arXiv subjects

Jian Sun

Publications and source records attributed to Jian Sun.

At least 343 records · Page 19Linked to original sources

arXiv2020

Rethinking Learnable Tree Filter for Generic Feature Transform

The Learnable Tree Filter presents a remarkable approach to model structure-preserving relations for semantic segmentation. Nevertheless, the intrinsic geometric constraint forces it to focus on the regions with close spatial distance, hindering the effective long-range interactions. To relax the geometric constraint, we give the analysis by reformulating it as a Markov Random Field and introduce a learnable unary term. Besides, we propose a learnable spanning tree algorithm to replace the original non-differentiable one, which further improves the flexibility and robustness. With the above improvements, our method can better capture long-range dependencies and preserve structural details with linear complexity, which is extended to several vision tasks for more generic feature transform. Extensive experiments on object detection/instance segmentation demonstrate the consistent improvements over the original version. For semantic segmentation, we achieve leading performance (82.1% mIoU) on the Cityscapes benchmark without bells-and-whistles. Code is available at https://github.com/StevenGrove/LearnableTreeFilterV2.

cs.CV↗

arXiv2020

Fine-Grained Dynamic Head for Object Detection

The Feature Pyramid Network (FPN) presents a remarkable approach to alleviate the scale variance in object representation by performing instance-level assignments. Nevertheless, this strategy ignores the distinct characteristics of different sub-regions in an instance. To this end, we propose a fine-grained dynamic head to conditionally select a pixel-level combination of FPN features from different scales for each instance, which further releases the ability of multi-scale feature representation. Moreover, we design a spatial gate with the new activation function to reduce computational complexity dramatically through spatially sparse convolutions. Extensive experiments demonstrate the effectiveness and efficiency of the proposed method on several state-of-the-art detection benchmarks. Code is available at https://github.com/StevenGrove/DynamicHead.

cs.CV↗

arXiv2020

Spherical Motion Dynamics: Learning Dynamics of Neural Network with Normalization, Weight Decay, and SGD

In this work, we comprehensively reveal the learning dynamics of neural network with normalization, weight decay (WD), and SGD (with momentum), named as Spherical Motion Dynamics (SMD). Most related works study SMD by focusing on "effective learning rate" in "equilibrium" condition, where weight norm remains unchanged. However, their discussions on why equilibrium condition can be reached in SMD is either absent or less convincing. Our work investigates SMD by directly exploring the cause of equilibrium condition. Specifically, 1) we introduce the assumptions that can lead to equilibrium condition in SMD, and prove that weight norm can converge at linear rate with given assumptions; 2) we propose "angular update" as a substitute for effective learning rate to measure the evolving of neural network in SMD, and prove angular update can also converge to its theoretical value at linear rate; 3) we verify our assumptions and theoretical results on various computer vision tasks including ImageNet and MSCOCO with standard settings. Experiment results show our theoretical findings agree well with empirical observations.

stat.ML↗

arXiv2020

AutoAssign: Differentiable Label Assignment for Dense Object Detection

Determining positive/negative samples for object detection is known as label assignment. Here we present an anchor-free detector named AutoAssign. It requires little human knowledge and achieves appearance-aware through a fully differentiable weighting mechanism. During training, to both satisfy the prior distribution of data and adapt to category characteristics, we present Center Weighting to adjust the category-specific prior distributions. To adapt to object appearances, Confidence Weighting is proposed to adjust the specific assign strategy of each instance. The two weighting modules are then combined to generate positive and negative weights to adjust each location's confidence. Extensive experiments on the MS COCO show that our method steadily surpasses other best sampling strategies by large margins with various backbones. Moreover, our best model achieves 52.1% AP, outperforming all existing one-stage detectors. Besides, experiments on other datasets, e.g., PASCAL VOC, Objects365, and WiderFace, demonstrate the broad applicability of AutoAssign.

cs.CV↗

arXiv2020

Deep Positional and Relational Feature Learning for Rotation-Invariant Point Cloud Analysis

