arXiv subjects
Publications and source records attributed to Lu Yang.
We consider recurrent solutions of the nonautonomous modified Swift-Hohenberg equation $$u_t+Δ^2u+2Δu+au+b|\nabla u|^2+u^3=g(t,x).$$ We employ Conley index theory to show that, if the forcing $g:\mathbb{R}\rightarrow L^2(Ω)$ is a recurrent function, then there are at least two recurrent solutions in $H_0^2(Ω)$ under appropriate assumptions on the parameters $a$, $b$ and $g$.
Scene text detection has received attention for years and achieved an impressive performance across various benchmarks. In this work, we propose an efficient and accurate approach to detect multioriented text in scene images. The proposed feature fusion mechanism allows us to use a shallower network to reduce the computational complexity. A self-attention mechanism is adopted to suppress false positive detections. Experiments on public benchmarks including ICDAR 2013, ICDAR 2015 and MSRA-TD500 show that our proposed approach can achieve better or comparable performances with fewer parameters and less computational cost.
In this work, we construct a large-scale dataset for vehicle re-identification (ReID), which contains 137k images of 13k vehicle instances captured by UAV-mounted cameras. To our knowledge, it is the largest UAV-based vehicle ReID dataset. To increase intra-class variation, each vehicle is captured by at least two UAVs at different locations, with diverse view-angles and flight-altitudes. We manually label a variety of vehicle attributes, including vehicle type, color, skylight, bumper, spare tire and luggage rack. Furthermore, for each vehicle image, the annotator is also required to mark the discriminative parts that helps them to distinguish this particular vehicle from others. Besides the dataset, we also design a specific vehicle ReID algorithm to make full use of the rich annotation information. It is capable of explicitly detecting discriminative parts for each specific vehicle and significantly outperforms the evaluated baselines and state-of-the-art vehicle ReID approaches.
Vision-based person, hand or face detection approaches have achieved incredible success in recent years with the development of deep convolutional neural network (CNN). In this paper, we take the inherent correlation between the body and body parts into account and propose a new framework to boost up the detection performance of the multi-level objects. In particular, we adopt a region-based object detection structure with two carefully designed detectors to separately pay attention to the human body and body parts in a coarse-to-fine manner, which we call Detector-in-Detector network (DID-Net). The first detector is designed to detect human body, hand, and face. The second detector, based on the body detection results of the first detector, mainly focus on the detection of small hand and face inside each body. The framework is trained in an end-to-end way by optimizing a multi-task loss. Due to the lack of human body, face and hand detection dataset, we have collected and labeled a new large dataset named Human-Parts with 14,962 images and 106,879 annotations. Experiments show that our method can achieve excellent performance on Human-Parts.
With the broad use of face recognition, its weakness gradually emerges that it is able to be attacked. So, it is important to study how face recognition networks are subject to attacks. In this paper, we focus on a novel way to do attacks against face recognition network that misleads the network to identify someone as the target person not misclassify inconspicuously. Simultaneously, for this purpose, we introduce a specific attentional adversarial attack generative network to generate fake face images. For capturing the semantic information of the target person, this work adds a conditional variational autoencoder and attention modules to learn the instance-level correspondences between faces. Unlike traditional two-player GAN, this work introduces face recognition networks as the third player to participate in the competition between generator and discriminator which allows the attacker to impersonate the target person better. The generated faces which are hard to arouse the notice of onlookers can evade recognition by state-of-the-art networks and most of them are recognized as the target person.
Instance-level human analysis is common in real-life scenarios and has multiple manifestations, such as human part segmentation, dense pose estimation, human-object interactions, etc. Models need to distinguish different human instances in the image panel and learn rich features to represent the details of each instance. In this paper, we present an end-to-end pipeline for solving the instance-level human analysis, named Parsing R-CNN. It processes a set of human instances simultaneously through comprehensive considering the characteristics of region-based approach and the appearance of a human, thus allowing representing the details of instances. Parsing R-CNN is very flexible and efficient, which is applicable to many issues in human instance analysis. Our approach outperforms all state-of-the-art methods on CIHP (Crowd Instance-level Human Parsing), MHP v2.0 (Multi-Human Parsing) and DensePose-COCO datasets. Based on the proposed Parsing R-CNN, we reach the 1st place in the COCO 2018 Challenge DensePose Estimation task. Code and models are public available.
