arXiv subjects

Yang Long

Publications and source records attributed to Yang Long.

84 records · Page 5Linked to original sources

arXiv2018

Learning RoI Transformer for Detecting Oriented Objects in Aerial Images

Object detection in aerial images is an active yet challenging task in computer vision because of the birdview perspective, the highly complex backgrounds, and the variant appearances of objects. Especially when detecting densely packed objects in aerial images, methods relying on horizontal proposals for common object detection often introduce mismatches between the Region of Interests (RoIs) and objects. This leads to the common misalignment between the final object classification confidence and localization accuracy. Although rotated anchors have been used to tackle this problem, the design of them always multiplies the number of anchors and dramatically increases the computational complexity. In this paper, we propose a RoI Transformer to address these problems. More precisely, to improve the quality of region proposals, we first designed a Rotated RoI (RRoI) learner to transform a Horizontal Region of Interest (HRoI) into a Rotated Region of Interest (RRoI). Based on the RRoIs, we then proposed a Rotated Position Sensitive RoI Align (RPS-RoI-Align) module to extract rotation-invariant features from them for boosting subsequent classification and regression. Our RoI Transformer is with light weight and can be easily embedded into detectors for oriented object detection. A simple implementation of the RoI Transformer has achieved state-of-the-art performances on two common and challenging aerial datasets, i.e., DOTA and HRSC2016, with a neglectable reduction to detection speed. Our RoI Transformer exceeds the deformable Position Sensitive RoI pooling when oriented bounding-box annotations are available. Extensive experiments have also validated the flexibility and effectiveness of our RoI Transformer. The results demonstrate that it can be easily integrated with other detector architectures and significantly improve the performances.

cs.CV↗

arXiv2018

ActionXPose: A Novel 2D Multi-view Pose-based Algorithm for Real-time Human Action Recognition

We present ActionXPose, a novel 2D pose-based algorithm for posture-level Human Action Recognition (HAR). The proposed approach exploits 2D human poses provided by OpenPose detector from RGB videos. ActionXPose aims to process poses data to be provided to a Long Short-Term Memory Neural Network and to a 1D Convolutional Neural Network, which solve the classification problem. ActionXPose is one of the first algorithms that exploits 2D human poses for HAR. The algorithm has real-time performance and it is robust to camera movings, subject proximity changes, viewpoint changes, subject appearance changes and provide high generalization degree. In fact, extensive simulations show that ActionXPose can be successfully trained using different datasets at once. State-of-the-art performance on popular datasets for posture-related HAR problems (i3DPost, KTH) are provided and results are compared with those obtained by other methods, including the selected ActionXPose baseline. Moreover, we also proposed two novel datasets called MPOSE and ISLD recorded in our Intelligent Sensing Lab, to show ActionXPose generalization performance.

eess.IV↗

arXiv2018

Seeing Topological Order and Band Inversion in Optical Diatomic Chain

Recent realizations of exotic topological states in condensed matter and cold atoms have advanced the exploration for topological characteristics, such as invariant topological orders and band inversion. Here we construct a 1D optical diatomic system made of split ring resonators with electric-magnetic couplings to implement the topological features. We experimentally measure the dispersion relationship and sub-lattice pseudo-spin vectors by detecting field strength distributions, which determines the accompanying winding number of bulk band as the topological order. The sub-lattice pseudo-spin vector inversion observed at band-edges evidences the band inversion. We further reconfirm the band-inversion-induced topological phase transition by measuring the symmetry exchanges at band-edges. These results have shown the great potential of split ring resonator platform to experimentally explore the advanced topological characteristics, such as high winding numbers or even high Chern numbers in more exotic setups.