In this paper we propose a rotation-invariant deep network for point clouds analysis. Point-based deep networks are commonly designed to recognize roughly aligned 3D shapes based on point coordinates, but suffer from performance drops with shape rotations. Some geometric features, e.g., distances and angles of points as inputs of network, are rotation-invariant but lose positional information of points. In this work, we propose a novel deep network for point clouds by incorporating positional information of points as inputs while yielding rotation-invariance. The network is hierarchical and relies on two modules: a positional feature embedding block and a relational feature embedding block. Both modules and the whole network are proven to be rotation-invariant when processing point clouds as input. Experiments show state-of-the-art classification and segmentation performances on benchmark datasets, and ablation studies demonstrate effectiveness of the network design.

cs.CV↗

arXiv2020

Joint COCO and Mapillary Workshop at ICCV 2019: COCO Instance Segmentation Challenge Track

In this report, we present our object detection/instance segmentation system, MegDetV2, which works in a two-pass fashion, first to detect instances then to obtain segmentation. Our baseline detector is mainly built on a new designed RPN, called RPN++. On the COCO-2019 detection/instance-segmentation test-dev dataset, our system achieves 61.0/53.1 mAP, which surpassed our 2018 winning results by 5.0/4.2 respectively. We achieve the best results in COCO Challenge 2019 and 2020.

cs.CV↗

arXiv2020

Discrete Conformal Geometry of Polyhedral Surfaces and Its Convergence

The paper proves a result on the convergence of discrete conformal maps to the Riemann mappings for Jordan domains. It is a counterpart of Rodin-Sullivan's theorem on convergence of circle packing mappings to the Riemann mapping in the new setting of discrete conformality. The proof follows the same strategy that Rodin-Sullivan used by establishing a rigidity result for regular hexagonal triangulations of the plane and estimating the quasiconformal constants associated to the discrete conformal maps.

math.GT↗

arXiv2020

Pressure Engineering of the Dirac Fermions in Quasi-One-Dimensional Tl$_2$Mo$_6$Se$_6$

Topological band dispersions other than the standard Dirac or Weyl fermions have garnered the increasing interest in materials science. Among them, the cubic Dirac fermions were recently proposed in the family of quasi-one-dimensional conductors A$_2$Mo$_6$X$_6$ (A= Na, K, In, Tl; X= S, Se, Te), where the band crossing is characterized by a linear dispersion in one $k$-space direction but the cubic dispersion in the plane perpendicular to it. It is not yet clear, however, how the external perturbations can alter these nontrivial carriers and ultimately induce a new distinct quantum phase. Here we study the evolution of Dirac fermions, in particular the cubic Dirac crossing, under external pressure in the representative quasi-one-dimensional Tl$_2$Mo$_6$Se$_6$ via the first-principles calculations. Specifically, it is found that the topological properties, including the bulk Dirac crossings and the topological surface states, change progressively under pressure up to 50 GPa where it undergoes a structural transition from the hexagonal phase to body-centered tetragonal phase. Above 50 GPa, the system is more likely to be topologically trivial. Further, we also investigate its phonon spectra, which reveals a gradual depletion of the negative phonon modes with pressure, consistent with the more three-dimensional Fermi surface in the high-pressure phase. Our work may provide a useful guideline for further experimental search and the band engineering of the topologically nontrivial fermions in this intriguing state of matter.

cond-mat.mtrl-sci↗

arXiv2020

End-to-end Interpretable Learning of Non-blind Image Deblurring

Non-blind image deblurring is typically formulated as a linear least-squares problem regularized by natural priors on the corresponding sharp picture's gradients, which can be solved, for example, using a half-quadratic splitting method with Richardson fixed-point iterations for its least-squares updates and a proximal operator for the auxiliary variable updates. We propose to precondition the Richardson solver using approximate inverse filters of the (known) blur and natural image prior kernels. Using convolutions instead of a generic linear preconditioner allows extremely efficient parameter sharing across the image, and leads to significant gains in accuracy and/or speed compared to classical FFT and conjugate-gradient methods. More importantly, the proposed architecture is easily adapted to learning both the preconditioner and the proximal operator using CNN embeddings. This yields a simple and efficient algorithm for non-blind image deblurring which is fully interpretable, can be learned end to end, and whose accuracy matches or exceeds the state of the art, quite significantly, in the non-uniform case.

