arXiv subjects

Xingang Wang

Publications and source records attributed to Xingang Wang.

At least 73 records · Page 4Linked to original sources

arXiv2020

Learning Dynamic Routing for Semantic Segmentation

Recently, numerous handcrafted and searched networks have been applied for semantic segmentation. However, previous works intend to handle inputs with various scales in pre-defined static architectures, such as FCN, U-Net, and DeepLab series. This paper studies a conceptually new method to alleviate the scale variance in semantic representation, named dynamic routing. The proposed framework generates data-dependent routes, adapting to the scale distribution of each image. To this end, a differentiable gating function, called soft conditional gate, is proposed to select scale transform paths on the fly. In addition, the computational cost can be further reduced in an end-to-end manner by giving budget constraints to the gating function. We further relax the network level routing space to support multi-path propagations and skip-connections in each forward, bringing substantial network capacity. To demonstrate the superiority of the dynamic property, we compare with several static architectures, which can be modeled as special cases in the routing space. Extensive experiments are conducted on Cityscapes and PASCAL VOC 2012 to illustrate the effectiveness of the dynamic framework. Code is available at https://github.com/yanwei-li/DynamicRouting.

cs.CV↗

arXiv2020

Long-term prediction of chaotic systems with recurrent neural networks

Reservoir computing systems, a class of recurrent neural networks, have recently been exploited for model-free, data-based prediction of the state evolution of a variety of chaotic dynamical systems. The prediction horizon demonstrated has been about half dozen Lyapunov time. Is it possible to significantly extend the prediction time beyond what has been achieved so far? We articulate a scheme incorporating time-dependent but sparse data inputs into reservoir computing and demonstrate that such rare "updates" of the actual state practically enable an arbitrarily long prediction horizon for a variety of chaotic systems. A physical understanding based on the theory of temporal synchronization is developed.

cs.LG↗

arXiv2020

Directional Deep Embedding and Appearance Learning for Fast Video Object Segmentation

Most recent semi-supervised video object segmentation (VOS) methods rely on fine-tuning deep convolutional neural networks online using the given mask of the first frame or predicted masks of subsequent frames. However, the online fine-tuning process is usually time-consuming, limiting the practical use of such methods. We propose a directional deep embedding and appearance learning (DDEAL) method, which is free of the online fine-tuning process, for fast VOS. First, a global directional matching module, which can be efficiently implemented by parallel convolutional operations, is proposed to learn a semantic pixel-wise embedding as an internal guidance. Second, an effective directional appearance model based statistics is proposed to represent the target and background on a spherical embedding space for VOS. Equipped with the global directional matching module and the directional appearance model learning module, DDEAL learns static cues from the labeled first frame and dynamically updates cues of the subsequent frames for object segmentation. Our method exhibits state-of-the-art VOS performance without using online fine-tuning. Specifically, it achieves a J & F mean score of 74.8% on DAVIS 2017 dataset and an overall score G of 71.3% on the large-scale YouTube-VOS dataset, while retaining a speed of 25 fps with a single NVIDIA TITAN Xp GPU. Furthermore, our faster version runs 31 fps with only a little accuracy loss. Our code and trained networks are available at https://github.com/YingjieYin/Directional-Deep-Embedding-and-Appearance-Learning-for-Fast-Video-Object-Segmentation.

cs.CV↗

arXiv2019

Multiple receptive fields and small-object-focusing weakly-supervised segmentation network for fast object detection

Object detection plays an important role in various visual applications. However, the precision and speed of detector are usually contradictory. One main reason for fast detectors' precision reduction is that small objects are hard to be detected. To address this problem, we propose a multiple receptive field and small-object-focusing weakly-supervised segmentation network (MRFSWSnet) to achieve fast object detection. In MRFSWSnet, multiple receptive fields block (MRF) is used to pay attention to the object and its adjacent background's different spatial location with different weights to enhance the feature's discriminability. In addition, in order to improve the accuracy of small object detection, a small-object-focusing weakly-supervised segmentation module which only focuses on small object instead of all objects is integrated into the detection network for auxiliary training to improve the precision of small object detection. Extensive experiments show the effectiveness of our method on both PASCAL VOC and MS COCO detection datasets. In particular, with a lower resolution version of 300x300, MRFSWSnet achieves 80.9% mAP on VOC2007 test with an inference speed of 15 milliseconds per frame, which is the state-of-the-art detector among real-time detectors.