Millimeter wave (mmWave) communication by utilizing lens antenna arrays is a promising technique for realizing cost-effective 5G wireless systems with large MIMO (multiple-input multiple-output) but only limited radio frequency (RF) chains. This paper studies an uplink multi-user mmWave single-sided lens MIMO system, where only the base station (BS) is equipped with a full-dimensional (FD) lens antenna array with both elevation and azimuth angle resolution capabilities, and each mobile station (MS) employs the conventional uniform planar array (UPA) without the lens. By exploiting the angle-dependent energy focusing property of the lens antenna array at the BS as well as the multi-path sparsity of mmWave channels, we propose a low-complexity path division multiple access (PDMA) scheme, which enables virtually interference-free multi-user communications when the angle of arrivals (AoAs) of all MS multi-path signals are sufficiently separable at the BS. To this end, a new technique called path delay compensation is proposed at the BS to effectively transform the multi-user frequency-selective MIMO channels to parallel frequency-flat small-size MIMO channels for different MSs, for each of which the low-complexity single-carrier(SC) transmission is applied. For general scenarios with insufficient AoA separations, analog beamforming at the MSs and digital combining at the BS are jointly designed to maximize the achievable sum-rate of the MSs based on their effective MIMO channels resulting from path delay compensation. In addition, we propose a new and efficient channel estimation scheme tailored for PDMA, which requires negligible training overhead in practical mmWave systems and yet leads to comparable performance as that based on perfect channel state information (CSI).
Given a multi-variant polynomial inequality with a parameter, how to find the best possible value of this parameter that satisfies the inequality? For instance, find the greatest number $k$ that satisfies $ a^3+b^3+c^3+ k(a^2b+b^2c+c^2a)-(k+1)(ab^2+bc^2+ca^2)\geq 0 $ for all nonnegative real numbers $ a,b,c $. Analogues problems often appeared in studies of inequalities and were dealt with by various methods. In this paper, a general algorithm is proposed for finding the required best possible constant. The algorithm can be easily implemented by computer algebra tools such as Maple.
Although the latest high-end smartphone has powerful CPU and GPU, running deeper convolutional neural networks (CNNs) for complex tasks such as ImageNet classification on mobile devices is challenging. To deploy deep CNNs on mobile devices, we present a simple and effective scheme to compress the entire CNN, which we call one-shot whole network compression. The proposed scheme consists of three steps: (1) rank selection with variational Bayesian matrix factorization, (2) Tucker decomposition on kernel tensor, and (3) fine-tuning to recover accumulated loss of accuracy, and each step can be easily implemented using publicly available tools. We demonstrate the effectiveness of the proposed scheme by testing the performance of various compressed CNNs (AlexNet, VGGS, GoogLeNet, and VGG-16) on the smartphone. Significant reductions in model size, runtime, and energy consumption are obtained, at the cost of small loss in accuracy. In addition, we address the important implementation level issue on 1?1 convolution, which is a key operation of inception module of GoogLeNet as well as CNNs compressed by our proposed scheme.
High index differential algebraic equations (DAEs) are ordinary differential equations (ODEs) with constraints and arise frequently from many mathematical models of physical phenomenons and engineering fields. In this paper, we generalize the idea of differential elimination with Dixon resultant to polynomially nonlinear DAEs. We propose a new algorithm for index reduction of DAEs and establish the notion of differential algebraic elimination, which can provide the differential algebraic resultant of the enlarged system of original equations. To make use of structure of DAEs, variable pencil technique is given to determine the termination of differentiation. Moreover, we also provide a heuristics method for removing the extraneous factors from differential algebraic resultant. The experimentation shows that the proposed algorithm outperforms existing ones for many examples taken from the literature.
This note shows the equivalence of two projection operators which both can be used in cylindrical algebraic decomposition (CAD) . One is known as Brown's Projection (C. W. Brown (2001)); the other was proposed by Lu Yang in his earlier work (L.Yang and S.~H. Xia (2000)) that is sketched as follows: given a polynomial $f$ in $x_1,\,x_2,\,\cdots$, by $f_1$ denote the resultant of $f$ and its partial derivative with respect to $x_1$ (removing the multiple factors), by $f_2$ denote the resultant of $f_1$ and its partial derivative with respect to $x_2$, (removing the multiple factors), $\cdots$, repeat this procedure successively until the last resultant becomes a univariate polynomial. Making use of an identity, the equivalence of these two projection operators is evident.