physics.optics↗

arXiv2018

Observation of acoustic spin

Unlike optical waves, acoustic waves in fluids are described by scalar pressure fields, and therefore are considered spinless. Here, we demonstrate experimentally the existence of spin in acoustics. In the interference of two acoustic waves propagating perpendicularly to each other, we observed the spin angular momentum in free space as a result of the rotation of local particle velocity. We successfully measured the acoustic spin, and spin induced torque acting on a lossy acoustic meta-atom that results from absorption of the spin angular momentum. The acoustic spin is also observed in the evanescent field of a guided mode traveling along a metamaterial waveguide. We found spin-momentum locking in acoustic waves whose propagation direction is determined by the sign of spin. The observed acoustic spin could open a new door in acoustics and their applications for the control of wave propagation and particle rotation.

physics.class-ph↗

arXiv2018

Simulating Quantum Spin Hall Effect in Topological Lieb Lattice of Linear Circuit

Inspired by the topological insulator circuit proposed and experimentally verified by Jia., et al. \cite{1}, we theoretically realized the topological Lieb lattice, a line centered square lattice with rich topological properties, in a radio-frequency circuit. We open the topological nontrivial band-gap through specific capacitor-inductor network, which resembles adding intrinsic spin orbit coupling term into the tight binding model. Finally, we discuss the extension of the $ϕ=π/2$ phase change of hopping between sites to arbitrary value, and investigate the topological phase transition of the band structure vary with capacitance, thereby paving the way for designing tunable lattices using the presented framework.

cond-mat.mes-hall↗

arXiv2018

Towards Universal Representation for Unseen Action Recognition

Unseen Action Recognition (UAR) aims to recognise novel action categories without training examples. While previous methods focus on inner-dataset seen/unseen splits, this paper proposes a pipeline using a large-scale training source to achieve a Universal Representation (UR) that can generalise to a more realistic Cross-Dataset UAR (CD-UAR) scenario. We first address UAR as a Generalised Multiple-Instance Learning (GMIL) problem and discover 'building-blocks' from the large-scale ActivityNet dataset using distribution kernels. Essential visual and semantic components are preserved in a shared space to achieve the UR that can efficiently generalise to new datasets. Predicted UR exemplars can be improved by a simple semantic adaptation, and then an unseen action can be directly recognised using UR during the test. Without further training, extensive experiments manifest significant improvements over the UCF101 and HMDB51 benchmarks.

cs.CV↗

arXiv2018

Exploring Hidden Acoustic Spin Underwater

Investigating wave propagation in fluid enables a variety of important applications in underwater communications, object detections and unmanned robot control. Conventionally, momentum and spin reveal fundamental physical properties about propagating waves. Yet, vast previous research focused on the orbital angular momentum of acoustics without thinking about the existence possibility of spin due to the longitudinal wave nature. Here, we show that underwater acoustic wave processes the non-trivial spin angular momentum intrinsically, which is associated with its special spin-orbital coupling relation for longitudinal waves. Furthermore, we demonstrate that this intrinsic spin, although unobservable in plane wave form, can be detected by four approaches: wave interference, Gaussian exponential decay form, boundary evanescent wave, and waveguide mode. We further show that the strong spin-orbital coupling can be exploited to achieve unidirectional excitation and backscattering immune transport. We hope the present results can improve the geometric and topological understanding about underwater acoustic wave and pave the way on the spin-related underwater applications.

physics.app-ph↗

arXiv2018

Quantum Spin of Elastic Wave

Unveiling intrinsic spins of propagating waves usually offers people a fundamental understanding of the geometrical and topological properties of waves from classical to quantum aspects. A great variety of research has shown that transverse waves can possess non-trivial quantum spins and topology without help of strong wave-matter interaction. However, until now we still lack essential physical insights about the spin and topological nature of longitudinal waves. Here, demonstrated by elastic waves we uncover unique quantum spins for longitudinal waves and the mixed longitudinal-transverse waves that play essential roles in topological spin-momentum locking. Based on this quantum spin perspective, several abnormal phenomena beyond pure transverse waves are attributed to the hybrid spin induced by mixed longitudinal-transverse waves. The intrinsic hybrid spin reveals the complex spin essence in elastic waves and advances our understanding about their fundamental topological properties. We also show these spin-dependent phenomena can be exploited to control the wave propagation, such as non-symmetric elastic wave excitation by spin pairs, uni-directional Rayleigh wave and spin-selected elastic wave routing. These findings are generally applicable for arbitrary waves with longitudinal and transverse components.