cs.CV↗

arXiv2019

Attention-guided Unified Network for Panoptic Segmentation

This paper studies panoptic segmentation, a recently proposed task which segments foreground (FG) objects at the instance level as well as background (BG) contents at the semantic level. Existing methods mostly dealt with these two problems separately, but in this paper, we reveal the underlying relationship between them, in particular, FG objects provide complementary cues to assist BG understanding. Our approach, named the Attention-guided Unified Network (AUNet), is a unified framework with two branches for FG and BG segmentation simultaneously. Two sources of attentions are added to the BG branch, namely, RPN and FG segmentation mask to provide object-level and pixel-level attentions, respectively. Our approach is generalized to different backbones with consistent accuracy gain in both FG and BG segmentation, and also sets new state-of-the-arts both in the MS-COCO (46.5% PQ) and Cityscapes (59.0% PQ) benchmarks.

cs.CV↗

arXiv2019

Multi-loss-aware Channel Pruning of Deep Networks

Channel pruning, which seeks to reduce the model size by removing redundant channels, is a popular solution for deep networks compression. Existing channel pruning methods usually conduct layer-wise channel selection by directly minimizing the reconstruction error of feature maps between the baseline model and the pruned one. However, they ignore the feature and semantic distributions within feature maps and real contribution of channels to the overall performance. In this paper, we propose a new channel pruning method by explicitly using both intermediate outputs of the baseline model and the classification loss of the pruned model to supervise layer-wise channel selection. Particularly, we introduce an additional loss to encode the differences in the feature and semantic distributions within feature maps between the baseline model and the pruned one. By considering the reconstruction error, the additional loss and the classification loss at the same time, our approach can significantly improve the performance of the pruned model. Comprehensive experiments on benchmark datasets demonstrate the effectiveness of the proposed method.

cs.CV↗

arXiv2019

Cluster Regularized Quantization for Deep Networks Compression

Deep neural networks (DNNs) have achieved great success in a wide range of computer vision areas, but the applications to mobile devices is limited due to their high storage and computational cost. Much efforts have been devoted to compress DNNs. In this paper, we propose a simple yet effective method for deep networks compression, named Cluster Regularized Quantization (CRQ), which can reduce the presentation precision of a full-precision model to ternary values without significant accuracy drop. In particular, the proposed method aims at reducing the quantization error by introducing a cluster regularization term, which is imposed on the full-precision weights to enable them naturally concentrate around the target values. Through explicitly regularizing the weights during the re-training stage, the full-precision model can achieve the smooth transition to the low-bit one. Comprehensive experiments on benchmark datasets demonstrate the effectiveness of the proposed method.

cs.CV↗

arXiv2019

Agent-based simulations of China inbound tourism network

Based on the results of a large-scale survey, we construct an agent-based network model for the independent inbound tourism of China and, by the approach of numerical simulation, investigate the dynamical responses of the tourist flows to external perturbations in different scenarios, including the closure of a tourist city, the opening of a new port in western China, and the increase of the tourism attractiveness of a specific city. Numerical results show that: (1) the closure of a single city in general will affect the tourist visitations of many other cities in the network and, comparing to the non-port cities, the overall visitation volume of the system is more influenced by closing a port city; (2) the opening of a new port city in western China will attract more tourists to the western cities, but has a negligible impact on either the overall visitation volume or the imbalanced tourist distribution; and (3) the increase of the tourism attractiveness of a non-port (port) city normally increases (decreases) the overall visitation volume, yet there are exceptions due to the spillover effect. Furthermore, by increasing the tourism attractiveness of a few cities simultaneously, we investigate also the strategy of multiple-city-upgrade in tourism development. Numerical results show that the overall tourist volume is better improved by upgrading important non-port cities that are geographically distant from each other. The study reveals the rich dynamics inherent in complex tourism network, and the findings could be helpful to the development and management of China inbound tourism.

physics.soc-ph↗

arXiv2018

Identity-Enhanced Network for Facial Expression Recognition

Facial expression recognition is a challenging task, arguably because of large intra-class variations and high inter-class similarities. The core drawback of the existing approaches is the lack of ability to discriminate the changes in appearance caused by emotions and identities. In this paper, we present a novel identity-enhanced network (IDEnNet) to eliminate the negative impact of identity factor and focus on recognizing facial expressions. Spatial fusion combined with self-constrained multi-task learning are adopted to jointly learn the expression representations and identity-related information. We evaluate our approach on three popular datasets, namely Oulu-CASIA, CK+ and MMI. IDEnNet improves the baseline consistently, and achieves the best or comparable state-of-the-art on all three datasets.

cs.CV↗

arXiv2018

Adversarial Feature Sampling Learning for Efficient Visual Tracking

The tracking-by-detection framework usually consist of two stages: drawing samples around the target object in the first stage and classifying each sample as the target object or background in the second stage. Current popular trackers based on tracking-by-detection framework typically draw samples in the raw image as the inputs of deep convolution networks in the first stage, which usually results in high computational burden and low running speed. In this paper, we propose a new visual tracking method using sampling deep convolutional features to address this problem. Only one cropped image around the target object is input into the designed deep convolution network and the samples is sampled on the feature maps of the network by spatial bilinear resampling. In addition, a generative adversarial network is integrated into our network framework to augment positive samples and improve the tracking performance. Extensive experiments on benchmark datasets demonstrate that the proposed method achieves a comparable performance to state-of-the-art trackers and accelerates tracking-by-detection trackers based on raw-image samples effectively.

