arXiv subjects

Yan Lu

Publications and source records attributed to Yan Lu.

At least 271 records · Page 15Linked to original sources

arXiv2018

Real-Time Anomaly Detection With HMOF Feature

Anomaly detection is a challenging problem in intelligent video surveillance. Most existing methods are computation consuming, which cannot satisfy the real-time requirement. In this paper, we propose a real-time anomaly detection framework with low computational complexity and high efficiency. A new feature, named Histogram of Magnitude Optical Flow (HMOF), is proposed to capture the motion of video patches. Compared with existing feature descriptors, HMOF is more sensitive to motion magnitude and more efficient to distinguish anomaly information. The HMOF features are computed for foreground patches, and are reconstructed by the auto-encoder for better clustering. Then, we use Gaussian Mixture Model (GMM) Classifiers to distinguish anomalies from normal activities in videos. Experimental results show that our framework outperforms state-of-the-art methods, and can reliably detect anomalies in real-time.

cs.CV↗

arXiv2018

In Defense of the Classification Loss for Person Re-Identification

The recent research for person re-identification has been focused on two trends. One is learning the part-based local features to form more informative feature descriptors. The other is designing effective metric learning loss functions such as the triplet loss family. We argue that learning global features with classification loss could achieve the same goal, even with some simple and cost-effective architecture design. In this paper, we first explain why the person re-id framework with standard classification loss usually has inferior performance compared to metric learning. Based on that, we further propose a person re-id framework featured by channel grouping and multi-branch strategy, which divides global features into multiple channel groups and learns the discriminative channel group features by multi-branch classification layers. The extensive experiments show that our framework outperforms prior state-of-the-arts in terms of both accuracy and inference speed.

cs.CV↗

arXiv2018

Affinity Derivation and Graph Merge for Instance Segmentation

We present an instance segmentation scheme based on pixel affinity information, which is the relationship of two pixels belonging to a same instance. In our scheme, we use two neural networks with similar structure. One is to predict pixel level semantic score and the other is designed to derive pixel affinities. Regarding pixels as the vertexes and affinities as edges, we then propose a simple yet effective graph merge algorithm to cluster pixels into instances. Experimental results show that our scheme can generate fine-grained instance mask. With Cityscapes training data, the proposed scheme achieves 27.3 AP on test set.

cs.CV↗

arXiv2018

MVPNet: Multi-View Point Regression Networks for 3D Object Reconstruction from A Single Image

In this paper, we address the problem of reconstructing an object's surface from a single image using generative networks. First, we represent a 3D surface with an aggregation of dense point clouds from multiple views. Each point cloud is embedded in a regular 2D grid aligned on an image plane of a viewpoint, making the point cloud convolution-favored and ordered so as to fit into deep network architectures. The point clouds can be easily triangulated by exploiting connectivities of the 2D grids to form mesh-based surfaces. Second, we propose an encoder-decoder network that generates such kind of multiple view-dependent point clouds from a single image by regressing their 3D coordinates and visibilities. We also introduce a novel geometric loss that is able to interpret discrepancy over 3D surfaces as opposed to 2D projective planes, resorting to the surface discretization on the constructed meshes. We demonstrate that the multi-view point regression network outperforms state-of-the-art methods with a significant improvement on challenging datasets.

cs.CV↗

arXiv2018

Silver nanowires with optimized silica coating as versatile plasmonic resonators

Metal nanoparticles are the most frequently used nanostructures in plasmonics. However, besides nanoparticles, metal nanowires feature several advantages for applications. Their elongation offers a larger interaction volume, their resonances can reach higher quality factors, and their mode structure provides better coupling into integrated hybrid dielectric-plasmonic circuits. It is crucial though, to control the distance of the wire to a supporting substrate, to another metal layer or to active materials with sub-nanometer precision. A dielectric coating can be utilized for distance control, but it must not degrade the plasmonic properties. In this paper, we introduce a controlled synthesis and coating approach for silver nanowires to fulfill these demands. We synthesize and characterize silver nanowires of around 70 nm in diameter. These nanowires are coated with nm-sized silica shells using a modified Stöber method to achieve a homogeneous and smooth surface quality. We use transmission electron microscopy, dark-field microscopy and electron-energy loss spectroscopy to study morphology and plasmonic resonances of individual nanowires and quantify the influence of the silica coating. Thorough numerical simulations support the experimental findings showing that the coating does not deteriorate the plasmonic properties and thus introduce silver nanowires as usable building blocks for integrated hybrid plasmonic systems.

physics.optics↗

arXiv2018

Interaction of Proteins with Polyelectrolytes: A Comparison between Theory and Experiment

