VR/AR-AdaptFace:面向虚拟现实与增强现实的自适应多模态面部替换模型

随着VR/AR技术的迅猛发展,用户对于沉浸式体验的需求日益增长。同时,虚拟人脸技术亦趋成熟。基于此,本文探索将高度拟真的虚拟人脸融入VR/AR,以增强用户体验的自然度与沉浸感。然而,在虚拟数字人领域,图像生成及换脸技术在VR/AR环境下仍遇诸多挑战,尤其是唇形合成模型在动态场景及多语言环境下的性能需进一步优化。为解决上述问题,本文提出VR/AR-AdaptFace模型,一个面向虚拟现实与增强现实的自适应多模态面部替换方案。该模型由两大模块构成:“文颜绘真”模块,采用先进的文本至图像转换技术和特定类别先验保存策略,优化虚拟人脸生成,并通过注意力机制大幅提升图像质量;“语唇映生”模块,依托强大的生成器、唇形同步判别器及视觉质量判别器,实现语音与唇形的精准同步,为VR/AR场景中的动态交互带来更加逼真的体验。

3D Face Reconstruction from a Single Image Using a Combined PCA-LPP Method

In this paper, we proposed a combined PCA-LPP algorithm toimprove 3D face reconstruction performance. Principal component analysis(PCA) is commonly used to compress images and extract features. Onedisadvantage of PCA is local feature loss. To address this, various studies haveproposed combining a PCA-LPP-based algorithm with a locality preservingprojection (LPP). However, the existing PCA-LPP method is unsuitable for3D face reconstruction because it focuses on data classification and clustering.In the existing PCA-LPP, the adjacency graph, which primarily shows the connectionrelationships between data, is composed of the e-or k-nearest neighbortechniques. By contrast, in this study, complex and detailed parts, such aswrinkles around the eyes and mouth, can be reconstructed by composing thetopology of the 3D face model as an adjacency graph and extracting localfeatures from the connection relationship between the 3D model vertices.Experiments verified the effectiveness of the proposed method. When theproposed method was applied to the 3D face reconstruction evaluation set,a performance improvement of 10% to 20% was observed compared with theexisting PCA-based method.

Time Series Facebook Prophet Model and Python for COVID-19 Outbreak Prediction

COVID-19 comes from a large family of viruses identied in 1965;to date,seven groups have been recorded which have been found to affect humans.In the healthcare industry,there is much evidence that Al or machine learning algorithms can provide effective models that solve problems in order to predict conrmed cases,recovered cases,and deaths.Many researchers and scientists in the eld of machine learning are also involved in solving this dilemma,seeking to understand the patterns and characteristics of virus attacks,so scientists may make the right decisions and take specic actions.Furthermore,many models have been considered to predict the Coronavirus outbreak,such as the retro prediction model,pandemic Kaplan’s model,and the neural forecasting model.Other research has used the time series-dependent face book prophet model for COVID-19 prediction in India’s various countries.Thus,we proposed a prediction and analysis model to predict COVID-19 in Saudi Arabia.The time series dependent face book prophet model is used to t the data and provide future predictions.This study aimed to determine the pandemic prediction of COVID-19 in Saudi Arabia,using the Time Series Analysis to observe and predict the coronavirus pandemic’s spread daily or weekly.We found that the proposed model has a low ability to forecast the recovered cases of the COVID-19 dataset.In contrast,the proposed model of death cases has a high ability to forecast the COVID-19 dataset.Finally,obtaining more data could empower the model for further validation.

Emfacenet:一种轻量级人脸识别的卷积神经网络

随着计算机技术日益发展,计算机视觉逐渐融入人们的生活,深度卷积神经网络在计算机视觉领域得到了广泛的应用.然而计算资源和内存的限制,为卷积神经网络在嵌入式设备的部署带来了巨大的困难.本文提出了一种新的轻量级的人脸识别的卷积神经网络——Emfacenet,通过在CASIA-WebFace数据集上进行卷积神经网络的训练,并在计算机CPU平台以及嵌入式平台上利用LFW数据集对模型的预测效果分别进行测试,Emfacenet在CPU平台下识别速度分别是Resnet50、Mobilenetv3以及Mobilefacenets这3种模型的2.07倍、1.67倍、1.63倍,在嵌入式平台下识别速度分别56.65倍、2.09倍、3.41倍.而且Emfacenet卷积神经网络模型大小仅为138.1KB,保持较高精度的同时运行效率显著提高,可以适用于嵌入式等硬件资源受限领域来实现人脸识别.