cs.CV↗

arXiv2018

Enhancing network synchronization by phase modulation

Due to time delays in signal transmission and processing, phase lags are inevitable in realistic complex oscillator networks. Conventional wisdom is that phase lags are detrimental to network synchronization. Here we show that judiciously chosen phase lag modulations can result in significantly enhanced network synchronization. We justify our strategy of phase modulation, demonstrate its power in facilitating and enhancing network synchronization with synthetic and empirical network models, and provide an analytic understanding of the underlying mechanism. Our work provides a new approach to synchronization optimization in complex networks, with insights into control of complex nonlinear networks.

nlin.CD↗

arXiv2018

A novel channel pruning method for deep neural network compression

In recent years, deep neural networks have achieved great success in the field of computer vision. However, it is still a big challenge to deploy these deep models on resource-constrained embedded devices such as mobile robots, smart phones and so on. Therefore, network compression for such platforms is a reasonable solution to reduce memory consumption and computation complexity. In this paper, a novel channel pruning method based on genetic algorithm is proposed to compress very deep Convolution Neural Networks (CNNs). Firstly, a pre-trained CNN model is pruned layer by layer according to the sensitivity of each layer. After that, the pruned model is fine-tuned based on knowledge distillation framework. These two improvements significantly decrease the model redundancy with less accuracy drop. Channel selection is a combinatorial optimization problem that has exponential solution space. In order to accelerate the selection process, the proposed method formulates it as a search problem, which can be solved efficiently by genetic algorithm. Meanwhile, a two-step approximation fitness function is designed to further improve the efficiency of genetic process. The proposed method has been verified on three benchmark datasets with two popular CNN models: VGGNet and ResNet. On the CIFAR-100 and ImageNet datasets, our approach outperforms several state-of-the-art methods. On the CIFAR-10 and SVHN datasets, the pruned VGGNet achieves better performance than the original model with 8 times parameters compression and 3 times FLOPs reduction.

cs.CV↗

arXiv2018

Network synchronization with periodic coupling

The synchronization behavior of networked chaotic oscillators with periodic coupling is investigated. It is observed in simulations that the network synchronizability could be significantly influenced by tuning the coupling frequency, even making the network alternated between the synchronous and non-synchronous states. By the method of master stability function, we conduct a detailed analysis on the influence of coupling frequency on network synchronizability, and find that the network synchronizability is maximized at some characteristic frequencies comparable to the intrinsic frequency of the local dynamics. Moremore, it is found that as the amplitude of the coupling increases, the characteristic frequencies are gradually decreased. By the technique of finite-time Lyapunov exponent, we investigate further the mechanism for the maximized synchronizability, and find that at the characteristic frequencies the power spectrum of the finite-time Lyapunov exponent is abruptly changed from the localized to broad distributions. When this feature is absent or not prominent, the network synchronizability is less influenced by the periodic coupling. Our study shows the efficiency of finite-time Lyapunov exponent in exploring the synchronization behavior of temporally coupled oscillators, and sheds lights on the interplay between the system dynamics and structure in the general temporal networks.

nlin.CD↗

arXiv2018

Huygens' synchronization experiment revisited: Luck or skill?

353 years ago, in a letter to the Royal Society of London, Christiaan Huygens described "an odd kind of sympathy" between two pendulums mounted side by side on a wooden beam, which inspired the modern studies of synchronization in coupled nonlinear oscillators. Despite the blooming of synchronization study in a variety of disciplines, the original phenomenon described by Huygens remains a puzzle to researchers. Here, by placing two mechanical metronomes on top of a freely moving plastic board, we revisit the synchronization experiment conducted by Huygens. Experimental results show that by introducing a small mismatch to the natural frequencies of the metronomes, the probability for generating the anti-phase synchronization (APS) state, i.e., the "odd sympathy" described by Huygens, can be clearly increased. By numerical simulations of the system dynamics, we conduct a detailed analysis on the influence of frequency mismatch on APS. It is found that as the frequency mismatch increases from $0$, the attracting basin of APS is gradually enlarged and, in the meantime, the basin of in-phase synchronization (IPS) is reduced. However, as the frequency mismatch exceeds some critical value, both the basins of APS and IPS are suddenly disappeared, resulting in the desynchronization states. The impacts of friction coefficient and synchronization precision on APS are also studied, and it is found that with the increases of the friction coefficient and the precision requirement of APS, the critical frequency mismatch for desynchronization will be decreased. Our study indicates that, instead of luck, Huygens might have introduced, deliberately and elaborately, a small frequency mismatch to the pendulums in his experiment for generating the "odd sympathy".

