arXiv subjects

Kun Luo

Publications and source records attributed to Kun Luo.

43 records · Page 3Linked to original sources

arXiv2020

Cross-Domain Complementary Learning Using Pose for Multi-Person Part Segmentation

Supervised deep learning with pixel-wise training labels has great successes on multi-person part segmentation. However, data labeling at pixel-level is very expensive. To solve the problem, people have been exploring to use synthetic data to avoid the data labeling. Although it is easy to generate labels for synthetic data, the results are much worse compared to those using real data and manual labeling. The degradation of the performance is mainly due to the domain gap, i.e., the discrepancy of the pixel value statistics between real and synthetic data. In this paper, we observe that real and synthetic humans both have a skeleton (pose) representation. We found that the skeletons can effectively bridge the synthetic and real domains during the training. Our proposed approach takes advantage of the rich and realistic variations of the real data and the easily obtainable labels of the synthetic data to learn multi-person part segmentation on real images without any human-annotated labels. Through experiments, we show that without any human labeling, our method performs comparably to several state-of-the-art approaches which require human labeling on Pascal-Person-Parts and COCO-DensePose datasets. On the other hand, if part labels are also available in the real-images during training, our method outperforms the supervised state-of-the-art methods by a large margin. We further demonstrate the generalizability of our method on predicting novel keypoints in real images where no real data labels are available for the novel keypoints detection. Code and pre-trained models are available at https://github.com/kevinlin311tw/CDCL-human-part-segmentation

cs.CV↗

arXiv2019

Tunable anisotropic behaviors in phosphorene under periodic potentials in arbitrary directions

We investigate theoretically the anisotropic electronic and optical behaviors of a monolayer black phosphorus (phosphorene) modulated by periodic potential superlattices in arbitrary directions. We demonstrate that different strength and orientation of the phosphorene potential superlattice can give rise to distinct energy spectra, i.e., tuning the intrinsic electronic anisotropy. Accordingly, the anisotropic effective mass, and optical absorption modulated by superlattice strength and orientation are addressed systematically. This feature enables tuning capability more than one order of magnitude in the optical absorption spectrum. Our findings should be useful in building phosphorene optical and (opto)electronic devices by applying external potential superlattice.

cond-mat.mes-hall↗

arXiv2019

Three metallic BN polymorphs: 1D multi-threaded conduction in 3D network

In this paper, three novel metallic sp2/sp3 -hybridized Boron Nitride (BN) polymorphs are proposed by first-principles calculations. One of them, named as tP-BN, is predicted based on the evolutionary particle swarm structural search. tP-BN is constructed by two interlocked rings forming a tube-like 3D network. The stability and band structure calculations show tP-BN is metastable and metallic at zero pressure. Calculations for the density of states and electron orbits confirm that the metallicity originates from the sp2 -hybridized B and N atoms, and forming 1D linear conductive channels in the 3D network. According to the relationship between the atomic structure and electronic properties, another two 3D metastable metallic sp2/sp3 -hybridized BN structures are constructed manually. Electronic properties calculations show that both of these structures have 1D conductive channel along different axes. The polymorphs predicted in this study enrich the structures and provide a different picture of the conductive mechanism of BN compounds.

cond-mat.mtrl-sci↗

arXiv2019

Graph Nets for Partial Charge Prediction

Atomic partial charges are crucial parameters for Molecular Dynamics (MD) simulations, molecular mechanics calculations, and virtual screening, as they determine the electrostatic contributions to interaction energies. Current methods for calculating partial charges, however, are either slow and scale poorly with molecular size (quantum chemical methods) or unreliable (empirical methods). Here, we present a new charge derivation method based on Graph Nets---a set of update and aggregate functions that operate on molecular topologies and propagate information thereon---that could approximate charges derived from Density Functional Theory (DFT) calculations with high accuracy and an over 500-fold speed up.

physics.comp-ph↗

arXiv2018

Proposal for detecting nodal-line semimetal surface-states with resonant spin-flipped reflection

Topological nodal-line semimetals are predicted to exhibit unique drumhead-like surface states (DSS). Yet, a direct detection of such states remains a challenge. Here, we propose spin-resolved transport in a junction between a normal metal and a spin-orbit coupled nodal-line semimetal as the mechanism for their detection. Specifically, we find that in such a device, the DSS induce resonant spin-flipped reflection. This effect can be probed by both vertical spin transport and lateral charge transport between anti-parallel magnetic terminals. In the tunneling limit of the junction, both spin and charge conductances exhibit a resonant peak around zero energy, providing a unique evidence of the DSS. This signature is robust to both dispersive DSS and interface disorder. Based on numerical calculations, we show that the scheme can be implemented in the topological semimetal HgCr$_2$Se$_4$.

cond-mat.mes-hall↗

arXiv2018

Treatment of solid objects in the Pencil Code using an immersed boundary method and overset grids

Two methods for solid body representation in flow simulations available in the Pencil Code are the immersed boundary method and overset grids. These methods are quite different in terms of computational cost, flexibility and numerical accuracy. We present here an investigation of the use of the different methods with the purpose of assessing their strengths and weaknesses. At present, the overset grid method in the Pencil Code can only be used for representing cylinders in the flow. For this task it surpasses the immersed boundary method in yielding highly accurate solutions at moderate computational costs. This is partly due to local grid stretching and a body-conformal grid, and partly due to the possibility of working with local time step restrictions on different grids. The immersed boundary method makes up the lack of computational efficiency with flexibility in regards to application to complex geometries, due to a recent extension of the method that allows our implementation of it to represent arbitrarily shaped objects in the flow.

physics.comp-ph↗

arXiv2017

Probing valley filtering effect by Andreev reflection in zigzag graphene nanoribbon

Ballistic point contact (BPC) with zigzag edges in graphene is a main candidate of a valley filter, in which the polarization of the valley degree of freedom can be selected by using a local gate voltage. Here, we propose to detect the valley filtering effect by Andreev reflection. Because electrons in the lowest conduction band and the highest valence band of the BPC possess opposite chirality, the inter-band Andreev reflection is strongly suppressed, after multiple scattering and interference. We draw this conclusion by both the scattering matrix analysis and the numerical simulation. The Andreev reflection as a function of the incident energy of electrons and the local gate voltage at the BPC is obtained, by which the parameter region for a perfect valley filter and the direction of valley polarization can be determined. The Andreev reflection exhibits an oscillatory decay with the length of the BPC, indicating a negative correlation to valley polarization.

cond-mat.mes-hall↗