Controllable face privacy

Facial identity recognition is one of the challenging problems in the domain of computer vision. Facial identity comprises the facial attributes of a person’s face ranging from age progression, gender, hairstyle, etc. Manipulating facial attributes such as changing the gender, hairstyle, expressions, and makeup changes the entire facial identity of a person which is often used by law offenders to commit crimes. Leveraging the deep learning-based approaches, this work proposes a one-step solution for facial attribute manipulation and detection leading to facial identity recognition in few-shot and traditional scenarios. As a first step towards performing facial identity recognition, we created the Facial Attribute Manipulation Detection (FAM) Dataset which consists of twenty unique identities with thirty-eight facial attributes generated by the StyleGAN3 inversion. The Facial Attribute Detection (FAM) Dataset has 11,560 images richly annotated in YOLO format. To perform facial attribute and identity detection, we developed the Spatial Transformer Block (STB) and Squeeze-Excite Spatial Pyramid Pooling (SE-SPP)-based Tiny YOLOv7 model and proposed as FIR-Tiny YOLOv7 (Facial Identity Recognition-Tiny YOLOv7) model. The proposed model is an improvised variant of the Tiny YOLOv7 model. For facial identity recognition, the proposed model achieved 10.0% higher mAP in the one-shot scenario, 30.4% higher mAP in the three-shot scenario, 15.3% higher mAP in the five-shot scenario, and 0.1% higher mAP in the traditional 70% − 30% split scenario as compared to the Tiny YOLOv7 model. The results obtained with the proposed model are promising for general facial identity recognition under varying facial attribute manipulation.

In this paper, we study the problem of facial attribute learning. In particular, we propose a Face Recognition guided facial Attribute classification Network, called FR-ANet. All the attributes share low-level features, while high-level features are specially learned for attribute groups. Further, to utilize the identity information, high-level features are merged to perform face identity recognition. The experimental results on CelebA and LFWA datasets demonstrate the promise of the FR-ANet.

Funding Acknowledgements Type of funding sources: None. Introduction Segmentation of cardiac magnetic resonance (CMR) images is an essential step for evaluating dimensional and functional ventricular parameters but may limited by noise artefacts, such as the increasing diffusion of implantable devices causing distortion of the magnetic field. Purpose The aim of this paper is to investigate the accuracy of a deep learning (DL) approach for automatic segmentation of cardiac structures from CMR images characterized by magnetic susceptibility artifact images. Methods A retrospective database of SAX (short axis) cine CMR images obtained from 216 patients using a 1.5 Tesla scanner was used to develop and validate the proposed DL approach, subdividing patients into two subgroups, considering the presence (n=86) or absence (n=130) of a cardiac implantable electronic device. A novel convolutional neural network (CNN) was proposed to extract the left (LV) and right (RV) ventricle endocardium and left ventricle epicardium (LV mass). The proposed network takes advantage from a spatial attention module which introduces an adaptive feature refinement through a spatial weighting strategy to identify salient image regions even in the presence of artifacts. To improve segmentation, especially for images with artifacts, a compounded loss function is introduced. Segmentation results were assessed against manual tracings and commercial CMR analysis software. Results Results of correlation and Bland-Altman analyses on images with artifacts were reported for the proposed CNN and the commercial software (CS) in respect to the manual gold standard (GT). Also interobserver variability between O1 and O2 was tested. The proposed method reached higher segmentation accuracy than CS, with performance comparable to expert inter-observer variability (Picture 1). It is possible to appreciate how all the contours were properly depicted using the proposed CNN, with results comparable to the manual tracings. On the opposite, the CS resulted in imprecise segmentations of LV and RV endocardium and LV epicardium in presence of susceptibility artefacts caused by cardiac implantable electronic device affecting images' quality (Picture 2). Furthermore, the automatic approach was at least 100 times faster than manual segmentation in providing cardiac parameters. Conclusion The proposed method reached promising performance in cardiac segmentation from CMR images with susceptibility artifacts and alleviates time consuming expert physician contourn segmentation.

