arXiv subjects

Xiaodong Yang

Publications and source records attributed to Xiaodong Yang.

At least 109 records · Page 6Linked to original sources

arXiv2018

Models Matter, So Does Training: An Empirical Study of CNNs for Optical Flow Estimation

We investigate two crucial and closely related aspects of CNNs for optical flow estimation: models and training. First, we design a compact but effective CNN model, called PWC-Net, according to simple and well-established principles: pyramidal processing, warping, and cost volume processing. PWC-Net is 17 times smaller in size, 2 times faster in inference, and 11\% more accurate on Sintel final than the recent FlowNet2 model. It is the winning entry in the optical flow competition of the robust vision challenge. Next, we experimentally analyze the sources of our performance gains. In particular, we use the same training procedure of PWC-Net to retrain FlowNetC, a sub-network of FlowNet2. The retrained FlowNetC is 56\% more accurate on Sintel final than the previously trained one and even 5\% more accurate than the FlowNet2 model. We further improve the training procedure and increase the accuracy of PWC-Net on Sintel by 10\% and on KITTI 2012 and 2015 by 20\%. Our newly trained model parameters and training protocols will be available on https://github.com/NVlabs/PWC-Net

cs.CV↗

arXiv2018

PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume

We present a compact but effective CNN model for optical flow, called PWC-Net. PWC-Net has been designed according to simple and well-established principles: pyramidal processing, warping, and the use of a cost volume. Cast in a learnable feature pyramid, PWC-Net uses the cur- rent optical flow estimate to warp the CNN features of the second image. It then uses the warped features and features of the first image to construct a cost volume, which is processed by a CNN to estimate the optical flow. PWC-Net is 17 times smaller in size and easier to train than the recent FlowNet2 model. Moreover, it outperforms all published optical flow methods on the MPI Sintel final pass and KITTI 2015 benchmarks, running at about 35 fps on Sintel resolution (1024x436) images. Our models are available on https://github.com/NVlabs/PWC-Net.

cs.CV↗

arXiv2018

Budget-Aware Activity Detection with A Recurrent Policy Network

In this paper, we address the challenging problem of efficient temporal activity detection in untrimmed long videos. While most recent work has focused and advanced the detection accuracy, the inference time can take seconds to minutes in processing each single video, which is too slow to be useful in real-world settings. This motivates the proposed budget-aware framework, which learns to perform activity detection by intelligently selecting a small subset of frames according to a specified time budget. We formulate this problem as a Markov decision process, and adopt a recurrent network to model the frame selection policy. We derive a recurrent policy gradient based approach to approximate the gradient of the non-decomposable and non-differentiable objective defined in our problem. In the extensive experiments, we achieve competitive detection accuracy, and more importantly, our approach is able to substantially reduce computation time and detect multiple activities with only 0.35s for each untrimmed long video.

cs.CV↗

arXiv2017

MoCoGAN: Decomposing Motion and Content for Video Generation

Visual signals in a video can be divided into content and motion. While content specifies which objects are in the video, motion describes their dynamics. Based on this prior, we propose the Motion and Content decomposed Generative Adversarial Network (MoCoGAN) framework for video generation. The proposed framework generates a video by mapping a sequence of random vectors to a sequence of video frames. Each random vector consists of a content part and a motion part. While the content part is kept fixed, the motion part is realized as a stochastic process. To learn motion and content decomposition in an unsupervised manner, we introduce a novel adversarial learning scheme utilizing both image and video discriminators. Extensive experimental results on several challenging datasets with qualitative and quantitative comparison to the state-of-the-art approaches, verify effectiveness of the proposed framework. In addition, we show that MoCoGAN allows one to generate videos with same content but different motion as well as videos with different content and same motion.

cs.CV↗

arXiv2017

Hybrid Quantum-Classical Approach to Quantum Optimal Control

A central challenge in quantum computing is to identify more computational problems for which utilization of quantum resources can offer significant speedup. Here, we propose a hybrid quantum-classical scheme to tackle the quantum optimal control problem. We show that the most computationally demanding part of gradient-based algorithms, namely computing the fitness function and its gradient for a control input, can be accomplished by the process of evolution and measurement on a quantum simulator. By posing queries to and receiving messages from the quantum simulator, classical computing devices update the control parameters until an optimal control solution is found. To demonstrate the quantum-classical scheme in experiment, we use a nine-spin nuclear magnetic resonance system, on which we have succeeded in preparing a seven-correlated quantum state without involving classical computation of the large Hilbert space evolution.

