arXiv subjects

Ze Liu

Publications and source records attributed to Ze Liu.

51 records · Page 3Linked to original sources

arXiv2021

Video Swin Transformer

The vision community is witnessing a modeling shift from CNNs to Transformers, where pure Transformer architectures have attained top accuracy on the major video recognition benchmarks. These video models are all built on Transformer layers that globally connect patches across the spatial and temporal dimensions. In this paper, we instead advocate an inductive bias of locality in video Transformers, which leads to a better speed-accuracy trade-off compared to previous approaches which compute self-attention globally even with spatial-temporal factorization. The locality of the proposed video architecture is realized by adapting the Swin Transformer designed for the image domain, while continuing to leverage the power of pre-trained image models. Our approach achieves state-of-the-art accuracy on a broad range of video recognition benchmarks, including on action recognition (84.9 top-1 accuracy on Kinetics-400 and 86.1 top-1 accuracy on Kinetics-600 with ~20x less pre-training data and ~3x smaller model size) and temporal modeling (69.6 top-1 accuracy on Something-Something v2). The code and models will be made publicly available at https://github.com/SwinTransformer/Video-Swin-Transformer.

cs.CV↗

arXiv2021

Enhanced spin-orbit coupling and orbital moment in ferromagnets by electron correlations

In atomic physics, the Hund rule says that the largest spin and orbital state is realized due to the interplay of the spin-orbit coupling (SOC) and the Coulomb interactions. Here, we show that in ferromagnetic solids the effective SOC and the orbital magnetic moment can be dramatically enhanced by a factor of $1/[1-(2U^\prime-U-J_H)ρ_0]$, where $U$ and $U^\prime$ are the on-site Coulomb interaction within the same oribtals and between different orbitals, respectively, $J_H$ is the Hund coupling, and $ρ_0$ is the average density of states. This factor is obtained by using the two-orbital as well as five-orbital Hubbard models with SOC. We also find that the spin polarization is more favorable than the orbital polarization, being consistent with experimental observations. This present work provides a fundamental basis for understanding the enhancements of SOC and orbital moment by Coulomb interactions in ferromagnets, which would have wide applications in spintronics.

cond-mat.str-el↗

arXiv2021

Group-Free 3D Object Detection via Transformers

Recently, directly detecting 3D objects from 3D point clouds has received increasing attention. To extract object representation from an irregular point cloud, existing methods usually take a point grouping step to assign the points to an object candidate so that a PointNet-like network could be used to derive object features from the grouped points. However, the inaccurate point assignments caused by the hand-crafted grouping scheme decrease the performance of 3D object detection. In this paper, we present a simple yet effective method for directly detecting 3D objects from the 3D point cloud. Instead of grouping local points to each object candidate, our method computes the feature of an object from all the points in the point cloud with the help of an attention mechanism in the Transformers \cite{vaswani2017attention}, where the contribution of each point is automatically learned in the network training. With an improved attention stacking scheme, our method fuses object features in different stages and generates more accurate object detection results. With few bells and whistles, the proposed method achieves state-of-the-art 3D object detection performance on two widely used benchmarks, ScanNet V2 and SUN RGB-D. The code and models are publicly available at \url{https://github.com/zeliu98/Group-Free-3D}

cs.CV↗

arXiv2021

RGB-D Salient Object Detection via 3D Convolutional Neural Networks

RGB-D salient object detection (SOD) recently has attracted increasing research interest and many deep learning methods based on encoder-decoder architectures have emerged. However, most existing RGB-D SOD models conduct feature fusion either in the single encoder or the decoder stage, which hardly guarantees sufficient cross-modal fusion ability. In this paper, we make the first attempt in addressing RGB-D SOD through 3D convolutional neural networks. The proposed model, named RD3D, aims at pre-fusion in the encoder stage and in-depth fusion in the decoder stage to effectively promote the full integration of RGB and depth streams. Specifically, RD3D first conducts pre-fusion across RGB and depth modalities through an inflated 3D encoder, and later provides in-depth feature fusion by designing a 3D decoder equipped with rich back-projection paths (RBPP) for leveraging the extensive aggregation ability of 3D convolutions. With such a progressive fusion strategy involving both the encoder and decoder, effective and thorough interaction between the two modalities can be exploited and boost the detection accuracy. Extensive experiments on six widely used benchmark datasets demonstrate that RD3D performs favorably against 14 state-of-the-art RGB-D SOD approaches in terms of four key evaluation metrics. Our code will be made publicly available: https://github.com/PPOLYpubki/RD3D.