The development of data communication technologies promotes large-scale sensitive data collection and transmission in various application areas. The sensitivity and criticism of the exchanged data raise several privacy issues. A lack of privacy may cause moral and emotional damage and discrimination. It can even create an unequal society. To fill this gap, a better understanding of privacy concept and its requirements is required. This chapter presents a comprehensive survey of privacy-preserving in the modern era. It deals with this concept of privacy-preserving from all perspectives, classifying its requirements into content-oriented and context-oriented ones. Based on the taxonomy, privacy attacks are described, and approaches and mechanisms for privacy protection are reviewed. A future research direction about privacy preserving in various fields is finally exposed.

Artificial intelligence (AI) used in autonomous vehicles (AV) to check for driver alertness is a critical piece of technology that makes the decision to hand over control to the human if there is a disengagement of the autonomous capability. It is important that this AI be inclusive without bias, because treating drivers differently will have an impact on the safety of humans not only in the vehicle but also on the road. This paper evaluates the AI that powers driver attention systems in the car to check if the AI treats all humans inclusively the same way beyond their ethnicity, gender and age, and whether it follows AI ethics principles of trust, transparency and fairness. Driver attention is built using two different AI models. One uses camera data to recognise humans and the other evaluates whether the human is alert. We found that both these AI models are biased and not inclusive of all people in all situations. We also found that there are unethical practices in how humans are tracked to check for alertness by using infrared sensors that track their retina movements without any concept of consent or privacy of people being tracked in the vehicles. The paper builds upon prior research on face detection outside the car and research that shows that car cognition AI does not recognise all humans on the road equally. We present research results about how the car is biased against some humans in its face identification and how the assertion of alertness of humans to hand over control during an emergency is fundamentally flawed in its definition of alertness. We recommend mitigation techniques and call for further research to build upon our work to make the AV inclusive with bias mitigation of bias in all forms of AI in AVs.

Fatigue damage assessment at the flexible riser hang-off location, where the pipe frequently endures maximum tension and curvature variations, is key to verify design integrity for service life. Traditionally, the fatigue analysis is performed in a separate local structure model, as the commercial global analysis software lacks the capability of handling the local behavior of the riser structural component, which is dependent on materials and manufacturing processes. During global fatigue analysis, the riser configuration is built with pinned connected at the hang-off point. The resulting tension and angle responses at the hang-off location, are then input to a local model to perform the stress and fatigue analysis, where the detailed pipe layer structure and bend stiffener are modeled. This traditional approach is conservative, time costly and is often limited to regular wave approach. Wellstream developed an external function to work with specialized commercial riser dynamic analysis software. The external function simulates the detailed behaviour of flexible pipe structure components and the resulting bending hysteresis during dynamic simulation in the time domain. Therefore, the stress time history of the tensile armour becomes available at the end of global simulation in the time domain and is ready for fatigue damage assessment by rain flow counting. This paper presents a study case where the fatigue assessment is performed directly at the hang-off region within the riser global dynamic simulation. The riser hang-off is situated at the top of the I-tube and a bend stiffener is fixed at the bottom of the I-tube. I-tube and pipe section are precisely modeled as pipe-in-pipe facility, where the interaction of riser/I-tube can be captured.

Nowadays, simulation is becoming more and more important in industries. Here we consider a typical industrial application in the field of sheet metal bending. A high number of simulations is necessary during the development process to perform parameter studies and optimizations. On the other hand, simulation tools should be also available for the customers of these machines, e.g., to plan the production of very specific profiles. In such cases, the optimal process parameters only can be found by simulation. Very important in this context are the license costs for commercial simulation software. Frequently, the simulations are not limited by computational power but by the number of available licenses, such that the duration for parameter studies is elongated. Also, with license costs it very expensive to provide a simulation platform to the customers. The presented case study has been carried out with the goal of comparing possible open source alternatives to expensive commercial Finite Element software. Exemplarily, we consider the elasto-plastic bending of a cantilever, using the Johnson Cook constitutive law. For this test case, a three dimensional Finite Element analysis is performed, comparing the results of open-source software (Salome-Meca) and a com mercial counterpart (Abaqus). Different element types and mesh sizes are compared, the usability of both tools, and the computational time. Considering the obvious price difference, both platforms show comparable results. Comparing the functionality of both programs, both are capable for modelling highly detailed and complex models for elasto-plastic material processing. However, for under standing the structure of the user interface of Salome-Meca is far more time consuming. Additionally, the performance of Salome-Meca on different operating systems is com pared: Salome-Meca on Linux, Salome-Meca on Linux, installed in a virtual machine on Windows, and finally Salome-Meca on Windows. All in all, it turned out that depending on the specific application Salome-Meca can be a powerful alternative to Abaqus for the considered industrial application.