quant-ph↗

arXiv2015

Nonlocal effective medium approximation for metallic nanorod metamaterials

We present an analytical nonlocal effective medium approximation to describe the optical nonlocal effects in metallic nanorod metamaterials based on Mie scattering theory. It is shown that the developed nonlocal effective medium theory can predict a coexistence state of two modes around the epsilon-near-zero region, where strong optical nonlocal effects lead to the behavior of both positive refraction and negative refraction in the nanorod metamaterials. Outside of the coexistence region, only one mode can be excited and its behavior can be well described using the local effective medium theory.

physics.optics↗

arXiv2015

Nonlocal effective medium analysis in symmetric metal-dielectric multilayer metamaterials

The optical nonlocality in symmetric metal-dielectric multilayer metamaterials is theoretically and experimentally investigated with respect to transverse-magnetic-polarized incident light. A nonlocal effective medium theory is derived from the transfer-matrix method to determine the nonlocal effective permittivity depending on both the frequency and wave vector in a symmetric metal-dielectric multilayer stack. In contrast to the local effective medium theory, our proposed nonlocal effective medium theory can accurately predict measured incident angle-dependent reflection spectra from a fabricated multilayer stack and provide nonlocal dispersion relations. Moreover, the bulk plasmon polaritons with large wave vectors supported in the multilayer stack are also investigated with the nonlocal effective medium theory through the analysis of the dispersion relation and eigenmode.

physics.optics↗

arXiv2015

Diffraction-free optical beam propagation with near-zero phase variation in extremely anisotropic metamaterials

Extremely anisotropic metal-dielectric multilayer metamaterials are designed to have the effective permittivity tensor of a transverse component (parallel to the interfaces of the multilayer) with zero real part and a longitudinal component (normal to the interfaces of the multilayer) with ultra-large imaginary part at the same wavelength, including the optical nonlocality analysis based on the transfer-matrix method. The diffraction-free deep-subwavelength optical beam propagation with near-zero phase variation in the designed multilayer stack due to the near-flat iso-frequency contour is demonstrated and analyzed, including the effects of the multilayer period and the material loss.

physics.optics↗

arXiv2014

Realizing broadband electromagnetic transparency with a graded-permittivity sphere

Broadband electromagnetic transparency phenomenon is realized with a well-designed graded-permittivity sphere, which has an extremely low scattering cross section over a wide frequency range, based on the generalized Mie scattering theory and numerical simulation in full-wave condition. The dynamic polarization cancellation is revealed by studying the variation of the polarization with respect to the frequency. Furthermore, a properly-designed multi-shell sphere is also proposed and examined in order to reduce the rigorous conditions for realizing the broadband transparency in experiments.

physics.optics↗

arXiv2014

Experimental realization of a carpet cloak for temperature field and heat flux

Based on transformation optics (TO), we present and experimentally realize a new thermal carpet cloak. The device, which we call a "thermal carpet", provides a considerable cloaking effect. The device is designed, fabricated and measured to verify the thermal cloaking performance. In comparison with previous experimental work, the advantage of this design is that the required medium parameter is inherently isotropic and thus easier to fabricate.

physics.optics↗

arXiv2014

Plasmonic Brownian ratchet

Here we present a Brownian ratchet based on plasmonic interactions. By periodically turning on and off a laser beam that illuminates a periodic array of plasmonic nanostructures with broken spatial symmetry, the random thermal motion of a subwavelength dielectric bead is rectified into one direction. By means of the Molecular Dynamics technique we show a statistical directed drift in particle flow.

cond-mat.mes-hall↗

arXiv2013

Quantum entanglement in plasmonic waveguides with near-zero mode indices

We investigate the quantum entanglement between two quantum dots in a plasmonic waveguide with near-zero mode index, considering the dependence of concurrence on interdot distance, quantum dot-waveguide frequency detuning and coupling strength ratio. High concurrence is achieved for a wide range of interdot distance due to the near-zero mode index, which largely relaxes the strict requirement of interdot distance in conventional dielectric waveguides or metal nanowires. The proposed quantum dot-waveguide system with near-zero phase variation along the waveguide near the mode cutoff frequency shows very promising potential in quantum optics and quantum information processing.

physics.optics↗

arXiv2013

Experimental demonstration of near-infrared epsilon-near-zero multilayer metamaterial slabs