Gaussian Mixture Models for Human Face Recognition under Illumination Variations

The appearance of a face is severely altered by illumination conditions that makes automatic face recognition a challenging task. In this paper we propose a Gaussian Mixture Models (GMM)-based human face identification technique built in the Fourier or frequency domain that is robust to illumination changes and does not require “illumination normalization” (removal of illumination effects) prior to application unlike many existing methods. The importance of the Fourier domain phase in human face identification is a well-established fact in signal processing. A maximum a posteriori (or, MAP) estimate based on the posterior likelihood is used to perform identification, achieving misclassification error rates as low as 2% on a database that contains images of 65 individuals under 21 different illumination conditions. Furthermore, a misclassification rate of 3.5% is observed on the Yale database with 10 people and 64 different illumination conditions. Both these sets of results are significantly better than those obtained from traditional PCA and LDA classifiers. Statistical analysis pertaining to model selection is also presented.

Realistic face modeling based on multiple deformations

On the basis of the assumption that the human face belongs to a linear class, a multiple-deformation model is proposed to recover face shape from a few points on a single 2D image. Compared to the conventional methods, this study has the following advantages. First, the proposed modified 3D sparse deforming model is a noniterative approach that can compute global translation efficiently and accurately. Subsequently, the overfitting problem can be alleviated based on the proposed multiple deformation model. Finally, by keeping the main features, the texture generated is realistic. The comparison results show that this novel method outperforms the existing methods by using ground truth data and that realistic 3D faces can be recovered efficiently from a single photograph.

A Combined Discrete Event—Agent Based Approach to Modeling Tensile Strength of One-Dimensional Fibrous Materials. Face Validation and Effect of the Basic Model Parameters

A combined method of discrete event and agent based modelling has been applied to the computer modelling and simulation of the tensile strength of one-dimensional fibrous materials (ODFM). This combined method is based on the concept of discrete event simulation as being applied to the modeling of the structure of the fiber flow and on the concept of agent based modelling for modelling and simulation of the fiber interaction within the structure of the fibrous material. Frictional and traction forces arise as the result of this fiber interaction. A model of the ODFM tensile strength, which is based on the slippage effect, is created and studied in this research. Only frictional and traction forces determine the tensile strength in this kind of the model. The article examines the validation problem of the slippage effect based tensile strength model and questions regarding the strength potential estimation through variation in the parameters of the model.

A simplified physically-based breach model for a high concrete-faced rockfill dam:A case study

A simplified physically-based model was developed to simulate the breaching process of the Gouhou concrete-faced rockfill dam （CFRD）, which is the only breach case of a high CFRD in the world. Considering the dam height, a hydraulic method was chosen to simulate the initial scour position on the downstream slope, with the steepening of the downstream slope taken into account; a headcut erosion formula was adopted to simulate the backward erosion as well. The moment equilibrium method was utilized to calculate the ultimate length of a concrete slab under its self-weight and water loads. The calculated results of the Gouhou CFRD breach case show that the proposed model provides reasonable peak breach flow, final breach width, and failure time, with relative errors less than 15% as compared with the measured data. Sensitivity studies show that the outputs of the proposed model are more or less sensitive to different parameters. Three typical parametric models were compared with the proposed model, and the comparison demonstrates that the proposed physically-based breach model performs better and provides more detailed results than the parametric models.

CYCLIC DEFORMATION OF FACE CENTERED CUBIC CRYSTALS AND ITS DISLOCATION INTERACTION MODEL——Ⅱ.DISLOCATION INTERACTION MODEL OF CYCLIC DEFORMATION

A dislocation interaction model has been proposed for cyclic deformation of fcc crystals.Ac- cording to this model,cyclic stress-strain responses and saturation dislocation structures of a crystal are associated with the modes and intensities of dislocation interactions between slip systems active in the crystal; and,hence,may be predicted by the location of its tensile axis in the crystallographic triangle.This model has successfully explained the different behaviours of double-slip crystals and multi-slip behaviours of some crystals with orientations usually con- sidered as single-slip ones.

