arXiv subjects

Xin Lu

Publications and source records attributed to Xin Lu.

At least 199 records · Page 11Linked to original sources

arXiv2019

Grid R-CNN Plus: Faster and Better

Grid R-CNN is a well-performed objection detection framework. It transforms the traditional box offset regression problem into a grid point estimation problem. With the guidance of the grid points, it can obtain high-quality localization results. However, the speed of Grid R-CNN is not so satisfactory. In this technical report we present Grid R-CNN Plus, a better and faster version of Grid R-CNN. We have made several updates that significantly speed up the framework and simultaneously improve the accuracy. On COCO dataset, the Res50-FPN based Grid R-CNN Plus detector achieves an mAP of 40.4%, outperforming the baseline on the same model by 3.0 points with similar inference time. Code is available at https://github.com/STVIR/Grid-R-CNN .

cs.CV↗

arXiv2019

Single full gap with mixed type-I and type-II superconductivity on surface of the type-II Dirac semimetal PdTe2 by point-contact spectroscopy

We report our point-contact spectroscopy (PCS) study on the superconducting state of the type-II Dirac semimetal PdTe$_2$ with a superconducting transition temperature $T\rm_c \sim$ 1.65 K. Both mechanical- and soft- PCS differential conductance curves at 0.3 K show a consistent double-peak structure and they can be perfectly fitted by a single s-wave gap based on the Blonder-Tinkham-Klapwijk model. The gap follows a typical Bardeen-Cooper-Schrieffer temperature behavior, yielding $Δ_0 \sim$ 0.29 meV and 2$Δ_0$/$k\rm_{B}$$T\rm_{c}$ = 4.15 in the strong coupling regime. A sudden suppression of the superconducting gap in magnetic field around $H{\rm_{c1}}$ $\sim$ 130 Oe is observed for most point-contacts on PdTe$_2$, characteristic of a first-order transition for type-I superconductor in field. However, for other contacts, a smooth evolution of the PCS conductance persists up to $H{\rm_{c2}}$ $\sim$ 600 Oe, signaling a local type-II superconductivity. The observed admixture of type-I and type-II superconductivity can possibly arise from an inhomogeneous electron mean free path on the surface of PdTe$_2$ due to its topological surface states.

cond-mat.supr-con↗

arXiv2019

Evidence for nematic superconductivity of topological surface states in PbTaSe2

Spontaneous symmetry breaking has been a paradigm to describe the phase transitions in condensed matter physics. In addition to the continuous electromagnetic gauge symmetry, an unconventional superconductor can break discrete symmetries simultaneously, such as time reversal and lattice rotational symmetry. In this work we report a characteristic in-plane 2-fold behaviour of the resistive upper critical field and point-contact spectra on the superconducting semimetal PbTaSe2 with topological nodal-rings, despite its hexagonal lattice symmetry (or D_3h in bulk while C_3v on surface, to be precise). However, we do not observe any lattice rotational symmetry breaking signal from field-angle-dependent specific heat. It is worth noting that such surface-only electronic nematicity is in sharp contrast to the observation in the topological superconductor candidate, CuxBi2Se3, where the nematicity occurs in various bulk measurements. In combination with theory, superconducting nematicity is likely to emerge from the topological surface states of PbTaSe2, rather than the proximity effect. The issue of time reversal symmetry breaking is also addressed. Thus, our results on PbTaSe2 shed new light on possible routes to realize nematic superconductivity with nontrivial topology.

cond-mat.supr-con↗

arXiv2019

Foreground-aware Image Inpainting

Existing image inpainting methods typically fill holes by borrowing information from surrounding pixels. They often produce unsatisfactory results when the holes overlap with or touch foreground objects due to lack of information about the actual extent of foreground and background regions within the holes. These scenarios, however, are very important in practice, especially for applications such as the removal of distracting objects. To address the problem, we propose a foreground-aware image inpainting system that explicitly disentangles structure inference and content completion. Specifically, our model learns to predict the foreground contour first, and then inpaints the missing region using the predicted contour as guidance. We show that by such disentanglement, the contour completion model predicts reasonable contours of objects, and further substantially improves the performance of image inpainting. Experiments show that our method significantly outperforms existing methods and achieves superior inpainting results on challenging cases with complex compositions.

cs.CV↗

arXiv2019

A deep learning framework for quality assessment and restoration in video endoscopy

