arXiv subjects

Feng Lu

Publications and source records attributed to Feng Lu.

At least 109 records · Page 6Linked to original sources

arXiv2019

Real-Time Semantic Segmentation via Multiply Spatial Fusion Network

Real-time semantic segmentation plays a significant role in industry applications, such as autonomous driving, robotics and so on. It is a challenging task as both efficiency and performance need to be considered simultaneously. To address such a complex task, this paper proposes an efficient CNN called Multiply Spatial Fusion Network (MSFNet) to achieve fast and accurate perception. The proposed MSFNet uses Class Boundary Supervision to process the relevant boundary information based on our proposed Multi-features Fusion Module which can obtain spatial information and enlarge receptive field. Therefore, the final upsampling of the feature maps of 1/8 original image size can achieve impressive results while maintaining a high speed. Experiments on Cityscapes and Camvid datasets show an obvious advantage of the proposed approach compared with the existing approaches. Specifically, it achieves 77.1% Mean IOU on the Cityscapes test dataset with the speed of 41 FPS for a 1024*2048 input, and 75.4% Mean IOU with the speed of 91 FPS on the Camvid test dataset.

cs.CV↗

arXiv2019

Separate In Latent Space: Unsupervised Single Image Layer Separation

Many real world vision tasks, such as reflection removal from a transparent surface and intrinsic image decomposition, can be modeled as single image layer separation. However, this problem is highly ill-posed, requiring accurately aligned and hard to collect triplet data to train the CNN models. To address this problem, this paper proposes an unsupervised method that requires no ground truth data triplet in training. At the core of the method are two assumptions about data distributions in the latent spaces of different layers, based on which a novel unsupervised layer separation pipeline can be derived. Then the method can be constructed based on the GANs framework with self-supervision and cycle consistency constraints, etc. Experimental results demonstrate its successfulness in outperforming existing unsupervised methods in both synthetic and real world tasks. The method also shows its ability to solve a more challenging multi-layer separation task.

cs.CV↗

arXiv2019

What I See Is What You See: Joint Attention Learning for First and Third Person Video Co-analysis

In recent years, more and more videos are captured from the first-person viewpoint by wearable cameras. Such first-person video provides additional information besides the traditional third-person video, and thus has a wide range of applications. However, techniques for analyzing the first-person video can be fundamentally different from those for the third-person video, and it is even more difficult to explore the shared information from both viewpoints. In this paper, we propose a novel method for first- and third-person video co-analysis. At the core of our method is the notion of "joint attention", indicating the learnable representation that corresponds to the shared attention regions in different viewpoints and thus links the two viewpoints. To this end, we develop a multi-branch deep network with a triplet loss to extract the joint attention from the first- and third-person videos via self-supervised learning. We evaluate our method on the public dataset with cross-viewpoint video matching tasks. Our method outperforms the state-of-the-art both qualitatively and quantitatively. We also demonstrate how the learned joint attention can benefit various applications through a set of additional experiments.

cs.CV↗

arXiv2018

Pathological Evidence Exploration in Deep Retinal Image Diagnosis

Though deep learning has shown successful performance in classifying the label and severity stage of certain disease, most of them give few evidence on how to make prediction. Here, we propose to exploit the interpretability of deep learning application in medical diagnosis. Inspired by Koch's Postulates, a well-known strategy in medical research to identify the property of pathogen, we define a pathological descriptor that can be extracted from the activated neurons of a diabetic retinopathy detector. To visualize the symptom and feature encoded in this descriptor, we propose a GAN based method to synthesize pathological retinal image given the descriptor and a binary vessel segmentation. Besides, with this descriptor, we can arbitrarily manipulate the position and quantity of lesions. As verified by a panel of 5 licensed ophthalmologists, our synthesized images carry the symptoms that are directly related to diabetic retinopathy diagnosis. The panel survey also shows that our generated images is both qualitatively and quantitatively superior to existing methods.

cs.CV↗

arXiv2018

Design and characterization of high energy micro-CT with a laser-based X-ray source

