arXiv subjects

Guo Lu

Publications and source records attributed to Guo Lu.

60 records · Page 4Linked to original sources

arXiv2020

Scope Head for Accurate Localization in Object Detection

Existing anchor-based and anchor-free object detectors in multi-stage or one-stage pipelines have achieved very promising detection performance. However, they still encounter the design difficulty in hand-crafted 2D anchor definition and the learning complexity in 1D direct location regression. To tackle these issues, in this paper, we propose a novel detector coined as ScopeNet, which models anchors of each location as a mutually dependent relationship. This approach quantizes the prediction space and employs a coarse-to-fine strategy for localization. It achieves superior flexibility as in the regression based anchor-free methods, while produces more precise prediction. Besides, an inherit anchor selection score is learned to indicate the localization quality of the detection result, and we propose to better represent the confidence of a detection box by combining the category-classification score and the anchor-selection score. With our concise and effective design, the proposed ScopeNet achieves state-of-the-art results on COCO

cs.CV↗

arXiv2020

Content Adaptive and Error Propagation Aware Deep Video Compression

Recently, learning based video compression methods attract increasing attention. However, the previous works suffer from error propagation due to the accumulation of reconstructed error in inter predictive coding. Meanwhile, the previous learning based video codecs are also not adaptive to different video contents. To address these two problems, we propose a content adaptive and error propagation aware video compression system. Specifically, our method employs a joint training strategy by considering the compression performance of multiple consecutive frames instead of a single frame. Based on the learned long-term temporal information, our approach effectively alleviates error propagation in reconstructed frames. More importantly, instead of using the hand-crafted coding modes in the traditional compression systems, we design an online encoder updating scheme in our system. The proposed approach updates the parameters for encoder according to the rate-distortion criterion but keeps the decoder unchanged in the inference stage. Therefore, the encoder is adaptive to different video contents and achieves better compression performance by reducing the domain gap between the training and testing datasets. Our method is simple yet effective and outperforms the state-of-the-art learning based video codecs on benchmark datasets without increasing the model size or decreasing the decoding speed.

eess.IV↗

arXiv2019

DVC: An End-to-end Deep Video Compression Framework

Conventional video compression approaches use the predictive coding architecture and encode the corresponding motion information and residual information. In this paper, taking advantage of both classical architecture in the conventional video compression method and the powerful non-linear representation ability of neural networks, we propose the first end-to-end video compression deep model that jointly optimizes all the components for video compression. Specifically, learning based optical flow estimation is utilized to obtain the motion information and reconstruct the current frames. Then we employ two auto-encoder style neural networks to compress the corresponding motion and residual information. All the modules are jointly learned through a single loss function, in which they collaborate with each other by considering the trade-off between reducing the number of compression bits and improving quality of the decoded video. Experimental results show that the proposed approach can outperform the widely used video coding standard H.264 in terms of PSNR and be even on par with the latest standard H.265 in terms of MS-SSIM. Code is released at https://github.com/GuoLusjtu/DVC.

eess.IV↗

arXiv2018

Shock-induced plasticity of semi-coherent 111 Cu-Ni multilayers

Using atomistic simulations, dislocation dynamics modeling, and continuum elastic-plastic stress-wave theory, we present a systematic investigation on shock-induced plasticity in semi-coherent CuNi multilayers. The features of stress wave evolutions in the multilayers, including wave-front stress attenuation and strong interfacial discontinuities, are revealed by atomistic simulations. Continuum models are proposed to explain the shockwave propagation features. The simulations provide insight into microplasticity behaviors including interactions between lattice and misfit dislocations. The formation of hybrid Lomer-Cottrell locks through the attraction and combination of lattice and misfit dislocations is a major mechanism for trapping gliding lattice dislocations at interfaces. The relationship between dislocation activity and dynamic stress wave evolution history is explored. The hybrid Lomer-Cottrell locks can dissociate under shock compression or reverse yielding. This dissociation facilitates slip transmission. The influence of coherent stress causes direction dependency in the slip transmission: a lattice dislocation is transmitted more smoothly across an interface from Ni to Cu than from Cu to Ni. The interaction forces between lattice and misfit dislocations are calculated using dislocation dynamics code. Lattice dislocation nucleation from semi-coherent interfaces under shock compression is also reported.

physics.comp-ph↗

arXiv2016

The Thermodynamical Instability Induced by Pressure Ionization in Fluid Helium

A systematic study of pressure ionization is carried out in the chemical picture by the example of fluid helium. By comparing the variants of the chemical model, it is demonstrated that the behavior of pressure ionization depends on the construction of the free energy function. In the chemical model with the Coulomb free energy described by the Padé interpolation formula, thermodynamical instability induced by pressure ionization is found to be manifested by a discontinuous drop or a continuous fall and rise along the pressure-density curve as well as the pressure-temperature curve, which is very much like the first order liquid-liquid phase transition of fluid hydrogen from the first principles simulations. In contrast, in the variant chemical model with the Coulomb free energy term empirically weakened, no thermodynamical instability is induced when pressure ionization occurs, and the resulting equation of state achieves good agreement with the first principles simulations of fluid helium.

physics.plasm-ph↗

arXiv2010

Cluster Identification and Characterization of Physical Fields

The description of complex configuration is a difficult issue. We present a powerful technique for cluster identification and characterization. The scheme is designed to treat with and analyze the experimental and/or simulation data from various methods. Main steps are as follows. We first divide the space using face or volume elements from discrete points. Then, combine the elements with the same and/or similar properties to construct clusters with special physical characterizations. In the algorithm, we adopt administrative structure of hierarchy-tree for spatial bodies such as points, lines, faces, blocks, and clusters. Two fast search algorithms with the complexity are realized. The establishing of the hierarchy-tree and the fast searching of spatial bodies are general, which are independent of spatial dimensions. Therefore, it is easy to extend the skill to other fields. As a verification and validation, we treated with and analyzed some two-dimensional and three-dimensional random data.

cond-mat.stat-mech↗