arXiv subjects

Li Zhang

Publications and source records attributed to Li Zhang.

At least 667 records · Page 37Linked to original sources

arXiv2022

SGM3D: Stereo Guided Monocular 3D Object Detection

Monocular 3D object detection aims to predict the object location, dimension and orientation in 3D space alongside the object category given only a monocular image. It poses a great challenge due to its ill-posed property which is critically lack of depth information in the 2D image plane. While there exist approaches leveraging off-the-shelve depth estimation or relying on LiDAR sensors to mitigate this problem, the dependence on the additional depth model or expensive equipment severely limits their scalability to generic 3D perception. In this paper, we propose a stereo-guided monocular 3D object detection framework, dubbed SGM3D, adapting the robust 3D features learned from stereo inputs to enhance the feature for monocular detection. We innovatively present a multi-granularity domain adaptation (MG-DA) mechanism to exploit the network's ability to generate stereo-mimicking features given only on monocular cues. Coarse BEV feature-level, as well as the fine anchor-level domain adaptation, are both leveraged for guidance in the monocular domain.In addition, we introduce an IoU matching-based alignment (IoU-MA) method for object-level domain adaptation between the stereo and monocular predictions to alleviate the mismatches while adopting the MG-DA. Extensive experiments demonstrate state-of-the-art results on KITTI and Lyft datasets.

cs.CV↗

arXiv2022

Learning from Mistakes -- A Framework for Neural Architecture Search

Learning from one's mistakes is an effective human learning technique where the learners focus more on the topics where mistakes were made, so as to deepen their understanding. In this paper, we investigate if this human learning strategy can be applied in machine learning. We propose a novel machine learning method called Learning From Mistakes (LFM), wherein the learner improves its ability to learn by focusing more on the mistakes during revision. We formulate LFM as a three-stage optimization problem: 1) learner learns; 2) learner re-learns focusing on the mistakes, and; 3) learner validates its learning. We develop an efficient algorithm to solve the LFM problem. We apply the LFM framework to neural architecture search on CIFAR-10, CIFAR-100, and Imagenet. Experimental results strongly demonstrate the effectiveness of our model.

cs.LG↗

arXiv2021

Multi-band Emission up to PeV Energy from the Crab Nebula in a Spatially Dependent Lepto-hadronic Model

Multi-band emission from radio to ultra-high energy gamma-rays in the Crab Nebula has been detected. To explain the observed results, non-thermal photon production \textbf{in} the Crab Nebula is carefully studied in a spatially dependent lepto-hadronic model. In our model, the dynamical evolution of the PWN is simulated in a spherically symmetric system. Both electrons and protons are accelerated at the termination shock. The relevant particle propagation equations as well as the photon evolving equation are simultaneously solved. For the Crab Nebula, our results reveal that the observed multi-band photon spectra can be well reproduced with reasonable model parameters. In particular, the photons with energy $\gtrsim 200$ TeV are mainly contributed by the hadronic component via proton-proton interaction. The contribution of the hadronic component depends on both proton spectral index $α_{\rm p}$ and number density $n_{\rm H}$ of medium within the PWN. Besides, high energy neutrino fluxes are predicted with variable proton spectral indices. The predicted fluxes are not only far below the sensitivities of current neutrino observatories, but also beneath the atmospheric neutrino background with energy less than $\sim 40$ TeV. Moreover, the calculated radial profiles of surface brightness and spectral index are presented.

astro-ph.HE↗

arXiv2021

Validity of equivalent photon spectra and the photoproduction processes in p-p collisions

Through a consistent analysis of the terms neglected in going from the accurate expression to the one of Weizsäcker-Williams approximation (WWA), the validity of equivalent photon spectra is studied, and a modified photon flux of proton is also derived. We take the photoproductions of photons and dileptons as examples, to provide the comparison between the exact results and the ones based on various photon fluxes. We present the results for the distributions in $Q^{2}$ (virtuality of photons), $y$ and $p_{T}$, the total cross sections are also estimated. The numerical results show that the modified equivalent photon spectrum reproduces the exact result within less than one percent. And the corrections of photoproduction processes to the dileptons and photons productions are about $20\%$.

hep-ph↗

arXiv2021

The Devil is in the Task: Exploiting Reciprocal Appearance-Localization Features for Monocular 3D Object Detection