We discuss recent investigations of the interaction of polyelectrolytes with proteins. In particular, we review our recent studies on the interaction of simple proteins such as human serum albumin (HSA) or lysozyme with linear polyelectrolytes, charged dendrimers, charged networks, and polyelectrolyte brushes. In all cases discussed here we combined experimental work with molecular dynamics (MD) simulations and mean-field theories. In particular, isothermal titration calorimetry (ITC) has been employed to obtain the respective binding constants Kb and the Gibbs free energy of binding. MD-simulations with explicit counterions but implicit water demonstrate that counterion release is the main driving force for the binding of proteins to strongly charged polyelectrolytes: Patches of positive charges located on the surface of the protein become a multivalent counterion of the polyelectrolyte thereby releasing a number of counterions condensed on the polyelectrolyte. The binding Gibbs free energy due to counterion release is predicted to scale with the logarithm of the salt concentration in the system which is verified both by simulations and experiment. In several cases, namely for the interaction of proteins with linear polyelectrolytes and highly charged hydrophilic dendrimers the binding constant could be calculated from simulations in very good approximation. This finding demonstrated that in these cases explicit hydration effects do not contribute to the Gibbs free energy of binding. The Gibbs free energy can also be used to predict the kinetics of protein uptake by microgels for a given system by applying dynamic density functional theory.

cond-mat.soft↗

arXiv2018

General Synthetic Route Towards Highly Dispersed Metal Clusters Enabled by Poly(ionic liquid)s

The capability to synthesize a broad spectrum of metal clusters (MCs) with their size controllable in a subnanometer scale presents an enticing prospect for exploring nanosize-dependent properties. Here we report an innovative design of a capping agent from a polytriazolium poly(ionic liquid) (PIL) in a vesicular form in solution that allows for crafting a variety of MCs including transition metals, noble metals, and their bimetallic alloy with precisely controlled sizes (~ 1 nm) and record-high catalytic performance. The ultrastrong stabilization power is result of an unusual synergy between the conventional binding sites in the heterocyclic cations in PIL, and an in-situ generated polycarbene structure induced simultaneously to the reduction reaction. INTRODUCTION

cond-mat.soft↗

arXiv2018

Deep Convolutional Neural Networks for Eigenvalue Problems in Mechanics

We show that deep convolutional neural networks (CNN) can massively outperform traditional densely-connected neural networks (both deep or shallow) in predicting eigenvalue problems in mechanics. In this sense, we strike out in a new direction in mechanics computations with strongly predictive NNs whose success depends not only on architectures being deep, but also being fundamentally different from the widely-used to date. We consider a model problem: predicting the eigenvalues of 1-D and 2-D phononic crystals. For the 1-D case, the optimal CNN architecture reaches $98\%$ accuracy level on unseen data when trained with just 20,000 samples, compared to $85\%$ accuracy even with $100,000$ samples for the typical network of choice in mechanics research. We show that, with relatively high data-efficiency, CNNs have the capability to generalize well and automatically learn deep symmetry operations, easily extending to higher dimensions and our 2D case. Most importantly, we show how CNNs can naturally represent mechanical material tensors, with its convolution kernels serving as local receptive fields, which is a natural representation of mechanical response. Strategies proposed are applicable to other mechanics' problems and may, in the future, be used to sidestep cumbersome algorithms with purely data-driven approaches based upon modern deep architectures.

physics.comp-ph↗

arXiv2018

Quantitative analysis on the disparity of regional economic development in China and its evolution from 1952 to 2000

Domestic and foreign scholars have already done much research on regional disparity and its evolution in China, but there is a big difference in conclusions. What is the reason for this? We think it is mainly due to different analytic approaches, perspectives, spatial units, statistical indicators and different periods for studies. On the basis of previous analyses and findings, we have done some further quantitative computation and empirical study, and revealed the inter-provincial disparity and regional disparity of economic development and their evolution trends from 1952-2000. The results shows that (a) Regional disparity in economic development in China, including the inter-provincial disparity, inter-regional disparity and intra-regional disparity, has existed for years; (b) Gini coefficient and Theil coefficient have revealed a similar dynamic trend for comparative disparity in economic development between provinces in China. From 1952 to 1978, except for the "Great Leap Forward" period, comparative disparity basically assumes a upward trend and it assumed a slowly downward trend from 1979 to1990. Afterwards from 1991 to 2000 the disparity assumed a slowly upward trend again; (c) A comparison between Shanghai and Guizhou shows that absolute inter-provincial disparity has been quite big for years; and (d) The Hurst exponent (H=0.5) in the period of 1966-1978 indicates that the comparative inter-provincial disparity of economic development showed a random characteristic, and in the Hurst exponent (H>0.5) in period of 1979-2000 indicates that in this period the evolution of the comparative inter-provincial disparity of economic development in China has a long-enduring characteristic.