nlin.CD↗

arXiv2018

Cluster synchronization in complex network of coupled chaotic circuits: an experimental study

By a small-size complex network of coupled chaotic Hindmarsh-Rose circuits, we study experimentally the stability of network synchronization to the removal of shortcut links. It is shown that the removal of a single shortcut link may destroy either completely or partially the network synchronization. Interestingly, when the network is partially desynchronized, it is found that the oscillators can be organized into different groups, with oscillators within each group being highly synchronized but are not for oscillators from different groups, showing the intriguing phenomenon of cluster synchronization. The experimental results are analyzed by the method of eigenvalue analysis, which implies that the formation of cluster synchronization is crucially dependent on the network symmetries. Our study demonstrates the observability of cluster synchronization in realistic systems, and indicates the feasibility of controlling network synchronization by adjusting network topology.

nlin.CD↗

arXiv2017

Pattern formation and transition in complex networks

Dynamical patterns in complex networks of coupled oscillators are both of theoretical and practical interest, yet to fully reveal and understand the interplay between pattern emergence and network structure remains to be an outstanding problem. A fundamental issue is the effect of network structure on the stability of the patterns. We address this issue by using the setting where random links are systematically added to a regular lattice and focusing on the dynamical evolution of spiral wave patterns. As the network structure deviates more from the regular topology (so that it becomes increasingly more complex), the original stable spiral wave pattern can disappear and a different type of pattern can emerge. Our main findings are the following. (1) Short-distance links added to a small region containing the spiral tip can have a more significant effect on the wave pattern than long-distance connections. (2) As more random links are introduced into the network, distinct pattern transitions can occur, which include the transition of spiral wave to global synchronization, to a chimera-like state, and then to a pinned spiral wave. (3) Around the transitions the network dynamics is highly sensitive to small variations in the network structure in the sense that the addition of even a single link can change the pattern from one type to another. These findings provide insights into the pattern dynamics in complex networks, a problem that is relevant to many physical, chemical, and biological systems.

nlin.CD↗

arXiv2017

Synchronization of coupled metronomes on two layers

Coupled metronomes serve as a paradigmatic model for exploring the collective behaviors of complex dynamical systems, as well as a classical setup for classroom demonstrations of synchronization phenomena. Whereas previous studies of metronome synchronization have been concentrating on symmetric coupling schemes, here we consider the asymmetric case by adopting the scheme of layered metronomes. Specifically, we place two metronomes on each layer, and couple two layers by placing one on top of the other. By varying the initial conditions of the metronomes and adjusting the friction between the two layers, a variety of synchronous patterns are observed in experiment, including the splay synchronization (SS) state, the generalized splay synchronization (GSS) state , the anti-phase synchronization (APS) state, the in-phase delay synchronization (IPDS) state, and the in-phase synchronization (IPS) state. In particular, the IPDS state, in which the metronomes on each layer are synchronized in phase but are of a constant phase delay to metronomes on the other layer, is observed for the first time. In addition, a new technique based on audio signals is proposed for pattern detection, which is more convenient and easier to apply than the existing acquisition techniques. Furthermore, a theoretical model is developed to explain the experimental observations, and is employed to explore the dynamical properties of the patterns, including the basin distributions and the pattern transitions. Our study sheds new lights on the collective behaviors of coupled metronomes, and the developed setup can be used in the classroom for demonstration purposes.

nlin.CD↗

arXiv2016

Evoking complex neuronal networks by stimulating a single neuron

The dynamical responses of complex neuronal networks to external stimulus injected on a \emph{single} neuron are investigated. Stimulating the largest-degree neuron in the network, it is found that as the intensity of the stimulus increases, the network will be transiting from the resting to firing states and then restoring to the resting state, showing a bounded firing region in the parameter space. Furthermore, it is found that as the coupling strength decreases, the firing region is gradually expanded and, at the weak couplings, separated into disconnected subregions. By a simplified network model, we conduct a detail analysis on the bifurcation diagram of the network dynamics in the two-dimensional parameter space spanned by stimulating intensity and coupling strength, and, by introducing a new coefficient named effective stimulus, explore the mechanisms of the modified firing region. It is revealed that the coupling strength and stimulating intensity are equally important in evoking the network, but with different mechanisms. Specifically, the effective stimuli are \emph{shifted up} globally with the increase of the stimulating intensity, while are \emph{drawn closer} with the increase of the coupling strength. The dynamical responses of small-world and random complex networks to external stimulus injected on the largest-degree neuron are also investigated, which confirm the generality of the observed phenomena.

nlin.CD↗