State-of-the-art face super-resolution methods employ deep convolutional neural networks to learn a mapping between low- and high-resolution facial patterns by exploring local appearance knowledge. However, most of these methods do not well exploit facial structures and identity information, and struggle to deal with facial images that exhibit large pose variations. In this paper, we propose a novel face super-resolution method that explicitly incorporates 3D facial priors which grasp the sharp facial structures. Our work is the first to explore 3D morphable knowledge based on the fusion of parametric descriptions of face attributes (e.g., identity, facial expression, texture, illumination, and face pose). Furthermore, the priors can easily be incorporated into any network and are extremely efficient in improving the performance and accelerating the convergence speed. Firstly, a 3D face rendering branch is set up to obtain 3D priors of salient facial structures and identity knowledge. Secondly, the Spatial Attention Module is used to better exploit this hierarchical information (i.e., intensity similarity, 3D facial structure, and identity content) for the super-resolution problem. Extensive experiments demonstrate that the proposed 3D priors achieve superior face super-resolution results over the state-of-the-arts.

State-of-the-art face restoration methods employ deep convolutional neural networks (CNNs) to learn a mapping between degraded and sharp facial patterns by exploring local appearance knowledge. However, most of these methods do not well exploit facial structures and identity information, and only deal with task-specific face restoration (e.g., face super-resolution or deblurring). In this paper, we propose cross-tasks and cross-models plug-and-play 3D facial priors to explicitly embed the network with the sharp facial structures for general face restoration tasks. Our 3D priors are the first to explore 3D morphable knowledge based on the fusion of parametric descriptions of face attributes (e.g., identity, facial expression, texture, illumination, and face pose). Furthermore, the priors can easily be incorporated into any network and are very efficient in improving the performance and accelerating the convergence speed. Firstly, a 3D face rendering branch is set up to obtain 3D priors of salient facial structures and identity knowledge. Secondly, for better exploiting this hierarchical information (i.e., intensity similarity, 3D facial structure, and identity content), a spatial attention module is designed for the image restoration problems. Extensive face restoration experiments including face super-resolution and deblurring demonstrate that the proposed 3D priors achieve superior face restoration results over the state-of-the-art algorithms.

FE models for complex or large-scaled structures that need detailed modeling of structural components are usually constructed using commercial analysis softwares. Updating of such FE model by conventional sensitivity-based methods is difficult since repeated computation for perturbed parameters and manual calculations are needed to obtain sensitivity matrix in each iteration. In this study, an FE model updating procedure avoiding such difficulties by using response surface (RS) method and a Pareto-based multiobjective optimization (MOO) was formulated and applied to FE models constructed with a commercial analysis package. The test building is a low-rise reinforced concrete building that has been seismically retrofitted. Dynamic properties of the building were extracted from vibration tests performed before and after the seismic retrofits, respectively. The elastic modulus of concrete and masonry, and spring constants for the expansion joint were updated. Two RS functions representing the errors in the natural frequencies and mode shape, respectively, were obtained and used as the objective functions for MOO. Among the Pareto solutions, the best compromise solution was determined using the TOPSIS (Technique for Order of Preference by Similarity to Ideal Solution) procedure. A similar task was performed for retrofitted building by taking the updating parameters as the stiffness of modified or added members. Obtained parameters of the existing building were reasonably comparable with the current code provisions. However, the stiffness of added concrete shear walls and steel section jacketed members were considerably lower than expectation. Such low values are seemingly because the bond between new and existing concrete was not as good as the monolithically casted members, even though they were connected by the anchoring bars.