The increasingly demand for machining accuracy and product quality excites a great interest in high-resolution non-destructive testing (NDT) methods, but spatial resolution of conventional high-energy computed tomography (CT) is limited to sub-millimeter because of large X-ray spot size. Therefore, we propose a novel high-resolution high-energy CT based on laser-driven X-ray source and prove its feasibility to allow high-spatial-resolution tomographic imaging of dense objects. A numerical model is developed with a consideration of realistic factors including parameter fluctuations, statistical noise and detecting efficiency. By using modulation transfer functions, the system performance is quantitatively characterized and optimized in terms of source characteristics, detector aperture, geometrical configuration and projection parameters. As a result, the simulated tomography for a high-density object (up to 19.35g/cm3) achieves a basic spatial resolution of 64.9μm. This concept expands the prospects of laser-based compact X-ray sources and shows a great potential to achieve high-perspectivity micro-CT imaging for various industrial applications.

physics.ins-det↗

arXiv2018

Detail Preserving Depth Estimation from a Single Image Using Attention Guided Networks

Convolutional Neural Networks have demonstrated superior performance on single image depth estimation in recent years. These works usually use stacked spatial pooling or strided convolution to get high-level information which are common practices in classification task. However, depth estimation is a dense prediction problem and low-resolution feature maps usually generate blurred depth map which is undesirable in application. In order to produce high quality depth map, say clean and accurate, we propose a network consists of a Dense Feature Extractor (DFE) and a Depth Map Generator (DMG). The DFE combines ResNet and dilated convolutions. It extracts multi-scale information from input image while keeping the feature maps dense. As for DMG, we use attention mechanism to fuse multi-scale features produced in DFE. Our Network is trained end-to-end and does not need any post-processing. Hence, it runs fast and can predict depth map in about 15 fps. Experiment results show that our method is competitive with the state-of-the-art in quantitative evaluation, but can preserve better structural details of the scene depth.

cs.CV↗

arXiv2018

VoxSegNet: Volumetric CNNs for Semantic Part Segmentation of 3D Shapes

Voxel is an important format to represent geometric data, which has been widely used for 3D deep learning in shape analysis due to its generalization ability and regular data format. However, fine-grained tasks like part segmentation require detailed structural information, which increases voxel resolution and thus causes other issues such as the exhaustion of computational resources. In this paper, we propose a novel volumetric convolutional neural network, which could extract discriminative features encoding detailed information from voxelized 3D data under a limited resolution. To this purpose, a spatial dense extraction (SDE) module is designed to preserve the spatial resolution during the feature extraction procedure, alleviating the loss of detail caused by sub-sampling operations such as max-pooling. An attention feature aggregation (AFA) module is also introduced to adaptively select informative features from different abstraction scales, leading to segmentation with both semantic consistency and high accuracy of details. Experiment results on the large-scale dataset demonstrate the effectiveness of our method in 3D shape part segmentation.

cs.CV↗

arXiv2017

Orbital-driven two-dome superconducting phases in iron-based superconductors

Recent several experiments revealed that novel bipartite magnetic/superconducting phases widely exist in iron pnictides and chalcogenides. Nevertheless, the origin of the two-dome superconducting phases in iron-based compounds still remains unclear. Here we theoretically investigated the electronic structures, magnetic and superconducting properties of three representative iron-based systems, i.e. LaFeAsO1-xHx, LaFeAs1-xPxO and KFe2As2. We found that in addition to the degenerate in-plane anisotropic xz/yz orbitals, the quasi-degenerate in-plane isotropic orbitals drive these systems entering into the second parent phase. Moreover, the second superconducting phase is contributed by the isotropic orbitals rather than the anisotropic ones in the first superconducting phase, indicating an orbital-selective pairing state. These results imply an orbital-driven mechanism and shed light on the understanding of the two-dome magnetic/superconducting phases in iron-based compounds.