Low-cost monocular 3D object detection plays a fundamental role in autonomous driving, whereas its accuracy is still far from satisfactory. In this paper, we dig into the 3D object detection task and reformulate it as the sub-tasks of object localization and appearance perception, which benefits to a deep excavation of reciprocal information underlying the entire task. We introduce a Dynamic Feature Reflecting Network, named DFR-Net, which contains two novel standalone modules: (i) the Appearance-Localization Feature Reflecting module (ALFR) that first separates taskspecific features and then self-mutually reflects the reciprocal features; (ii) the Dynamic Intra-Trading module (DIT) that adaptively realigns the training processes of various sub-tasks via a self-learning manner. Extensive experiments on the challenging KITTI dataset demonstrate the effectiveness and generalization of DFR-Net. We rank 1st among all the monocular 3D object detectors in the KITTI test set (till March 16th, 2021). The proposed method is also easy to be plug-and-play in many cutting-edge 3D detection frameworks at negligible cost to boost performance. The code will be made publicly available.

cs.CV↗

arXiv2021

A Possible Hadronic Origin of TeV Photon Emission from SNR G106.3+2.7

In our model, the acceleration and propagation of particles from the Bohm-like diffusion region inside the SNR to the Galactic diffusion region outside the SNR are described through nonlinear diffusive shock acceleration (NLDSA). The main content of our NLDSA model is to solve the hydrodynamic equations numerically for gas density, gas velocity, gas pressure as well as the equation for the quasi-isotropic particle momentum distribution. The consequent multi-band non-thermal emission is from two different regions, namely the acceleration region and the escaping region. Our model is capable of explaining the multi-band photon emission via the dominant synchrotron radiation of the electrons accelerated inside the SNR; and the photons with energy of $\gtrsim$ GeV are naturally produced by the protons inside and outside the SNR. Moreover, the photons in the energy range of $\sim 1 - \sim 100$ TeV are due to the interaction of escaped protons with dense molecular clouds. For the photons with energy of $E_γ\gtrsim $ 1 GeV from SNR G106.3+2.7, our results here favor a hadronic origin, where the photons in the energy range of $\sim 1$ GeV to $\sim 1$ TeV are produced inside the SNR through proton-proton interaction, while the photons with $E_γ\gtrsim 1$ TeV originate from the interaction of escaped protons with a dense molecular cloud.

astro-ph.HE↗

arXiv2021

Observation of topologically enabled complete polarization conversion

Exploiting topological ideas has been a major theme in modern photonics, which provides unprecedented opportunities to design photonic devices with robustness against defects and flaws. While most previous works in topological photonics have focused on band theory, recent theoretical advances extend the topological concepts to the analysis of scattering matrices and suggest a topological route to complete polarization conversion (CPC), a unique photonic phenomenon without an electronic counterpart. Here, we report on the experimental observation of the topological effect in reflection matrices of a photonic crystal slab, enabling CPC between two linear polarizations over a wide range of frequencies. Using angle-resolved reflection measurements, we observe CPC occurring at vortex singularities of reflection coefficients in momentum space, verifying the topological nature of CPC. In addition, the topological effect also guarantees the spin-preserved reflection of a circularly polarized wave. Remarkably, we experimentally establish a connection between two seemingly unrelated topological phenomena--CPC and bound states in the continuum (BICs): BICs lie on the critical coupling curves that define the condition for CPC. Our work paves the way to exploring the topological properties in scattering matrices for controlling light polarization and spin and creating robust photonic devices.

physics.class-ph↗

arXiv2021

Beijing ZKJ-NPU Speaker Verification System for VoxCeleb Speaker Recognition Challenge 2021

In this report, we describe the Beijing ZKJ-NPU team submission to the VoxCeleb Speaker Recognition Challenge 2021 (VoxSRC-21). We participated in the fully supervised speaker verification track 1 and track 2. In the challenge, we explored various kinds of advanced neural network structures with different pooling layers and objective loss functions. In addition, we introduced the ResNet-DTCF, CoAtNet and PyConv networks to advance the performance of CNN-based speaker embedding model. Moreover, we applied embedding normalization and score normalization at the evaluation stage. By fusing 11 and 14 systems, our final best performances (minDCF/EER) on the evaluation trails are 0.1205/2.8160% and 0.1175/2.8400% respectively for track 1 and 2. With our submission, we came to the second place in the challenge for both tracks.

cs.SD↗

arXiv2021

Acoustic non-Hermitian skin effect from twisted winding topology

The recently discovered non-Hermitian skin effect (NHSE) manifests the breakdown of current classification of topological phases in energy-nonconservative systems, and necessitates the introduction of non-Hermitian band topology. So far, all NHSE observations are based on one type of non-Hermitian band topology, in which the complex energy spectrum winds along a closed loop. As recently characterized along a synthetic dimension on a photonic platform, non-Hermitian band topology can exhibit almost arbitrary windings in momentum space, but their actual phenomena in real physical systems remain unclear. Here, we report the experimental realization of NHSE in a one-dimensional (1D) non-reciprocal acoustic crystal. With direct acoustic measurement, we demonstrate that a twisted winding, whose topology consists of two oppositely oriented loops in contact rather than a single loop, will dramatically change the NHSE, following previous predictions of unique features such as the bipolar localization and the Bloch point for a Bloch-wave-like extended state. This work reveals previously unnoticed features of NHSE, and provides the observation of physical phenomena originating from complex non-Hermitian winding topology.