cs.CV↗

arXiv2020

LabelEnc: A New Intermediate Supervision Method for Object Detection

In this paper we propose a new intermediate supervision method, named LabelEnc, to boost the training of object detection systems. The key idea is to introduce a novel label encoding function, mapping the ground-truth labels into latent embedding, acting as an auxiliary intermediate supervision to the detection backbone during training. Our approach mainly involves a two-step training procedure. First, we optimize the label encoding function via an AutoEncoder defined in the label space, approximating the "desired" intermediate representations for the target object detector. Second, taking advantage of the learned label encoding function, we introduce a new auxiliary loss attached to the detection backbones, thus benefiting the performance of the derived detector. Experiments show our method improves a variety of detection systems by around 2% on COCO dataset, no matter one-stage or two-stage frameworks. Moreover, the auxiliary structures only exist during training, i.e. it is completely cost-free in inference time. Code is available at: https://github.com/megvii-model/LabelEnc

cs.CV↗

arXiv2020

Scaling Behavior of the Quantum Phase Transition from a Quantum Anomalous Hall Insulator to an Axion Insulator

The phase transitions from one plateau to the next plateau or to an insulator in quantum Hall and quantum anomalous Hall (QAH) systems have revealed universal scaling behaviors. A magnetic-field-driven quantum phase transition from a QAH insulator to an axion insulator was recently demonstrated in magnetic topological insulator sandwich samples. Here, we show that the temperature dependence of the derivative of the longitudinal resistance on magnetic field at the transition point follows a characteristic power-law that indicates a universal scaling behavior for the QAH to axion insulator phase transition. Similar to the quantum Hall plateau to plateau transition, the QAH to axion insulator transition can also be understood by the Chalker-Coddington network model. We extract a critical exponent k~ 0.38 in agreement with recent high-precision numerical results on the correlation length exponent of the Chalker-Coddington model at v ~ 2.6, rather than the generally-accepted value of 2.33.

cond-mat.mes-hall↗

arXiv2020

Spin filtering in germanium/silicon core/shell nanowires with pseudo-helical gap

Semiconductors with strong spin-orbit interactions can exhibit a helical gap with spin-momentum locking opened by a magnetic field. Such a gap is highly spin selective as a result of a topologically protected spin-momentum locking, which can be used for spin filtering. We experimentally demonstrate such a spin filtering effect in a quasi-ballistic p-type germanium/silicon core/shell nanowire (NW), which possesses a pseudo-helical gap without the application of magnetic field. Polarized hole spin injection to the NW is achieved using cobalt ferromagnetic contacts with controlled natural surface oxide on the NW as a tunnel barrier. Local and nonlocal spin valve effects are measured as the verification of polarized spin transport in the NW outside the helical gap. We electrically tune the NW into the helical gap by scanning its chemical potential with a gate. A hysteresis loop with three resistance states is observed in the local spin valve geometry, as an evidence of spin filtering in the helical gap.

cond-mat.mes-hall↗

arXiv2020

Deep Reinforcement Learning for Dynamic Spectrum Sensing and Aggregation in Multi-Channel Wireless Networks

In this paper, the problem of dynamic spectrum sensing and aggregation is investigated in a wireless network containing N correlated channels, where these channels are occupied or vacant following an unknown joint 2-state Markov model. At each time slot, a single cognitive user with certain bandwidth requirement either stays idle or selects a segment comprising C (C < N) contiguous channels to sense. Then, the vacant channels in the selected segment will be aggregated for satisfying the user requirement. The user receives a binary feedback signal indicating whether the transmission is successful or not (i.e., ACK signal) after each transmission, and makes next decision based on the sensing channel states. Here, we aim to find a policy that can maximize the number of successful transmissions without interrupting the primary users (PUs). The problem can be considered as a partially observable Markov decision process (POMDP) due to without full observation of system environment. We implement a Deep Q-Network (DQN) to address the challenge of unknown system dynamics and computational expenses. The performance of DQN, Q-Learning, and the Improvident Policy with known system dynamics is evaluated through simulations. The simulation results show that DQN can achieve near-optimal performance among different system scenarios only based on partial observations and ACK signals.

eess.SP↗

arXiv2020

Multi-Frequency Multi-Scenario Millimeter Wave MIMO Channel Measurements and Modeling for B5G Wireless Communication Systems

