arXiv subjects

Zhi Yu

Publications and source records attributed to Zhi Yu.

42 records · Page 3Linked to original sources

arXiv2019

Lightweight Real-time Makeup Try-on in Mobile Browsers with Tiny CNN Models for Facial Tracking

Recent works on convolutional neural networks (CNNs) for facial alignment have demonstrated unprecedented accuracy on a variety of large, publicly available datasets. However, the developed models are often both cumbersome and computationally expensive, and are not adapted to applications on resource restricted devices. In this work, we look into developing and training compact facial alignment models that feature fast inference speed and small deployment size, making them suitable for applications on the aforementioned category of devices. Our main contribution lies in designing such small models while maintaining high accuracy of facial alignment. The models we propose make use of light CNN architectures adapted to the facial alignment problem for accurate two-stage prediction of facial landmark coordinates from low-resolution output heatmaps. We further combine the developed facial tracker with a rendering method, and build a real-time makeup try-on demo that runs client-side in smartphone Web browsers. More results and demo are in our project page: http://research.modiface.com/makeup-try-on-cvprw2019/

cs.CV↗

arXiv2017

Geometric field theory and weak Euler-Lagrange equation for classical relativistic particle-field systems

A manifestly covariant, or geometric, field theory for relativistic classical particle-field system is developed. The connection between space-time symmetry and energy-momentum conservation laws for the system is established geometrically without splitting the space and time coordinates, i.e., space-time is treated as one identity without choosing a coordinate system. To achieve this goal, we need to overcome two difficulties. The first difficulty arises from the fact that particles and field reside on different manifold. As a result, the geometric Lagrangian density of the system is a function of the 4-potential of electromagnetic fields and also a functional of particles' world-lines. The other difficulty associated with the geometric setting is due to the mass-shell condition. The standard Euler-Lagrange (EL) equation for a particle is generalized into the geometric EL equation when the mass-shell condition is imposed. For the particle-field system, the geometric EL equation is further generalized into a weak geometric EL equation for particles. With the EL equation for field and the geometric weak EL equation for particles, symmetries and conservation laws can be established geometrically. A geometric expression for the energy-momentum tensor for particles is derived for the first time, which recovers the non-geometric form in the existing literature for a chosen coordinate system.

physics.plasm-ph↗

arXiv2017

Joint Multiuser Downlink Beamforming and Admission Control for Green Cloud-RANs with Limited Fronthaul Based on Mixed Integer Semi-definite Program

With the dense deployment of the remote radio heads (RRHs), the huge network power consumption has become a great challenge for green cloud radio access networks (Cloud-RANs), and multiuser downlink beamforming has been proposed as a promising solution. Moreover, the increasing number of mobile users (MUs) causes that admission control is essential for Cloud-RAN with limited fronthaul capacity and predefined power budget. In this paper, we consider the problem of joint multiuser downlink beamforming and admission control (JBAC) to enhance the admitted MUs in the network and reduce the network power consumption, while taking into account the Quality of Service requirements of the MUs, the power budget constraints and fronthaul limitation. It is shown that the JBAC problem is a mixed integer nonlinear problem, and still non-convex even though the continuous relaxation is adopted. Therefore, we first transform the JBAC problem into a Mixed-Integer Semidefinite Program. Then, we propose a bound improving Branch and Bound algorithm to yield the near-optimal solution. For practical application, a polynomial-time heuristic algorithm is proposed to derive the sub-optimal solution. Extensive simulations are conducted with different system configurations to show the effectiveness of the proposed two schemes.

cs.IT↗

arXiv2016

Joint User Association and Downlink Beamforming for Green Cloud-RANs with Limited Fronthaul

With the explosive growth of smart devices and mobile data traffic, limited fronthaul capacity has become a notable bottleneck of green communication access networks, such as cloud radio access networks(C-RANs). In this paper, we proposed a joint user association and downlink beamforming scheme for green C-RANs to minimize the network power consumption with the limited fronthaul links. We first formulate the design problem as a mixed integer nonlinear programming (MINLP), and then transformed the MINLP problem into a mixed integer second-order cone program (MI-SOCP) which is a convex program when the integer variables are fixed. By relaxing the integer variables to continuous ones, an inflation algorithm, which can be finished within polynomial time, was proposed to solved the problem. The simulation results are presented to validate the effectiveness of our proposed algorithm compared with the the scheme adopted by LTE-A.

cs.IT↗

arXiv2016

Explicit high-order noncanonical symplectic algorithms for ideal two-fluid systems

An explicit high-order noncanonical symplectic algorithm for ideal two-fluid systems is developed. The fluid is discretized as particles in the Lagrangian description, while the electromagnetic fields and internal energy are treated as discrete differential form fields on a fixed mesh. With the assistance of Whitney interpolating forms, this scheme preserves the gauge symmetry of the electromagnetic field, and the pressure field is naturally derived from the discrete internal energy. The whole system is solved using the Hamiltonian splitting method discovered by He et al., which was been successfully adopted in constructing symplectic particle-in-cell schemes. Because of its structure preserving and explicit nature, this algorithm is especially suitable for large-scale simulations for physics problems that are multi-scale and require long-term fidelity and accuracy. The algorithm is verified via two tests: studies of the dispersion relation of waves in a two-fluid plasma system and the oscillating two-stream instability.

physics.plasm-ph↗

arXiv2015

Variational Symplectic Particle-in-cell Simulation of Nonlinear Mode Conversion from Extraordinary waves to Bernstein Waves

In this paper, the nonlinear mode conversion of extraordinary waves in nonuniform magnetized plasmas is studied using the variational symplectic particle-in-cell simulation. The accuracy of the nonlinear simulation is guaranteed by the long-term accuracy and conservativeness of the symplectic algorithm. The spectra of the electromagnetic wave, the evolution of the wave reflectivity, the energy deposition profile, and the parameter-dependent properties of radio-frequency waves during the nonlinear mode conversion are investigated. It is illustrated that nonlinear effects significantly modify the physics of the radio-frequency injection in magnetized plasmas. The evolutions of the radio-frequency wave reflectivity and the energy deposition are observed, as well as the self-interaction of the Bernstein waves and mode excitations. Even for waves with small magnitude, nonlinear effects can also become important after continuous wave injections, which are common in the realistic radio-frequency wave heating and current drive experiments.

physics.plasm-ph↗