Endoscopy is a routine imaging technique used for both diagnosis and minimally invasive surgical treatment. Artifacts such as motion blur, bubbles, specular reflections, floating objects and pixel saturation impede the visual interpretation and the automated analysis of endoscopy videos. Given the widespread use of endoscopy in different clinical applications, we contend that the robust and reliable identification of such artifacts and the automated restoration of corrupted video frames is a fundamental medical imaging problem. Existing state-of-the-art methods only deal with the detection and restoration of selected artifacts. However, typically endoscopy videos contain numerous artifacts which motivates to establish a comprehensive solution. We propose a fully automatic framework that can: 1) detect and classify six different primary artifacts, 2) provide a quality score for each frame and 3) restore mildly corrupted frames. To detect different artifacts our framework exploits fast multi-scale, single stage convolutional neural network detector. We introduce a quality metric to assess frame quality and predict image restoration success. Generative adversarial networks with carefully chosen regularization are finally used to restore corrupted frames. Our detector yields the highest mean average precision (mAP at 5% threshold) of 49.0 and the lowest computational time of 88 ms allowing for accurate real-time processing. Our restoration models for blind deblurring, saturation correction and inpainting demonstrate significant improvements over previous methods. On a set of 10 test videos we show that our approach preserves an average of 68.7% which is 25% more frames than that retained from the raw videos.

cs.CV↗

arXiv2019

When Energy Trading meets Blockchain in Electrical Power System: The State of the Art

With the rapid growth of renewable energy resources, the energy trading began to shift from centralized to distributed manner. Blockchain, as a distributed public ledger technology, has been widely adopted to design new energy trading schemes. However, there are many challenging issues for blockchain-based energy trading, i.e., low efficiency, high transaction cost, security & privacy issues. To tackle with the above challenges, many solutions have been proposed. In this survey, the blockchain-based energy trading in electrical power system is thoroughly investigated. Firstly, the challenges in blockchain-based energy trading are identified. Then, the existing energy trading schemes are studied and classified into three categories based on their main focus: energy transaction, consensus mechanism, and system optimization. And each category is presented in detail. Although existing schemes can meet the specific energy trading requirements, there are still many unsolved problems. Finally, the discussion and future directions are given.

cs.CR↗

arXiv2018

Grid R-CNN

This paper proposes a novel object detection framework named Grid R-CNN, which adopts a grid guided localization mechanism for accurate object detection. Different from the traditional regression based methods, the Grid R-CNN captures the spatial information explicitly and enjoys the position sensitive property of fully convolutional architecture. Instead of using only two independent points, we design a multi-point supervision formulation to encode more clues in order to reduce the impact of inaccurate prediction of specific points. To take the full advantage of the correlation of points in a grid, we propose a two-stage information fusion strategy to fuse feature maps of neighbor grid points. The grid guided localization approach is easy to be extended to different state-of-the-art detection frameworks. Grid R-CNN leads to high quality object localization, and experiments demonstrate that it achieves a 4.1% AP gain at IoU=0.8 and a 10.0% AP gain at IoU=0.9 on COCO benchmark compared to Faster R-CNN with Res50 backbone and FPN architecture.

cs.CV↗

arXiv2018

Optical control of a single spin-valley in charged WSe$_2$ quantum dots

Control and manipulation of single charges and their internal degrees of freedom, such as spins, is a fundamental goal of nanoscience with promising technological applications. Recently, atomically thin semiconductors such as WSe$_2$ have emerged as a platform for valleytronics, offering rich possibilities for optical, magnetic and electrical control of the valley index. While progress has been made in controlling valley index of ensemble of charge carriers, valley control of individual charges, crucial for valleytronics, remains unexplored. Here, we provide unambiguous evidence for localized holes with net spin in optically active WSe$_2$ quantum dots (QDs) and control their spin-valley state with the helicity of the excitation laser under small magnetic field. We estimate a lower bound on the valley lifetime of a single charge in QD from recombination time to be $\sim$ nanoseconds. Remarkably, neutral QDs do not exhibit such a control, demonstrating the role of excess charge in prolonging the valley lifetime. Our work extends the field of 2D valleytronics to the level of single spin-valley, relevant for quantum information and sensing applications

cond-mat.mes-hall↗

arXiv2018

Entanglement of single-photons and chiral phonons in atomically thin WSe$_2$

Quantum entanglement is a fundamental phenomenon which, on the one hand, reveals deep connections between quantum mechanics, gravity and the space-time; on the other hand, has practical applications as a key resource in quantum information processing. While it is routinely achieved in photon-atom ensembles, entanglement involving the solid-state or macroscopic objects remains challenging albeit promising for both fundamental physics and technological applications. Here, we report entanglement between collective, chiral vibrations in two-dimensional (2D) WSe$_2$ host --- chiral phonons (CPs) --- and single-photons emitted from quantum dots (QDs) present in it. CPs which carry angular momentum were recently observed in WSe$_2$ and are a distinguishing feature of the underlying honeycomb lattice. The entanglement results from a "which-way" scattering process, involving an optical excitation in a QD and doubly-degenerate CPs, which takes place via two indistinguishable paths. Our unveiling of entanglement involving a macroscopic, collective excitation together with strong interaction between CPs and QDs in 2D materials opens up ways for phonon-driven entanglement of QDs and engineering chiral or non-reciprocal interactions at the single-photon level.