Synaptotagmin (syt) 7 is one of three syt isoforms found in all metazoans; it is ubiquitously expressed, yet its function in neurons remains obscure. Here, we resolved Ca2+-dependent and Ca2+-independent synaptic vesicle (SV) replenishment pathways, and found that syt 7 plays a selective and critical role in the Ca2+-dependent pathway. Mutations that disrupt Ca2+-binding to syt 7 abolish this function, suggesting that syt 7 functions as a Ca2+-sensor for replenishment. The Ca2+-binding protein calmodulin (CaM) has also been implicated in SV replenishment, and we found that loss of syt 7 was phenocopied by a CaM antagonist. Moreover, we discovered that syt 7 binds to CaM in a highly specific and Ca2+-dependent manner; this interaction requires intact Ca2+-binding sites within syt 7. Together, these data indicate that a complex of two conserved Ca2+-binding proteins, syt 7 and CaM, serve as a key regulator of SV replenishment in presynaptic nerve terminals.
There have been some effective tools for solving (constant/parametric) semi-algebraic systems in Maple's library RegularChains since Maple 13. By using the functions of the library, e.g., RealRootClassfication, one can prove and discover polynomial inequalities. This paper is more or less a user guide on using RealRootClassfication to prove the nonnegativity of polynomials. We show by examples how to use this powerful tool to prove a polynomial is nonnegative under some polynomial inequality and/or equation constraints. Some tricks for using the tool are also provided.
In this paper, the achievable DoF of MIMO X channels for constant channel coefficients with $M_t$ antennas at transmitter $t$ and $N_r$ antennas at receiver $r$ ($t,r=1,2$) is studied. A spatial interference alignment and cancelation scheme is proposed to achieve the maximum DoF of the MIMO X channels. The scenario of $M_1\geq M_2\geq N_1\geq N_2$ is first considered and divided into 3 cases, $3N_2 N_1$, otherwise it is 1/2 or 1 less than the outer-bound; in Case $C$, the achievable DoF is equal to the outer-bound $2/3(M_1+M_2)$ if $(3N_2-M_1-M_2)\mod 3=0$, and it is 1/3 or 1/6 less than the outer-bound if $(3N_2-M_1-M_2)\mod 3=1 \mathrm{or} 2$. In the scenario of $M_t\leq N_r$, the exact symmetrical results of DoF can be obtained.
Cylindrical algebraic decomposition is one of the most important tools for computing with semi-algebraic sets, while triangular decomposition is among the most important approaches for manipulating constructible sets. In this paper, for an arbitrary finite set $F \subset {\R}[y_1, ..., y_n]$ we apply comprehensive triangular decomposition in order to obtain an $F$-invariant cylindrical decomposition of the $n$-dimensional complex space, from which we extract an $F$-invariant cylindrical algebraic decomposition of the $n$-dimensional real space. We report on an implementation of this new approach for constructing cylindrical algebraic decompositions.
Partitioning and transport of water and small solutes into and through nanopores is important to a variety of chemical and biological processes and applications. Here we study the partitioning of positive ions of increasing size into the pore interior using extensive molecular dynamics simulations. Despite the simplicity of the simulation system, the results provide new and useful insights on both the pore hydration and ion partitioning. For low negative charge densities on the pore, the driving force for partitioning of positive ions into the pore is weak, and no partitioning is observed. Increasing the pore charge gradually leads to partitioning of positive ions into the pore. Interestingly, over a range of intermediate negative charge densities, nanopores display both thermodynamic as well as kinetic selectivity toward partitioning of the larger K+ and Cs+ ions into their interior over the smaller Na+ ions. The variation of thermodynamic driving force and the average partitioning time with the pore charge density together suggest the presence of free energy barriers in the partitioning process. We discuss the role of ion hydration in the bulk and in the pore interior as well as of the pore hydration in determining the barrier heights for ion partitioning and the observed thermodynamic and kinetic selectivity.
The well-known "Generalized Champagne Problem" on simultaneous stabilization of linear systems is solved by using complex analysis \cite{A73,C78,G69,N52,R87} and Blondel's technique \cite{B94,BG93,BGMR94}. We give a complete answer to the open problem proposed by Patel et al. \cite{P99,PDV02}, which automatically includes the solution to the original "Champagne Problem" \cite{BG93,BGMR94,BSVW99,LKZ99,P99,PDV02}. Based on the recent development in automated inequality-type theorem proving \cite{Y98,Y99,YHX01,YX05,YX03}, a new stabilizing controller design method is established. Our numerical examples significantly improve the relevant results in the literature \cite{LKZ99,PDV02}.