cs.CV↗

arXiv2020

A Closer Look at Local Aggregation Operators in Point Cloud Analysis

Recent advances of network architecture for point cloud processing are mainly driven by new designs of local aggregation operators. However, the impact of these operators to network performance is not carefully investigated due to different overall network architecture and implementation details in each solution. Meanwhile, most of operators are only applied in shallow architectures. In this paper, we revisit the representative local aggregation operators and study their performance using the same deep residual architecture. Our investigation reveals that despite the different designs of these operators, all of these operators make surprisingly similar contributions to the network performance under the same network input and feature numbers and result in the state-of-the-art accuracy on standard benchmarks. This finding stimulate us to rethink the necessity of sophisticated design of local aggregation operator for point cloud processing. To this end, we propose a simple local aggregation operator without learnable weights, named Position Pooling (PosPool), which performs similarly or slightly better than existing sophisticated operators. In particular, a simple deep residual network with PosPool layers achieves outstanding performance on all benchmarks, which outperforms the previous state-of-the methods on the challenging PartNet datasets by a large margin (7.4 mIoU). The code is publicly available at https://github.com/zeliu98/CloserLook3D

cs.CV↗

arXiv2019

A new twisting phase transition in layered materials

Twisted two-dimensional (2D) layered materials exhibit many novel and unique phenomena, such as insulation and superconductivity transition, and superlubricity. However, the effect of twisting on these phenomena remains unclear. A key challenge is the lack of linkage between the twisting and the material properties. Here, twisting induced moiré pattern is directly correlated with the configuration entropy of the bilayer graphene, based on which a new twisting phase transition in layered materials is uncovered. Most critically, the twisting phase transition is found size dependent. The diversity of phase transition among the AB stacking and the twisted phases could provide insights into the twistronics and twistribology.

cond-mat.mes-hall↗

arXiv2018

Spatially Heterogeneous Dynamics in a Metallic Glass Forming Liquid Imaged by Electron Correlation Microscopy

Supercooled liquids exhibit spatial heterogeneity in the dynamics of their fluctuating atomic arrangements. The length and time scales of the heterogeneous dynamics are central to the glass transition and influence nucleation and growth of crystals from the liquid. We report direct experimental visualization of the spatially heterogeneous dynamics as a function of temperature in the supercooled liquid state of a Pt-based metallic glass, using electron correlation microscopy with sub-nanometer resolution. An experimental four point space-time intensity correlation function demonstrates a growing dynamic correlation length, $ξ$, upon cooling of the liquid toward the glass transition temperature. $ξ$ as a function of the relaxation time $τ$ data are in the good agreement with the Adam-Gibbs, inhomogeneous mode coupling theory and random first order transition theory of the glass transition. The same experiments demonstrate the existence of a nanometer thickness near surface layer with order of magnitude shorter relaxation time than inside the bulk.

cond-mat.mtrl-sci↗

arXiv2016

One-step fabrication of metal nanostructures by high-throughput imprinting

Direct nanoimprinting provides a simple and high-throughput route for producing uniform nanopatterns at great precision and at low costs. However, applying this technique to crystalline metals has been considered as impossible due to intrinsic limitation from grain size effect. Here we demonstrate direct superplastic nanoimprinting (SPNI) of crystalline metals well below their melting temperatures (Tm), generating ordered nanowire arrays with aspect ratio up to ~2000. Our investigations of replicating metal hierarchical nanostructures show the capability of imprinting features as small as 8 nm, far smaller than the grain size of bulk metals. Most surprisingly, the prepared metal hierarchical nanostructures were found possessing perfect monocrystalline structures. These findings indicate that nanoimprinting of crystalline metals below Tm might be from lattice diffusion. SPNI as a one-step and highly controlled high-throughput fabrication method, could facilitate the applications of metal nanostructures in bio-sensing, diagnostic imaging, catalysis, food industry and environmental conservation.

cond-mat.mes-hall↗

arXiv2015

Shear accelerated crystallization in a supercooled atomic liquid

A bulk metallic glass forming alloy is subjected to shear flow in its supercooled state by compression of a short rod to produce a flat disc. The resulting material exhibits enhanced crystallization kinetics during isothermal annealing as reflected in the decrease of the crystallization time relative to the non-deformed case. The transition from quiescent to shear-accelerated crystallization is linked to strain accumulated during shear flow above a critical shear rate $\dotγ_c\approx 0.3$ s$^{-1}$ which corresponds to Péclet number, $Pe\sim\mathcal{O}(1)$. The observation of shear accelerated crystallization in an atomic system at modest shear rates is uncommon. It is made possible here by the substantial viscosity of the supercooled liquid which increases strongly with temperature in the approach to the glass transition. We may therefore anticipate the encounter of non-trivial shear-related effects during thermoplastic deformation of similar systems.