cond-mat.supr-con↗

arXiv2016

A Sodium laser guide star coupling efficiency measurement method

Large telescope's adaptive optics (AO) system requires one or several bright artificial laser guide stars to improve its sky coverage. The recent advent of high power sodium laser is perfect for such application. However, besides the output power, other parameters of the laser also have significant impact on the brightness of the generated sodium laser guide star mostly in non-linear relationships. When tuning and optimizing these parameters it is necessary to tune based on a laser guide star generation performance metric. Although return photon flux is widely used, variability of atmosphere and sodium layer make it difficult to compare from site to site even within short time period for the same site. A new metric, coupling efficiency is adopted in our field tests. In this paper, we will introduce our method for measuring the coupling efficiency of a 20W class pulse sodium laser for AO application during field tests that were conducted during 2013-2015.

astro-ph.IM↗

arXiv2016

Teaching Robots to Do Object Assembly using Multi-modal 3D Vision

The motivation of this paper is to develop a smart system using multi-modal vision for next-generation mechanical assembly. It includes two phases where in the first phase human beings teach the assembly structure to a robot and in the second phase the robot finds objects and grasps and assembles them using AI planning. The crucial part of the system is the precision of 3D visual detection and the paper presents multi-modal approaches to meet the requirements: AR markers are used in the teaching phase since human beings can actively control the process. Point cloud matching and geometric constraints are used in the robot execution phase to avoid unexpected noises. Experiments are performed to examine the precision and correctness of the approaches. The study is practical: The developed approaches are integrated with graph model-based motion planning, implemented on an industrial robots and applicable to real-world scenarios.

cs.RO↗

arXiv2015

Experimental Demonstration of Phase Modulation and Motion Sensing Using Graphene-Integrated Metasurfaces

Plasmonic metasurfaces are able to modify the wavefront by altering the light intensity, phase and polarization state. Active plasmonic metasurfaces would allow dynamic modulation of the wavefront which give rise to interesting application such as beam-steering, holograms and tunable waveplates. Graphene is an interesting material with dynamic property which can be controlled by electrical gating at an ultra-fast speed. We use a graphene-integrated metasurface to induce a tunable phase change to the wavefront. The metasurface supports a Fano resonance which produces high-quality resonances around 7.7 microns. The phase change is measured using a Michleson interferometry setup. It is shown that the reflection phase can change up to 55 degrees. In particular the phase can change by 28 degrees while the amplitude is nearly constant. The anisotropic optical response of the metasurface is used to modulate the ellipticity of the reflected wave in response to an incident field at 45 degree. We show a proof of concept application of our system in potentially ultra-fast laser interferometry with sub-micron accuracy.

physics.optics↗

arXiv2014

A New Way to Factorize Linear Cameras

The implementation details of factorizing the 3x4 projection matrices of linear cameras into their left matrix factors and the 4x4 homogeneous central(also parallel for infinite center cases) projection factors are presented in this work. Any full row rank 3x4 real matrix can be factorized into such basic matrices which will be called LC factors. A further extension to multiple view midpoint triangulation, for both pinhole and affine camera cases, is also presented based on such camera factorizations.

cs.CV↗

arXiv2013

Topological Structure of Urban Street Networks from the Perspective of Degree Correlations

Many complex networks demonstrate a phenomenon of striking degree correlations, i.e., a node tends to link to other nodes with similar (or dissimilar) degrees. From the perspective of degree correlations, this paper attempts to characterize topological structures of urban street networks. We adopted six urban street networks (three European and three North American), and converted them into network topologies in which nodes and edges respectively represent individual streets and street intersections, and compared the network topologies to three reference network topologies (biological, technological, and social). The urban street network topologies (with the exception of Manhattan) showed a consistent pattern that distinctly differs from the three reference networks. The topologies of urban street networks lack striking degree correlations in general. Through reshuffling the network topologies towards for example maximum or minimum degree correlations while retaining the initial degree distributions, we found that all the surrogate topologies of the urban street networks, as well as the reference ones, tended to deviate from small world properties. This implies that the initial degree correlations do not have any positive or negative effect on the networks' performance or functions. Keywords: Scale free, small world, rewiring, rich club effect, reshuffle, and complex networks

physics.soc-ph↗

arXiv2013

Correlation effects in the electronic structure of the Ni-based superconducting KNi2S2