Millimeter wave (mmWave) bands have been utilized for the fifth generation (5G) communication systems and will no doubt continue to be deployed for beyond 5G (B5G). However, the underlying channels are not fully investigated at multifrequency bands and in multi-scenarios by using the same channel sounder, especially for the outdoor, multiple-input multiple-output (MIMO), and vehicle-to-vehicle (V2V) conditions. In this paper, we conduct multi-frequency multi-scenario mmWave MIMO channel measurements with 4*4 antennas at 28, 32, and 39 GHz bands for three cases, i.e., the human body and vehicle blockage measurements, outdoor path loss measurements, and V2V measurements. The channel characteristics, including blockage effect, path loss and coverage range, and non-stationarity and spatial consistency, are thoroughly studied. The blockage model, path loss model, and time-varying channel model are proposed for mmWave MIMO channels. The channel measurement and modeling results will be of great importance for further mmWave communication system deployments in indoor hotspot, outdoor, and vehicular network scenarios for B5G.

eess.SP↗

arXiv2020

A Non-Stationary VVLC MIMO Channel Model for Street Corner Scenarios

In recent years, the application potential of visible light communication (VLC) technology as an alternative and supplement to radio frequency (RF) technology has attracted people's attention. The study of the underlying VLC channel is the basis for designing the VLC communication system. In this paper, a new non-stationary geometric street corner model is proposed for vehicular VLC (VVLC) multiple-input multiple-output (MIMO) channel. The proposed model takes into account changes in vehicle speed and direction. The category of scatterers includes fixed scatterers and mobile scatterers (MS). Based on the proposed model, we derive the channel impulse response (CIR) and explore the statistical characteristics of the VVLC channel. The channel gain and root mean square (RMS) delay spread of the VVLC channel are studied. In addition, the influence of velocity change on the statistical characteristics of the model is also investigated. The proposed channel model can guide future vehicle-to-infrastructure (V2I) and vehicle-to-vehicle (V2V) optical communication system design.

eess.SP↗

arXiv2020

A Survey on Complex Question Answering over Knowledge Base: Recent Advances and Challenges

Question Answering (QA) over Knowledge Base (KB) aims to automatically answer natural language questions via well-structured relation information between entities stored in knowledge bases. In order to make KBQA more applicable in actual scenarios, researchers have shifted their attention from simple questions to complex questions, which require more KB triples and constraint inference. In this paper, we introduce the recent advances in complex QA. Besides traditional methods relying on templates and rules, the research is categorized into a taxonomy that contains two main branches, namely Information Retrieval-based and Neural Semantic Parsing-based. After describing the methods of these branches, we analyze directions for future research and introduce the models proposed by the Alime team.

cs.CL↗

arXiv2020

WeightNet: Revisiting the Design Space of Weight Networks

We present a conceptually simple, flexible and effective framework for weight generating networks. Our approach is general that unifies two current distinct and extremely effective SENet and CondConv into the same framework on weight space. The method, called WeightNet, generalizes the two methods by simply adding one more grouped fully-connected layer to the attention activation layer. We use the WeightNet, composed entirely of (grouped) fully-connected layers, to directly output the convolutional weight. WeightNet is easy and memory-conserving to train, on the kernel space instead of the feature space. Because of the flexibility, our method outperforms existing approaches on both ImageNet and COCO detection tasks, achieving better Accuracy-FLOPs and Accuracy-Parameter trade-offs. The framework on the flexible weight space has the potential to further improve the performance. Code is available at https://github.com/megvii-model/WeightNet.

cs.CV↗

arXiv2020

Funnel Activation for Visual Recognition

We present a conceptually simple but effective funnel activation for image recognition tasks, called Funnel activation (FReLU), that extends ReLU and PReLU to a 2D activation by adding a negligible overhead of spatial condition. The forms of ReLU and PReLU are y = max(x, 0) and y = max(x, px), respectively, while FReLU is in the form of y = max(x,T(x)), where T(x) is the 2D spatial condition. Moreover, the spatial condition achieves a pixel-wise modeling capacity in a simple way, capturing complicated visual layouts with regular convolutions. We conduct experiments on ImageNet, COCO detection, and semantic segmentation tasks, showing great improvements and robustness of FReLU in the visual recognition tasks. Code is available at https://github.com/megvii-model/FunnelAct.

cs.CV↗