Seepage simulation of high concrete-faced rockfill dams based on generalized equivalent continuum model

This research focused on the three-dimensional(3 D) seepage field simulation of a high concrete-faced rockfill dam(CFRD) under complex hydraulic conditions. A generalized equivalent continuum model of fractured rock mass was used for equivalent continuous seepage field analysis based on the improved node virtual flow method. Using a high CFRD as an example, the generalized equivalent continuum range was determined, and a finite element model was established based on the terrain and geological conditions, as well as structural face characteristics of the dam area. The equivalent seepage coefficients of different material zones or positions in the dam foundation were calculated with the Snow model or inverse analysis. Then, the 3 D seepage field in the dam area was calculated under the normal water storage conditions, and the corresponding water head distribution, seepage flow, seepage gradient, and seepage characteristics in the dam area were analyzed. The results show that the generalized equivalent continuum model can effectively simulate overall seepage patterns of the CFRD under complex hydraulic conditions and provide a reference for seepage analysis of similar CFRDs.

基于边塌陷减面的实景三维模型轻量化技术

实景三维模型由于其丰富的场景展现能力,广泛应用在数字孪生和地理测绘等领域。然而,丰富的三维场景和细节所包含的数据量十分庞大,给实景三维模型的存储、传输、渲染与展示带来了极大的挑战。针对上述问题,本文提出了一种面向实景三维模型的轻量化处理方法,首先引入模型预处理技术,去除模型噪音和非流行边;然后通过边塌陷减面的轻量化算法,精简模型拓扑结构;最后加入反转检测算法,防止边塌陷过程中出现面反转的情况。对张家界景区4个真实场景的测试结果表明,该技术可以在保留模型整体外观和场景细节的基础上,压缩至原始模型大小的3.7%~10.9%,使得复杂场景的实景三维模型能够在计算机中更加快速真实地反映或显示世界的地理地貌或城市建筑。

基于ThetaMEX全局池化的人脸识别神经网络——ShuffaceNet

针对目前大规模网络不适合在手机、平板电脑等资源匮乏的移动设备上使用,以及池化层会导致特征图的稀疏性最终影响神经网络识别精度的问题,提出了一个轻量级人脸识别神经网络ShuffaceNet,设计了一个非线性平滑Log-Mean-Exp函数ThetaMEX,并提出了一种端到端可训练的ThetaMEX全局池化层(TGPL),从而在保证算法精度的前提下,减少网络参数、提高运算速度,进而达到有效地将该网络部署在资源匮乏的移动设备上的目的。ShuffaceNet约有3 600个参数,模型大小仅为3.5 MB。在LFW(Labled Faces in the Wild)、AgeDB-30 (Age Database-30)、CFP (Celebrities in Frontal Profile)人脸数据集上的识别测试的结果表明,ShuffaceNet的精度分别达到了99.32%、93.17%、94.51%。与MobileNetV1、SqueezeNet、Xception相比,所提网络的大小分别缩减了73.1%、82.1%、78.5%,在AgeDB-30数据集上的精度分别提高了5.0%、6.3%、6.7%。可见,基于ThetaMEX全局池化的所提网络能够提高模型精度。

A New Modified Conductivity Model for Prediction of Shear Yield Stress of Electrorheological Fluids Based on Face-center Square Structure

A new modified conductivity model was established to predict the shear yield stress of electrorheological fluids (ERF). By using a cell equivalent method, the present model can deal with the face-center square structure of ERF. Combining the scheme of the classical conductivity model for the single-chain structure, a new formula for the prediction of the shear yield stress of ERF was set up. The influences of the separation distance of the particles, the volume fraction of the particles and the applied electric field on the shear yield stress were investigated.

基于特征融合的轻量级新残差人脸识别方法

针对现有轻量级模型在嵌入式设备的人脸识别应用中存在识别精度难以提升的问题,提出一种融合人脸对齐关键特征点信息的轻量级新残差网络模型(Lightweight New Residual Network,LNRN).LNRN利用深度残差网络结构能够解决网络退化且避免干扰因素影响的优势,结合人脸对齐环节产生的关键特征点信息,对深度残差网络结构进行简化和合理设计,实现对关键特征信息和全局信息的提取.为避免特征提取过程中丢失重要特征信息,该模型在新残差网络中加入结合空间和通道的注意力机制进行辅助.在公开的四个标准人脸数据集上的仿真实验表明,该模型识别速度在接近主流轻量级人脸识别方法的同时,平均识别精度比MobiFace提高了0.6%.