cond-mat.mtrl-sci↗

arXiv2014

Mechanical buckling induced periodic kinking/stripe microstructures in mechanically peeled graphite flakes from HOPG

Mechanical exfoliation is a widely used method to isolate high quality graphene layers from bulk graphite. In our recent experiments, some ordered microstructures, consisting of a periodic alternation of kinks and stripes, were observed in thin graphite flakes that were mechanically peeled from highly oriented pyrolytic graphite (HOPG). A theoretical model is presented in this paper to understand the formation of such ordered microstructures, based on elastic buckling of a graphite flake being subjected to a bending moment. The width of the stripes predicted from this model agrees reasonably well with our experimental measurements.

cond-mat.mtrl-sci↗

arXiv2013

Binding and interlayer force in the near-contact region of two graphite slabs: experiment and theory

Via a novel experiment, Liu \emph{et al.} [Phys. Rev. B, {\bf 85}, 205418 (2012)] estimated the graphite binding energy, specifically the cleavage energy, an important physical property of bulk graphite. We re-examine the data analysis and note that within the standard Lennard-Jones model employed, there are difficulties in achieving internal consistency in the reproduction of the graphite elastic properties. By employing similar models which guarantee consistency with the elastic constant, we find a wide range of model dependent binding energy values from the same experimental data. We attribute some of the difficulty in the determination of the binding energy to: i) limited theoretical understanding of the van der Waals dispersion of graphite cleavage, ii) the mis-match between the strong bending stiffness of the graphite-SiO$_2$ cantilever and the weak asymptotic inter-layer forces that are integrated over to produce the binding energy. We find, however, that the data does support determination of a maximum inter-layer force that is relatively model independent. We conclude that the peak force per unit area is $1.1 \pm 0.15$GPa for cleavage, and occurs at an inter-layer spacing of $0.377\pm 0.013$nm.

cond-mat.mtrl-sci↗

arXiv2012

Interlayer shear strength of single crystalline graphite

Reported values (0.2 MPa ~ 7.0 GPa) of the interlayer shear strength (ISS) of graphite are very dispersed. The main challenge to obtain a reliable value of ISS is the lack of precise experimental methods. Here we present a novel experimental approach to measure the ISS, and obtain the value as 0.14 GPa. Our result can serve as an important basis for understanding mechanical behavior of graphite or graphene-based materials.

cond-mat.mtrl-sci↗

arXiv2011

Self-retracting motion of graphite micro-flakes: superlubricity in micrometer scale

Through experimental study, we reveal superlubricity as the mechanism of self-retracting motion of micrometer sized graphite flakes on graphite platforms by correlating respectively the lock-up or self-retraction states with the commensurate or incommensurate contacts. We show that the scale-dependent loss of self-retractability is caused by generation of contact interfacial defects. A HOPG structure is also proposed to understand our experimental observations, particularly in term of the polycrystal structure. The realisation of the superlubricity in micrometer scale in our experiments will have impact in the design and fabrication of micro/nanoelectromechanical systems based on graphitic materials.

cond-mat.mes-hall↗

arXiv2011

Interlayer binding energy of graphite -- A direct experimental determination

Despite interlayer binding energy is one of the most important material properties for graphite, there is still lacking report on its direct experimental determination. In this paper, we present a novel experimental method to directly measure the interlayer binding energy of highly oriented pyrolytic graphite (HOPG). The obtained values of the binding energy are 0.27($\pm $0.02)J/m$^{2}$, which can serve as a benchmark for other theoretical and experimental works.

cond-mat.mes-hall↗

arXiv2010

Graphite Nanoeraser

We present here a method for cleaning intermediate-size (5~50nm) contamination from highly oriented pyrolytic graphite. Electron beam deposition causes a continuous increase of carbonaceous material on graphene and graphite surfaces, which is difficult to remove by conventional techniques. Direct mechanical wiping using a graphite nanoeraser is observed to drastically reduce the amount of contamination. After the mechanical removal of contamination, the graphite surfaces were able to self-retract after shearing, indicating that van der Waals contact bonding is restored. Since contact bonding provides an indication of a level of cleanliness normally only attainable in a high-quality clean-room, we discuss potential applications in preparation of ultraclean surfaces.

cond-mat.mtrl-sci↗