The LDA plus Gutzwiller variational method is used to investigate the groundstate physical properties of the newly discovered superconducting KNi2S2. Five Ni-3d Wannier-orbital basis are constructed by the density-functional theory, to combine with local Coulomb interaction to describe the normal state electronic structure of Ni-based superconductor. The band structure and the mass enhanced are studied based on a multiorbital Hubbard model by using Gutzwiller approximation method. Our results indicate that the correlation effects lead to the mass enhancement of KNi2S2. Different from the band structure calculated from the LDA results, there are three energy bands across the Fermi level along the X-Z line due to the existence of the correlation effects, which induces a very complicated Fermi surface along the X-Z line. We have also investigated the variation of the quasi-particle weight factor with the hole or electron doping and found that the mass enhancement character has been maintained with the doping.

cond-mat.supr-con↗

arXiv2013

Correlated Topological Insulators with Mixed Valence

We propose the local density approximation (LDA)+Gutzwiller method incorporating Green's function scheme to study the topological physics of correlated materials from the first-principles. Applying this method to typical mixed valence materials SmB6, we found its non-trivial Z2 topology, indicating that SmB6 is a strongly correlated topological insulator (TI). The unique feature of this compound is that its surface states contain three Dirac cones in contrast to most known TIs.

cond-mat.str-el↗

arXiv2013

Optical properties and Mechanical properties of C, Si, Ge and 3C-SiC Materials Calculated from First Principles Theory

First-principles calculations based on Density functional theory (DFT) and Heyd, Scuseria and Ernzerhof (HSE) adopt PBE approximation-a new version of the generalized gradient approximation (GGA). It has studied lattice parameter, electronic structure, optical properties and mechanical properties of materials, including diamond-C and zinc-blende structure of Si, Ge, 3C-SiC. The result of HSE calculation is obviously superior to DFT calculation and accord well with the existing experimental values. It indicates that they are indirect bandgap materials from the figure of band structure and density of states. The bandgap which it calculated using HSE accords well with the existing experimental values except Ge. In the calculation of optical property, it shows that they correspond with the existing experimental values from the imaginary part of analytical dielectric function to the refractive index and adsorption coefficient. It demonstrates that they exists the corresponding relationship among the peak of the imaginary part of analytical dielectric function, the refractive index and adsorption coefficient. The optical property has a direct relationship with the distribution of crystal bandgap and electronic state density. From the results of mechanical properties it can know that they are brittle materials. Though the hardness and stiffness of 3C-SiC is lower than diamond, it is superior to the materials of Si, Ge as excellent semiconductor materials. In addition, the mechanical anisotropy of four materials is inconspicuous; the anisotropy of Diamond-C is very inconspicuous from anisotropic properties of Young modulus.

cond-mat.mtrl-sci↗

arXiv2012

Transversal Propagation Behaviors of Helical Edge States In a QSH System

The transversal propagation of the edge states in a two-dimensional quantum spin Hall system are classified by decay characteristic quantity $λ$. Two different modes of the helical edge states exhibit distinct behaviors. The penetration depth is momentum dependent in normal edge states. The finite size gap decays monotonously with sample width, leading to the normal size effect. In contrast, the penetration depth maintains a uniform minimal value in the special edge states with much shorter length. The finite size gap decays non-monotonously with width, leading to the anomalous finite size effect. Real materials are compared in the phase diagram, which explicitly demonstrates their differences. We also propose an intuitive way to search the special edge states in two-dimensional quantum spin Hall system.

cond-mat.mes-hall↗

arXiv2011

Connectivity of edge and surface states in topological insulators

The edge states of a two-dimensional quantum spin Hall (QSH) insulator form a one-dimensional helical metal which is responsible for the transport property of the QSH insulator. Conceptually, such a one-dimensional helical metal can be attached to any scattering region as the usual metallic leads. We study the analytical property of the scattering matrix for such a conceptual multiterminal scattering problem in the presence of time reversal invariance. As a result, several theorems on the connectivity property of helical edge states in two-dimensional QSH systems as well as surface states of three-dimensional topological insulators are obtained. Without addressing real model details, these theorems, which are phenomenologically obtained, emphasize the general connectivity property of topological edge/surface states from the mere time reversal symmetry restriction.

cond-mat.mes-hall↗