堆石料本构模型对混凝土面板坝应力变形计算结果的影响研究

堆石料本构模型选择是影响混凝土面板坝应力变形计算结果的主要因素。以某典型面板坝为例,采用邓肯E-B非线性弹性模型和“南水”双屈服面弹塑性模型,对坝体填筑过程和蓄水过程进行了三维有限元模拟;研究了两种模型计算的坝体和面板的位移和应力分布差异;分析了两种模型计算结果呈现差异的原因。两种模型计算结果最显著的差异体现在两个方面:(1)“南水”模型计算的坝体沉降、面板挠度等位移指标小于E-B模型,其原因是“南水”模型计算的坝体小主应力和变形模量更高。(2)E-B模型计算结果显示蓄水后面板底部顺坡向受拉;而“南水”模型计算的面板顺坡向应力为全断面受压。前者是由E-B模型的各向同性弹性本质决定的,垫层料因水压力作用顺坡向膨胀,产生沿坡面向上的位移;后者是其F1屈服面持续扩张产生塑性体积收缩导致的,塑性体缩抵消了顺坡向弹性膨胀,使垫层料产生沿坡面向下的位移。

混凝土面板与堆石料接触面模型及其应用

混凝土面板是面板坝防渗安全的生命线,其应力变形特性主要取决于坝体堆石料的应力变形特性及面板与堆石料之间的接触特性。在面板坝的应力变形计算中,通常采用接触面模型模拟面板与堆石料之间的荷载传递机制,因此接触面模型是否合理,直接影响到面板应力变形计算结果的可靠性。分析了传统双曲线接触面模型存在的不足,提出了一个考虑双向剪切耦合效应,且符合强度准则和标架无差异原则的接触面本构模型,该模型参数少,数值实现方便,可以定性地模拟出实际面板坝工程中出现的混凝土挤压破坏区域。

基于三维渗流场下的某尾矿库安全性分析

尾矿库一般是通过筑坝拦截谷口或围地形成,作为堆存废石、废矿及其他工业废弃物的主要场地。通过一系列调查研究发现,坝体渗流的浸润面位置高低对尾矿库的安全稳定性有着很大影响。以河南省某尾矿库工程为例,建立了能够反映坝体结构和岩体特性的三维有限元模型,分析了标高1 280 m尾矿库坝体的渗流特性。结果表明,在堆积坝体内铺设排渗席垫,可大量排出坝体内渗流水,有效地控制浸润面高度,使其埋深满足规范要求,不同水位下渗透坡降均小于允许数值,不发生渗透破坏。

域迁移增强的综合假脸检测模型

提出一种跨域的综合假脸检测模型。首先,设计一种双域融合模型,该模型利用空间注意力机制实现RGB域和频域特征的融合。其次,在此基础上,结合数据增强技术,提出了一种跨域迁移策略。最后,提出的双域模型的精度,在5个通用数据集上均比单域模型有一定的提高,尤其在NT数据集上,该方法的精度比EfficientNet-B0方法提高了3.4%。此外,实验结果表明,与其他迁移学习方法相比,在FaceForensics++和Celeb-df数据集上,该方法在域迁移中具有更好的泛化性能。

FACE水稻茎蘖动态模型

借助农田开放式空气CO2浓度增高(FACE)技术平台,以武香粳14为供试水稻品种,设置不同施N量处理,研究大气CO2浓度为570μmol·mol-1(比对照高200μmol·mol-1)的FACE处理对水稻茎蘖动态的影响,并建立了相应的模拟模型:Tt=1+Ae1a1-b1t-1+Aea22-b2t+C×1+Ba31-b3t-1+Be2a4-b4t+D.模型以时间为驱动因子,描述了水稻茎蘖数随移栽天数的动态变化过程,对常规及CO2浓度增加条件下水稻茎蘖的变化均有很好的拟合性.通过不同年份试验数据对模型的检验,预测根均方差(RMSE)最大为44·27个·m-2,最小为13·96个·m-2,且相关系数均达到了极显著水平.表明模型的预测程度较高,具有很好的适用性.