cond-mat.mes-hall↗

arXiv2018

Flow-Grounded Spatial-Temporal Video Prediction from Still Images

Existing video prediction methods mainly rely on observing multiple historical frames or focus on predicting the next one-frame. In this work, we study the problem of generating consecutive multiple future frames by observing one single still image only. We formulate the multi-frame prediction task as a multiple time step flow (multi-flow) prediction phase followed by a flow-to-frame synthesis phase. The multi-flow prediction is modeled in a variational probabilistic manner with spatial-temporal relationships learned through 3D convolutions. The flow-to-frame synthesis is modeled as a generative process in order to keep the predicted results lying closer to the manifold shape of real video sequence. Such a two-phase design prevents the model from directly looking at the high-dimensional pixel space of the frame sequence and is demonstrated to be more effective in predicting better and diverse results. Extensive experimental results on videos with different types of motion show that the proposed algorithm performs favorably against existing methods in terms of quality, diversity and human perceptual evaluation.

cs.CV↗

arXiv2018

Contemplating Visual Emotions: Understanding and Overcoming Dataset Bias

While machine learning approaches to visual emotion recognition offer great promise, current methods consider training and testing models on small scale datasets covering limited visual emotion concepts. Our analysis identifies an important but long overlooked issue of existing visual emotion benchmarks in the form of dataset biases. We design a series of tests to show and measure how such dataset biases obstruct learning a generalizable emotion recognition model. Based on our analysis, we propose a webly supervised approach by leveraging a large quantity of stock image data. Our approach uses a simple yet effective curriculum guided training strategy for learning discriminative emotion features. We discover that the models learned using our large scale stock image dataset exhibit significantly better generalization ability than the existing datasets without the manual collection of even a single label. Moreover, visual representation learned using our approach holds a lot of promise across a variety of tasks on different image and video datasets.

cs.CV↗

arXiv2018

SLM-based Digital Adaptive Coronagraphy: Current Status and Capabilities

Active coronagraphy is deemed to play a key role for the next generation of high-contrast instruments, notably in order to deal with large segmented mirrors that might exhibit time-dependent pupil merit function, caused by missing or defective segments. To this purpose, we recently introduced a new technological framework called digital adaptive coronagraphy (DAC), making use of liquid-crystal spatial light modulators (SLMs) display panels operating as active focal-plane phase mask coronagraphs. Here, we first review the latest contrast performance, measured in laboratory conditions with monochromatic visible light, and describe a few potential pathways to improve SLM coronagraphic nulling in the future. We then unveil a few unique capabilities of SLM-based DAC that were recently, or are currently in the process of being, demonstrated in our laboratory, including NCPA wavefront sensing, aperture-matched adaptive phase masks, coronagraphic nulling of multiple star systems, and coherent differential imaging (CDI).

astro-ph.IM↗

arXiv2018

MAttNet: Modular Attention Network for Referring Expression Comprehension

In this paper, we address referring expression comprehension: localizing an image region described by a natural language expression. While most recent work treats expressions as a single unit, we propose to decompose them into three modular components related to subject appearance, location, and relationship to other objects. This allows us to flexibly adapt to expressions containing different types of information in an end-to-end framework. In our model, which we call the Modular Attention Network (MAttNet), two types of attention are utilized: language-based attention that learns the module weights as well as the word/phrase attention that each module should focus on; and visual attention that allows the subject and relationship modules to focus on relevant image components. Module weights combine scores from all three modules dynamically to output an overall score. Experiments show that MAttNet outperforms previous state-of-art methods by a large margin on both bounding-box-level and pixel-level comprehension tasks. Demo and code are provided.

cs.CV↗

arXiv2018

Generative Image Inpainting with Contextual Attention

Recent deep learning based approaches have shown promising results for the challenging task of inpainting large missing regions in an image. These methods can generate visually plausible image structures and textures, but often create distorted structures or blurry textures inconsistent with surrounding areas. This is mainly due to ineffectiveness of convolutional neural networks in explicitly borrowing or copying information from distant spatial locations. On the other hand, traditional texture and patch synthesis approaches are particularly suitable when it needs to borrow textures from the surrounding regions. Motivated by these observations, we propose a new deep generative model-based approach which can not only synthesize novel image structures but also explicitly utilize surrounding image features as references during network training to make better predictions. The model is a feed-forward, fully convolutional neural network which can process images with multiple holes at arbitrary locations and with variable sizes during the test time. Experiments on multiple datasets including faces (CelebA, CelebA-HQ), textures (DTD) and natural images (ImageNet, Places2) demonstrate that our proposed approach generates higher-quality inpainting results than existing ones. Code, demo and models are available at: https://github.com/JiahuiYu/generative_inpainting.