physics.app-ph↗

arXiv2021

From Cloud to Edge: A First Look at Public Edge Platforms

Public edge platforms have drawn increasing attention from both academia and industry. In this study, we perform a first-of-its-kind measurement study on a leading public edge platform that has been densely deployed in China. Based on this measurement, we quantitatively answer two critical yet unexplored questions. First, from end users' perspective, what is the performance of commodity edge platforms compared to cloud, in terms of the end-to-end network delay, throughput, and the application QoE. Second, from the edge service provider's perspective, how are the edge workloads different from cloud, in terms of their VM subscription, monetary cost, and resource usage. Our study quantitatively reveals the status quo of today's public edge platforms, and provides crucial insights towards developing and operating future edge services.

cs.NI↗

arXiv2021

Revisiting the Performance of iALS on Item Recommendation Benchmarks

Matrix factorization learned by implicit alternating least squares (iALS) is a popular baseline in recommender system research publications. iALS is known to be one of the most computationally efficient and scalable collaborative filtering methods. However, recent studies suggest that its prediction quality is not competitive with the current state of the art, in particular autoencoders and other item-based collaborative filtering methods. In this work, we revisit the iALS algorithm and present a bag of tricks that we found useful when applying iALS. We revisit four well-studied benchmarks where iALS was reported to perform poorly and show that with proper tuning, iALS is highly competitive and outperforms any method on at least half of the comparisons. We hope that these high quality results together with iALS's known scalability spark new interest in applying and further improving this decade old technique.

cs.IR↗

arXiv2021

iALS++: Speeding up Matrix Factorization with Subspace Optimization

iALS is a popular algorithm for learning matrix factorization models from implicit feedback with alternating least squares. This algorithm was invented over a decade ago but still shows competitive quality compared to recent approaches like VAE, EASE, SLIM, or NCF. Due to a computational trick that avoids negative sampling, iALS is very efficient especially for large item catalogues. However, iALS does not scale well with large embedding dimensions, d, due to its cubic runtime dependency on d. Coordinate descent variations, iCD, have been proposed to lower the complexity to quadratic in d. In this work, we show that iCD approaches are not well suited for modern processors and can be an order of magnitude slower than a careful iALS implementation for small to mid scale embedding sizes (d ~ 100) and only perform better than iALS on large embeddings d ~ 1000. We propose a new solver iALS++ that combines the advantages of iALS in terms of vector processing with a low computational complexity as in iCD. iALS++ is an order of magnitude faster than iCD both for small and large embedding dimensions. It can solve benchmark problems like Movielens 20M or Million Song Dataset even for 1000 dimensional embedding vectors in a few minutes.

cs.LG↗

arXiv2021

Photoproductions of dileptons, photons, and light vector mesons in p-p, p-Pb, and Pb-Pb collisions at Large Hadron Collider energies

The photoproduction of large $p_{T}$ dileptons, photons and light vector mesons in p-p, p-Pb and Pb-Pb collisions at LHC energies is studied, where the fragmentation processes and the ultra-incoherent photon channel are included. An exact treatment is developed for photoproduction processes in heavy ion collisions, which recovers the equivalent photon approximation (EPA) in the limit $Q^{2}\rightarrow0$ and can avoid double counting effectively. The full kinematical relations are also achieved. We present the results as the distributions in $Q^{2}$, $p_{T}$ and $y_{r}$, the total cross sections are also estimated. The numerical results indicate that: the contribution of photoproduction processes is evident in the large $p_{T}$ and $y_{r}$ regions, and starts to play a fundamental role in p-Pb collisions. The ultra-incoherent photon emission is a important channel of photoproduction processes, which can provide the meaningful contributions. EPA is only applicable in small $y$ and $Q^{2}$ domains, and is very sensitive to the values of $y_{\mathrm{max}}$ and $Q^{2}_{\mathrm{max}}$. The EPA errors appear when $y>0.29$ and $Q^{2}>0.1~\mathrm{GeV}^{2}$, and are rather serious in p-Pb and Pb-Pb collisions. When dealing with widely kinematical regions, the exact treatment needs to be adopted.