stat.AP↗

arXiv2018

Local Descriptors Optimized for Average Precision

Extraction of local feature descriptors is a vital stage in the solution pipelines for numerous computer vision tasks. Learning-based approaches improve performance in certain tasks, but still cannot replace handcrafted features in general. In this paper, we improve the learning of local feature descriptors by optimizing the performance of descriptor matching, which is a common stage that follows descriptor extraction in local feature based pipelines, and can be formulated as nearest neighbor retrieval. Specifically, we directly optimize a ranking-based retrieval performance metric, Average Precision, using deep neural networks. This general-purpose solution can also be viewed as a listwise learning to rank approach, which is advantageous compared to recent local ranking approaches. On standard benchmarks, descriptors learned with our formulation achieve state-of-the-art results in patch verification, patch retrieval, and image matching.

cs.CV↗

arXiv2018

Catalysis by metallic nanoparticles in solution: Thermosensitive microgels as nanoreactors

Metallic nanoparticles have been used as catalysts for various reactions, and the huge literature on the subject is hard to overlook. In many applications, the nanoparticles must be affixed to a colloidal carrier for easy handling during catalysis. These "passive carriers" (e.g., dendrimers) serve for a controlled synthesis of the nanoparticles and prevent coagulation during catalysis. Recently, hybrids from nanoparticles and polymers have been developed that allow us to change the catalytic activity of the nanoparticles by external triggers. In particular, single nanoparticles embedded in a thermosensitive network made from poly(N-isopropylacrylamide) (PNIPAM) have become the most-studied examples of such hybrids: Immersed in cold water, the PNIPAM network is hydrophilic and fully swollen. In this state, hydrophilic substrates can diffuse easily through the network, and react at the surface of the nanoparticles. Above the volume transition located at 32°C, the network becomes hydrophobic and shrinks. Now hydrophobic substrates will preferably diffuse through the network and react with other substrates in the reaction catalyzed by the enclosed nanoparticle. Such "active carriers", may thus be viewed as true nanoreactors that open new ways for the use of nanoparticles in catalysis.

physics.chem-ph↗

arXiv2017

Feature Selective Networks for Object Detection

Objects for detection usually have distinct characteristics in different sub-regions and different aspect ratios. However, in prevalent two-stage object detection methods, Region-of-Interest (RoI) features are extracted by RoI pooling with little emphasis on these translation-variant feature components. We present feature selective networks to reform the feature representations of RoIs by exploiting their disparities among sub-regions and aspect ratios. Our network produces the sub-region attention bank and aspect ratio attention bank for the whole image. The RoI-based sub-region attention map and aspect ratio attention map are selectively pooled from the banks, and then used to refine the original RoI features for RoI classification. Equipped with a light-weight detection subnetwork, our network gets a consistent boost in detection performance based on general ConvNet backbones (ResNet-101, GoogLeNet and VGG-16). Without bells and whistles, our detectors equipped with ResNet-101 achieve more than 3% mAP improvement compared to counterparts on PASCAL VOC 2007, PASCAL VOC 2012 and MS COCO datasets.

cs.CV↗

arXiv2017

Level Repulsion and Band Sorting in Phononic Crystals

In this paper we consider the problem of avoided crossings (level repulsion) in phononic crystals and suggest a computationally efficient strategy to distinguish them from normal cross points. This process is essential for the correct sorting of the phononic bands and, subsequently, for the accurate determination of mode continuation, group velocities, and emergent properties which depend on them such as thermal conductivity. Through explicit phononic calculations using generalized Rayleigh quotient, we identify exact locations of exceptional points in the complex wavenumber domain which results in level repulsion in the real domain. We show that in the vicinity of the exceptional point the relevant phononic eigenvalue surfaces resemble the surfaces of a 2 by 2 parameter-dependent matrix. Along a closed loop encircling the exceptional point we show that the phononic eigenvalues are exchanged, just as they are for the 2 by 2 matrix case. However, the behavior of the associated eigenvectors is shown to be more complex in the phononic case. Along a closed loop around an exceptional point, we show that the eigenvectors can flip signs multiple times unlike a 2 by 2 matrix where the flip of sign occurs only once. Finally, we exploit these eigenvector sign flips around exceptional points to propose a simple and efficient method of distinguishing them from normal crosses and of correctly sorting the band-structure. Our proposed method is roughly an order-of magnitude faster than the zoom-in method and correctly identifies > 97% of the cases considered. Both its speed and accuracy can be further improved and we suggest some ways of achieving this. Our method is general and, as such, would be directly applicable to other eigenvalue problems where the eigenspectrum needs to be correctly sorted.

cond-mat.mtrl-sci↗

arXiv2017

Combining plane wave expansion and variational techniques for fast phononic computations