Near-infrared epsilon-near-zero (ENZ) metamaterial slabs based on silver-germanium (Ag-Ge) multilayers are experimentally demonstrated. Transmission, reflection and absorption spectra are characterized and used to determine the complex refractive indices and the effective permittivities of the ENZ metamaterial slabs, which match the results obtained from both the numerical simulations and the optical nonlocalities analysis. A rapid post-annealing process is used to reduce the collision frequency of silver and therefore decrease the optical absorption loss of multilayer metamaterial slabs. Furthermore, multilayer grating structures are studied to enhance the optical transmission and also tune the location of ENZ wavelength. The demonstrated near-infrared ENZ multilayer metamaterial slabs are important for realizing many exotic applications, such as phase front shaping and engineering of photonic density of states.

physics.optics↗

arXiv2013

Giant optical nonlocality near the Dirac point in metal-dielectric multilayer metamaterials

The giant optical nonlocality near the Dirac point in lossless metal-dielectric multilayer metamaterials is revealed and investigated through the analysis of the band structure of the multilayer stack in the three-dimensional omega-k space, according to the transfer-matrix method with the optical nonlocal effect. The position of the Dirac point is analytically located in the omega-k space. It is revealed that the emergence of the Dirac point is due to the degeneracy of the symmetric and the asymmetric eigenmodes of the coupled surface plasmon polaritons. The optical nonlocality induced epsilon-near-zero frequency shift for the multilayer stack compared to the effective medium is studied. Furthermore, the giant optical nonlocality around the Dirac point is explored with the iso-frequency contour analysis, while the beam splitting phenomenon at the Dirac point due to the optical nonlocal effect is also demonstrated.

physics.optics↗

arXiv2013

Anomalous optical coupling between two silicon wires of a slot waveguide in epsilon-near-zero metamaterials

Anomalous optical coupling properties between two silicon wires in a silicon slot waveguide embedded in epsilon-near-zero (ENZ) metamaterials are proposed and demonstrated. The dependences of optical field enhancement in the slot region and transverse optical force on the slot size and the permittivity of surrounding material are studied in details. It is demonstrated that the optical field in the slot region is significantly enhanced due to the giant index contrast at the slot interface between silicon wires and ENZ metamaterials, but the optical mode coupling between silicon wires is greatly reduced so that the transverse optical force is suppressed into almost zero. Moreover, metal-dielectric multilayer structures are designed to realize ENZ metamaterials in the slot region for achieving the electric field enhancement.

physics.optics↗

arXiv2013

Experimental realization of epsilon-near-zero metamaterial slabs with metal-dielectric multilayers

Epsilon-near-zero (ENZ) metamaterial slabs at visible frequencies based on metal-dielectric multilayers are experimentally realized. Transmission, reflection and absorption spectra are measured and used to determine the complex refractive indices and the effective permittivities of the ENZ slabs, which agree with the results obtained from both the numerical simulations and the optical nonlocalities analysis. Furthermore, light propagation in ENZ slabs and directional emission from ENZ prisms are also analyzed. The accurate determination of the ENZ wavelength for metal-dielectric multilayer metamaterial slabs is important for realizing many unique applications, such as phase front manipulation and enhancement of photonic density of states.

physics.optics↗

arXiv2013

Broadband Epsilon-Near-Zero Metamaterials with Step-Like Metal-Dielectric Multilayer Structures

The concept of the broadband epsilon-near-zero meta-atom consisting of layered stacks with specified metallic filling ratio and thickness is proposed based on the Bergman spectral representation of the effective permittivity. The step-like metal-dielectric multilayer structures are designed to achieve realistic broadband epsilon-near-zero meta-atoms in optical frequency range. These meta-atoms can be integrated as building blocks for unconventional optical components with exotic electromagnetic properties over a wide frequency range, such as the demonstrated broadband directional emission and phase front shaping.

physics.optics↗

arXiv2013

Deep subwavelength beam propagation in extremely loss-anisotropic metamaterials

Metal-dielectric multilayer metamaterials with extreme loss-anisotropy, in which the longitudinal component of the permittivity tensor has ultra-large imaginary part, are proposed and designed. Diffraction-free deep subwavelength beam propagation and manipulation, due to the nearly flat iso-frequency contour (IFC), is demonstrated in such loss-anisotropic metamaterials. It is also shown that deep subwavelength beam propagation can be realized in practical multilayer structures with large multilayer period, when the nonlocal effect is considered.

physics.optics↗