Details of the mould design are explained and optimized as needed to produce the system on the two‐component injection moulding machine available in this laboratory. A number of adjustments were required in order to reduce the pressures needed to fill all parts without the danger of exceeding the maximum clamping force. For these hierarchical, subsequent optimization steps, a commercial analysis package was used. After the first successful production runs, the two major functional components of the device were tested: the splitting serpentine channels were fed with two fluids, one containing a fluorescent dye, and down‐flow analysis results showed that laminar mixing indeed proceeds efficiently. The peristaltic pumps, consisting of three air‐pressure actuated membranes, were also tested, and found to have a maximum operating frequency of 5 Hz. Above this critical frequency, the membranes do not sufficiently close on their hubs and begin trembling and vibrating in the pulsating flow. Using a mild vacuum to support the membrane motion shifts the critical frequency to 15 Hz. The volume displacement per stroke is in the order of 7 µL, yielding a maximum flux of ca. 100 µL · s−1.magnified image

A critical task in predicting and tailoring the acoustic absorption properties of porous media is the calculation of the frequency-dependent effective density and compressibility tensors, which are explicitly related to the micro-scale permeability properties. Although these two quantities exhibit strong sensitivity to physics occurring at complex micro-scale geometries, most of the existing literature focuses on employing very limited in-house and oftentimes multiple numerical analysis tools. In order to predict these parameters and acoustic absorption efficiently and conveniently, this article synthesizes multiple disparate approaches into a single unified formulation suitable for incorporation into a commercial analysis package. Numerical results computed herein for four close-packed porous media are compared to similar results available in the literature. These include simple cubic, body-centered cubic, and face-centered cubic structures, and also hexagonal close-packed, which has not appeared in the literature. Together with critical comparisons of a hybrid versus direct numerical approaches, the close agreement demonstrates the capabilities of the unified formulation to analyze and control the acoustic absorption properties at the microscopic level.

One of the main concerns in using commercial software for finite element analyses of dam-foundation-reservoir systems is that the simplifying assumptions of the massless foundation are unreliable. In this study, an appropriate direct finite element method is introduced for simulating the mass, radiation damping and wave propagation effect in foundations of dam-foundation-reservoir systems using commercial software ABAQUS. The free-field boundary condition is used for modeling the semi-infinite foundation and radiation damping, which is not a built-in boundary condition in most of the available commercial software for finite element analysis of structures such as ANSYS or ABAQUS and thus needs to be implemented differently. The different mechanism for modeling of the foundation, earthquake input and far-field boundary condition is described. Implementation of the free-field boundary condition in finite element software is verified by comparing it with analytical results. To investigation the feasibility of the proposed method in dam-foundation-reservoir system analysis, a series of analyses is accomplished in a variety of cases and the obtained results are compared with the substructure method by using the EAGD-84 program. Finally, the massed and massless foundation results are compared and it is concluded that the massless foundation approach leads to the overestimation of the displacements and stresses within the dam body.

Flight simulators can recreate aircraft flights for flight training and aircraft design. This paper reviews existing engineering flight simulators and introduces the development of a multi-purpose reconfigurable engineering flight simulator at Ryerson University. The multi-purpose engineering flight simulator named the Ryerson Fixed Base Simulator (RFBS) has been designed and built to teach and initiate research projects in the area of aircraft design, flight simulation, pilot training, and flight data analysis. It consists of three 46 inch high definition screens and six 22 inch touch screen panels to represent the instrument panel, the centre console, and the overhead panels of an actual aircraft flight deck. Several low-cost, commercial flight simulation software were tested and X-Plane was selected as the main flight simulation tool. This paper also introduces a research that utilizes the RFBS and a commercial flight analysis software. The objective of the research was to develop a flight data conversion methodology. A case study was commenced to verify the work flow of flight data generation and analysis with an example of hard-landing analysis.

In this paper, we present a new Matlab based toolbox for power system analysis, called MatDyn. It is opensource software, and available for everyone to download. Its design philosophy is based on the well-known open-source Matlab toolbox MATPOWER, but its focus is transient stability analysis and time-domain simulation of power systems, instead of steadystate calculations. MatDyn's philosophy, design criteria, program structure, and implementation, are discussed in detail. A trade-off is achieved between the flexibility of the program and readability of the code. MatDyn retains overall flexibility by for instance allowing user defined models, and custom integration methods. The software is validated by comparing its results with those obtained by the commercial grade power system analysis package, PSS/E. Despite the fact that MatDyn is fairly new, it has already been extensively used in research and education. This paper reports interesting results obtained with MatDyn in recent research, that would be hard to obtain using commercial software.