cs.CV↗

arXiv2018

Spin-glass ground state in a triangular-lattice compound YbZnGaO$_4$

We report on comprehensive results identifying the ground state of a triangular-lattice structured YbZnGaO$_4$ to be spin glass, including no long-range magnetic order, prominent broad excitation continua, and absence of magnetic thermal conductivity. More crucially, from the ultralow-temperature a.c. susceptibility measurements, we unambiguously observe frequency-dependent peaks around 0.1 K, indicating the spin-glass ground state. We suggest this conclusion to hold also for its sister compound YbMgGaO$_4$, which is confirmed by the observation of spin freezing at low temperatures. We consider disorder and frustration to be the main driving force for the spin-glass phase.

cond-mat.str-el↗

arXiv2018

Unconventional and conventional quantum criticalities in CeRh$_{0.58}$Ir$_{0.42}$In$_5$

An appropriate description of the state of matter that appears as a second order phase transition is tuned toward zero temperature, {\it viz.} quantum-critical point (QCP), poses fundamental and still not fully answered questions. Experiments are needed both to test basic conclusions and to guide further refinement of theoretical models. Here, charge and entropy transport properties as well as AC specific heat of the heavy-fermion compound CeRh$_{0.58}$Ir$_{0.42}$In$_5$, measured as a function of pressure, reveal two qualitatively different QCPs in a {\it single} material driven by a {\it single} non-symmetry-breaking tuning parameter. A discontinuous sign-change jump in thermopower suggests an unconventional QCP at $p_{c1}$ accompanied by an abrupt Fermi-surface reconstruction that is followed by a conventional spin-density-wave critical point at $p_{c2}$ across which the Fermi surface evolves smoothly to a heavy Fermi-liquid state. These experiments are consistent with some theoretical predictions, including the sequence of critical points and the temperature dependence of the thermopower in their vicinity.

cond-mat.str-el↗

arXiv2018

Rethinking the Smaller-Norm-Less-Informative Assumption in Channel Pruning of Convolution Layers

Model pruning has become a useful technique that improves the computational efficiency of deep learning, making it possible to deploy solutions in resource-limited scenarios. A widely-used practice in relevant work assumes that a smaller-norm parameter or feature plays a less informative role at the inference time. In this paper, we propose a channel pruning technique for accelerating the computations of deep convolutional neural networks (CNNs) that does not critically rely on this assumption. Instead, it focuses on direct simplification of the channel-to-channel computation graph of a CNN without the need of performing a computationally difficult and not-always-useful task of making high-dimensional tensors of CNN structured sparse. Our approach takes two stages: first to adopt an end-to- end stochastic training method that eventually forces the outputs of some channels to be constant, and then to prune those constant channels from the original neural network by adjusting the biases of their impacting layers such that the resulting compact model can be quickly fine-tuned. Our approach is mathematically appealing from an optimization perspective and easy to reproduce. We experimented our approach through several image learning benchmarks and demonstrate its interesting aspects and competitive performance.

cs.LG↗

arXiv2018

All-solid flexible fiber-shaped lithium ion batteries

We propose fabrication of the fiber-shaped lithium ion batteries assembled by twisting a cathode filament together with an anode filament. The cathode filament is fabricated by depositing a LiFePO4 (LFP)-composite layer onto a steel-filled polyester conductive thread (SPCT). As anode filaments, we propose several scenarios including a Li4Ti5O12 (LTO)-composite coated SPCT (dip-and-dry deposition), a tin-coated SPCT (PVD deposition) as well as a bare tin wire. An electrolyte composite layer consisting of LiPF6 and polyethylene oxide (PEO) is then deposited onto both the anode and cathode filament before the battery assembly. By twisting the cathode filament and anode filament together using a customized jig, the batteries are then assembled. The open-circuit voltage is found to be ~ 2.3 V for the battery using the LTO@SPCT anode, and ~3.3 V for the battery using the tin@SPCT anode and the tin wire anode. Charge-discharge tests are carried out at different C rates for each battery sample. Experimental results suggest that the LIBs using the LTO@SPCT anode, the tin@SPCT anode and the bare tin wire anode could achieve a specific capacity of ~64, ~67, and ~96 mAh/g, respectively, when charge-discharged at 0.5-C rate. The battery could retain well its capacity after 80 charge-discharge cycles. During operation of all the batteries reported in this paper, their coulombic efficiency remained above 80%. Among the advantages of the proposed LIB are light weight, ease of fabrication, high specific capacitance, high energy density, and good durability. Finally, employing cheap and commercially-available steel-filled polyester threads as a base material in our batteries, makes them potentially suitable for integration into wearables using various standard textile manufacturing techniques.

physics.app-ph↗