In this paper the salient features of the Plane Wave Expansion (PWE) method and the mixed variational technique are combined for the fast eigenvalue computations of arbitrarily complex phononic unit cells. This is done by expanding the material properties in a Fourier expansion, as is the case with PWE. The required matrix elements in the variational scheme are identified as the discrete Fourier transform coefficients of material properties, thus obviating the need for any explicit integration. The process allows us to provide succinct and closed form expressions for all the matrices involved in the mixed variational method. The scheme proposed here preserves both the simplicity of expression which is inherent in the PWE method and the superior convergence properties of the mixed variational scheme. We present numerical results and comment upon the convergence and stability of the current method. We show that the current representation renders the results of the method stable over the entire range of the expansion terms as allowed by the spatial discretization. When compared with a zero order numerical integration scheme, the present method results in greater computational accuracy of all eigenvalues. A higher order numerical integration scheme comes close to the accuracy of the present method but only with significantly more computational expense.

cond-mat.mtrl-sci↗

arXiv2017

3-D phononic crystals with ultra-wide band gaps

In this paper gradient based topology optimization (TO) is used to discover 3-D phononic structures that exhibit ultra-wide normalized all-angle all-mode band gaps. The challenging computational task of repeated 3-D phononic band-structure evaluations is accomplished by a combination of a fast mixed variational eigenvalue solver and distributed Graphic Processing Unit (GPU) parallel computations. The TO algorithm utilizes the material distribution-based approach and a gradient-based optimizer. The design sensitivity for the mixed variational eigenvalue problem is derived using the adjoint method and is implemented through highly efficient vectorization techniques. We present optimized results for two-material simple cubic (SC), body centered cubic (BCC), and face centered cubic (FCC) crystal structures and show that in each of these cases different initial designs converge to single inclusion network topologies within their corresponding primitive cells. The optimized results show that large phononic stop bands for bulk wave propagation can be achieved at lower than close packed spherical configurations leading to lighter unit cells. For tungsten carbide - epoxy crystals we identify all angle all mode normalized stop bands exceeding 100%, which is larger than what is possible with only spherical inclusions.

cond-mat.mtrl-sci↗

arXiv2016

Thermosensitive Cu2O-PNIPAM core-shell nanoreactors with tunable photocatalytic activity

We report a facile and novel method for the fabrication of Cu2O@PNIPAM core-shell nanoreactors using Cu2O nanocubes as the core. The PNIPAM shell not only effectively protects the Cu2O nanocubes from oxidation, but also improves the colloidal stability of the system. The Cu2O@PNIPAM core-shell microgels can work efficiently as photocatalyst for the decomposition of methyl orange under visible light. A significant enhancement in the catalytic activity has been observed for the core-shell microgels compared with the pure Cu2O nanocubes. Most importantly, the photocatalytic activity of the Cu2O nanocubes can be further tuned by the thermosensitive PNIPAM shell, as rationalized by our recent theory.

physics.chem-ph↗

arXiv2016

Spin exchange collision mixing of the K and Rb ac Stark shifts

In a hybrid pumping alkali vapor cell that both K and Rb are filled, K atom spins are optically pumped by laser and Rb atom spins are polarized by the K spins through spin exchange. We find that the AC Stark shift of the Rb atoms is composed of not only the AC Stark shift of the Rb atoms caused by the far off resonant pumping laser which is tuned to the K absorption lines, but also the AC Stark shift of the K atom spins. The mixing of the light shifts through fast spin exchange between K and Rb atoms are studied in this paper and we demonstrate a K-Rb-21Ne co-magnetometer in which the AC Stark shift of the Rb atoms are reduced by the collision mixing.

physics.atom-ph↗

arXiv2016

Variational Methods For Phononic Calculations

Three fundamental variational principles used for solving elastodynamic eigenvalue problems are studied within the context of elastic wave propagation in periodic composites (phononics). We study the convergence of the eigenvalue problems resulting from the displacement Rayleigh quotient, the stress Rayleigh quotient and the mixed quotient. The convergence rates of the three quotients are found to be related to the continuity and differentiability of the density and compliance variation over the unit cell. In general, the mixed quotient converges faster than both the displacement Rayleigh and the stress Rayleigh quotients, however, there exist special cases where either the displacement Rayleigh or the stress Rayleigh quotient shows the exact same convergence as the mixed-method. We show that all methods converge faster for smoother material property variations, but when density variation is rough, the difference between the mixed quotient and stress Rayleigh quotient is higher and similarly, when compliance variation is rough, the difference between the mixed quotient and displacement Rayleigh quotient is higher. Since eigenvalue problems such as those considered in this paper tend to be highly computationally intensive, it is expected that these results will lead to fast and efficient algorithms in the areas of phononics and photonics.

cond-mat.mtrl-sci↗