010554
We propose a robust face identification method based on high saliency extraction. The adopted algorithm is performed on enhanced tri-dimensional datamesh. The enhancement stage aims to upgrade the quality of the scanneddata by inhibiting noises, correcting missing information and smoothing the surface. Then, Gaussian curvatures and mean curvatures are calculated from principal curvature computation in furtherance of feature extraction. Next, referential curvature points are selected and utilized to perform the matching process. The results of our experimental essay have been evaluated by comparing them to similar advanced studies and have advantageously manifested bright levels of identification rates.
Metal forging is used to manufacture dies. We require the best set of input parameters for the process to be efficient. Currently, we predict the best parameters using the finite element method by generating simulatio...
详细信息
ISBN:
(纸本)9783031429408;9783031429415
Metal forging is used to manufacture dies. We require the best set of input parameters for the process to be efficient. Currently, we predict the best parameters using the finite element method by generating simulations for the different initial conditions, which is a time-consuming process. In this paper, introduce a hybrid approach that helps in processing and generating new data simulations using a surrogate graph neural network model based on graph convolutions, having a cheaper time cost. We also introduce a hybrid approach that helps in processing and generating new data simulations using the model. Given a dataset representing meshes, our focus is on the conversion of the available information into a graph or point cloud structure. This new representation enables deep learning. The predicted result is similar, with a low error when compared to that produced using the finite element method. The new models have outperformed existing PointNet and simple graph neural network models when applied to produce the simulations.
As one of the most significant local features, corners play a fundamental role in 3d computer vision and computer graphics. Two novel corner detection algorithms for 3d mesh data are presented in this Letter, in which...
详细信息
As one of the most significant local features, corners play a fundamental role in 3d computer vision and computer graphics. Two novel corner detection algorithms for 3d mesh data are presented in this Letter, in which the authors examine the geometric properties of the surface in multi-scale space and utilise simple triangle principle to detect mesh corners. Numerical experiments show that our corner detectors outperform six state-of-the-art methods.
Studying cultural heritage artefacts, using 3ddigital models, is gaining interest. It not only allows applications in documentation and visualisation, but also permits further contact-less examination. In this paper,...
详细信息
Studying cultural heritage artefacts, using 3ddigital models, is gaining interest. It not only allows applications in documentation and visualisation, but also permits further contact-less examination. In this paper, we are presenting a statistical analysis of stone engravings based on features that were semiautomatically extracted from 3d acquisition data. Our objects of study are two Neolithic stone stelae and a faithful replica that was created in the course of an archaeological study. We use common statistical methods and investigate the populations of depth anddiameter of the engraving traces, as well as their correlation. We observe that the erosion of the two prehistoric stelae results in a larger dispersion of the diameter distribution, as well as in a similarity of a linear regression model. Furthermore, we note a similar median in the height anddepth distribution between the replica and only one of the prehistoric stela. This finding supports an archaeological hypothesis regarding the presumed engraving tools. (C) 2015 Elsevier Masson SAS. All rights reserved.
Graph neural networks have emerged as a promising approach for the analysis of non-Euclidean data such as meshes. In medical imaging, mesh-like data plays an important role for modelling anatomical structures, and sha...
详细信息
Graph neural networks have emerged as a promising approach for the analysis of non-Euclidean data such as meshes. In medical imaging, mesh-like data plays an important role for modelling anatomical structures, and shape classification can be used in computer aideddiagnosis anddisease detection. However, with a plethora of options, the best architectural choices for medical shape analysis using GNNs remain unclear. We conduct a comparative analysis to provide practitioners with an overview of the current state-of-the-art in geometric deep learning for shape classification in neuroimaging. Using biological sex classification as a proof-of-concept task, we find that using FPFH as node features substantially improves GNN performance and generalisation to out-of-distribution data;we compare the performance of three alternative convolutional layers;and we reinforce the importance of data augmentation for graph based learning. We then confirm these results hold for a clinically relevant task, using the classification of Alzheimer's disease.
This thesis relates to research anddevelopment in the field of 3d mesh data forcomputer graphics. A review of existing storage and manipulation techniques for meshdata isgiven followed by a framework for mesh editin...
详细信息
This thesis relates to research anddevelopment in the field of 3d mesh data forcomputer graphics. A review of existing storage and manipulation techniques for meshdata isgiven followed by a framework for mesh editing. The proposed framework combines complexmesh editing techniques, automatic level of detail generation andmesh compression forstorage. These methods work coherently due to the underlying data structure. The problem ofstoring and manipulating data for 3d models is a highly researched field. Models are usuallyrepresented by sparse meshdata which consists of vertex position information, theconnectivity information to generate faces from those vertices, surface normal data andtexture coordinate information. This sparse data is sent to the graphics hardware forrendering but must be manipulated on the *** proposed framework is based upon geometry images and is designed to store andmanipulate the meshdata entirely on the graphics hardware. By utilizing the highly parallelnature of current graphics hardware and new hardware features, new levels of interactivitywith large meshes can be gained. Automatic level of detail rendering can be used to allowmodels upwards of 2 million polygons to be manipulated in real time while viewing a lowerlevel of detail. Through the use of pixels shaders the high detail is preserved in the surfacenormals while geometric detail is reduced. A compression scheme is then introduced whichutilizes the regular structure of the geometry image to compress the floating point data. Anumber of existing compression schemes are compared as well as custom bit *** is a TIF funded project which is partnered with Unlimited Realities, a PalmerstonNorth software development company. The project was to design a system to create,manipulate and store 3dmeshes in a compressed and easy to manipulate manner. The goal isto create the underlying technologies to allow for a 3d modelling system to become integratedinto the Umajin engine, not
暂无评论