cond-mat.mes-hall↗

arXiv2017

Photonic Spin Hall Effect in Waveguides Composed of Two Types of Single-Negative Metamaterials

The polarization controlled optical signal routing has many important applications in photonics such as polarization beam splitter. By using two-dimensional transmission lines with lumped elements, we experimentally demonstrate the selective excitation of guided modes in waveguides composed of two kinds of single-negative metamaterials. A localized, circularly polarized emitter placed near the interface of the two kinds of single-negative metamaterials only couples with one guided mode with a specific propagating direction determined by the polarization handedness of the source. Moreover, this optical spin-orbit locking phenomenon, also called the photonic spin Hall effect, is robust against interface fluctuations, which may be very useful in the manipulation of electromagnetic signals.

cond-mat.mes-hall↗

arXiv2017

Asymmetric Nonlinear System is Not sufficient for Non-Reciprocal Quantum Wave Diode

We demonstrate symmetric wave propagations in asymmetric nonlinear quantum systems. By solving the nonlinear Schördinger equation, we first analytically prove the existence of symmetric transmission in asymmetric systems with a single nonlinear delta-function interface. We then point out that a finite width of the nonlinear interface region is necessary to produce non-reciprocity in asymmetric systems. However, a geometrical resonant condition for breaking non-reciprocal propagation is then identified theoretically and verified numerically. With such a resonant condition, the nonlinear interface region of finite width behaves like a single nonlinear delta-barrier so that wave propagations in the forward and backward directions are identical under arbitrary incident wave intensity. As such, reciprocity re-emerges periodically in the asymmetric nonlinear system when changing the width of interface region. Finally, similar resonant conditions of discrete nonlinear Schördinger equation are discussed. Therefore, we have identified instances of Reciprocity Theorem that breaking spatial symmetry in nonlinear interface systems is not sufficient to produce non-reciprocal wave propagation.

cond-mat.mes-hall↗

arXiv2017

Topological Landau-Zener Bloch Oscillations in Photonic Floquet Lieb Lattices

The Lieb Lattice exhibits intriguing properties that are of general interest in both the fundamental physics and practical applications. Here, we investigate the topological Landau-Zener Bloch oscillation in a photonic Floquet Lieb lattice, where the dimerized helical waveguides is constructed to realize the synthetic spin-orbital interaction through the Floquet mechanism, rendering us to study the impacts of topological transition from trivial gaps to non-trivial ones. The compact localized states of flat bands supported by the local symmetry of Lieb lattice will be associated with other bands by topological invariants, Chern number, and involved into Landau-Zener transition during Bloch oscillation. Importantly, the non-trivial geometrical phases after topological transitions will be taken into account for constructive and destructive interferences of wave functions. The numerical calculations of continuum photonic medium demonstrate reasonable agreements with theoretical tight-binding model. Our results provide an ongoing effort to realize designed quantum materials with tailored properties.

cond-mat.mes-hall↗

arXiv2017

From Zero-shot Learning to Conventional Supervised Classification: Unseen Visual Data Synthesis

Robust object recognition systems usually rely on powerful feature extraction mechanisms from a large number of real images. However, in many realistic applications, collecting sufficient images for ever-growing new classes is unattainable. In this paper, we propose a new Zero-shot learning (ZSL) framework that can synthesise visual features for unseen classes without acquiring real images. Using the proposed Unseen Visual Data Synthesis (UVDS) algorithm, semantic attributes are effectively utilised as an intermediate clue to synthesise unseen visual features at the training stage. Hereafter, ZSL recognition is converted into the conventional supervised problem, i.e. the synthesised visual features can be straightforwardly fed to typical classifiers such as SVM. On four benchmark datasets, we demonstrate the benefit of using synthesised unseen data. Extensive experimental results suggest that our proposed approach significantly improve the state-of-the-art results.

cs.CV↗