hep-ph↗

arXiv2021

Text-Based Person Search with Limited Data

Text-based person search (TBPS) aims at retrieving a target person from an image gallery with a descriptive text query. Solving such a fine-grained cross-modal retrieval task is challenging, which is further hampered by the lack of large-scale datasets. In this paper, we present a framework with two novel components to handle the problems brought by limited data. Firstly, to fully utilize the existing small-scale benchmarking datasets for more discriminative feature learning, we introduce a cross-modal momentum contrastive learning framework to enrich the training data for a given mini-batch. Secondly, we propose to transfer knowledge learned from existing coarse-grained large-scale datasets containing image-text pairs from drastically different problem domains to compensate for the lack of TBPS training data. A transfer learning method is designed so that useful information can be transferred despite the large domain gap. Armed with these components, our method achieves new state of the art on the CUHK-PEDES dataset with significant improvements over the prior art in terms of Rank-1 and mAP. Our code is available at https://github.com/BrandonHanx/TextReID.

cs.CV↗

arXiv2021

Duality Temporal-channel-frequency Attention Enhanced Speaker Representation Learning

The use of channel-wise attention in CNN based speaker representation networks has achieved remarkable performance in speaker verification (SV). But these approaches do simple averaging on time and frequency feature maps before channel-wise attention learning and ignore the essential mutual interaction among temporal, channel as well as frequency scales. To address this problem, we propose the Duality Temporal-Channel-Frequency (DTCF) attention to re-calibrate the channel-wise features with aggregation of global context on temporal and frequency dimensions. Specifically, the duality attention - time-channel (T-C) attention as well as frequency-channel (F-C) attention - aims to focus on salient regions along the T-C and F-C feature maps that may have more considerable impact on the global context, leading to more discriminative speaker representations. We evaluate the effectiveness of the proposed DTCF attention on the CN-Celeb and VoxCeleb datasets. On the CN-Celeb evaluation set, the EER/minDCF of ResNet34-DTCF are reduced by 0.63%/0.0718 compared with those of ResNet34-SE. On VoxCeleb1-O, VoxCeleb1-E and VoxCeleb1-H evaluation sets, the EER/minDCF of ResNet34-DTCF achieve 0.36%/0.0263, 0.39%/0.0382 and 0.74%/0.0753 reductions compared with those of ResNet34-SE.

cs.SD↗

arXiv2021

Unsupervised Domain Adaptation in Semantic Segmentation Based on Pixel Alignment and Self-Training

This paper proposes an unsupervised cross-modality domain adaptation approach based on pixel alignment and self-training. Pixel alignment transfers ceT1 scans to hrT2 modality, helping to reduce domain shift in the training segmentation model. Self-training adapts the decision boundary of the segmentation network to fit the distribution of hrT2 scans. Experiment results show that PAST has outperformed the non-UDA baseline significantly, and it received rank-2 on CrossMoDA validation phase Leaderboard with a mean Dice score of 0.8395.

cs.CV↗

arXiv2021

Spectroscopic visualization of flat bands in magic-angle twisted monolayer-bilayer graphene: localization-delocalization coexisting electronic states

Recent transport studies have demonstrated the great potential of twisted monolayer-bilayer graphene (tMBG) as a new platform to host moiré flat bands with a higher tunability than twisted bilayer graphene (tBG). However, a direct visualization of the flat bands in tMBG and its comparison with the ones in tBG remain unexplored. Here, via fabricating on a single sample with exactly the same twist angle of ~1.13°, we present a direct comparative study between tMBG and tBG using scanning tunneling microscopy/spectroscopy. We observe a sharp density of states peak near the Fermi energy in tunneling spectroscopy, confirming unambiguously the existence of flat electronic bands in tMBG. The bandwidth of this flat-band peak is found to be slightly narrower than that of the tBG, validating previous theoretical predictions. Remarkably, by measuring spatially resolved spectroscopy, combined with continuum model calculation, we show that the flat-band states in tMBG exhibit a unique layer-resolved localization-delocalization coexisting feature, which offers an unprecedented possibility to utilize their cooperation on exploring novel correlation phenomena. Our work provides important microscopic insight of flat-band states for better understanding the emergent physics in graphene moiré systems.

cond-mat.mes-hall↗

arXiv2021

Visual Goal-Step Inference using wikiHow

Understanding what sequence of steps are needed to complete a goal can help artificial intelligence systems reason about human activities. Past work in NLP has examined the task of goal-step inference for text. We introduce the visual analogue. We propose the Visual Goal-Step Inference (VGSI) task, where a model is given a textual goal and must choose which of four images represents a plausible step towards that goal. With a new dataset harvested from wikiHow consisting of 772,277 images representing human actions, we show that our task is challenging for state-of-the-art multimodal models. Moreover, the multimodal representation learned from our data can be effectively transferred to other datasets like HowTo100m, increasing the VGSI accuracy by 15 - 20%. Our task will facilitate multimodal reasoning about procedural events.

cs.CV↗