Search

This paper develops a new frame for non-stationary signal separation, which is a combination of wavelet transform, clustering strategy and local maximum approximation. We provide a rigorous mathematical theoretical analysis and prove that the proposed algorithm can estimate instantaneous frequencies and sub-signal modes from a blind source signal. The error bounds for instantaneous frequency estimation and sub-signal recovery are provided. Numerical experiments on synthetic and real data demonstrate the effectiveness and efficiency of the proposed algorithm. Our method based on wavelet transform can be extended to other time–frequency transforms, which provides a new perspective of time–frequency analysis tools in solving the non-stationary signal separation problem. [ABSTRACT FROM AUTHOR]

In this article, we consider the mixed ratcheting-periodic dividend strategies for spectrally negative Lévy risk model, in which dividend payments can both be made continuously without falling and discretely at the jump times of an independent Poisson process. The expected net present value of dividends paid up to ruin and the Laplace transform of the ruin time are obtained by using Lévy fluctuation theory. All the results are expressed in terms of scale functions. Finally, numerical results for Brownian motion with drift are given. [ABSTRACT FROM AUTHOR]

In this article, we consider the extension of Shannon sampling series reconstruction theorem for nonhomogeneous random fields using local averages sampling, which helps improve certain earlier results. The upper bound of mean square truncation sampling approximation error is more precise, and we establish one approximation result in the almost sure sense. [ABSTRACT FROM AUTHOR]

Since magnetic resonance imaging (MRI) has superior soft tissue contrast, contouring (brain) tumor accurately by MRI images is essential in medical image processing. Segmenting tumor accurately is immensely challenging, since tumor and normal tissues are often inextricably intertwined in the brain. It is also extremely time consuming manually. Late deep learning techniques start to show reasonable success in brain tumor segmentation automatically. The purpose of this study is to develop a new region-of-interest-aided (ROI-aided) deep learning technique for automatic brain tumor MRI segmentation. The method consists of two major steps. Step one is to use a 2D network with U-Net architecture to localize the tumor ROI, which is to reduce the impact of normal tissue's disturbance. Then a 3D U-Net is performed in step 2 for tumor segmentation within identified ROI. The proposed method is validated on MICCAI BraTS 2015 Challenge with 220 high Gliomas grade (HGG) and 54 low Gliomas grade (LGG) patients' data. The Dice similarity coefficient and the Hausdorff distance between the manual tumor contour and that segmented by the proposed method are 0.876 ±0.068 and 3.594±1.347 mm, respectively. These numbers are indications that our proposed method is an effective ROI-aided deep learning strategy for brain MRI tumor segmentation, and a valid and useful tool in medical image processing. [ABSTRACT FROM AUTHOR]

In this paper, we propose an ambiguousness-aware state evolution (AASE) method which represents the uncertainty of the input sequence and evolves the subsequent skeletons to generate a reasonable full-length sequence for action prediction. Unlike most existing methods that enforce partial sequences with the labels of full-length videos and ignore the semantic information of the subsequent action, we develop an evolution method by predicting the instructional actions and generating the reasonable candidate subsequent actions, so that the ambiguity of the full sequence’s label supervising for the partial actions can be effectively alleviated. Our method generates the rational subsequent actions under the instructional action class to complement the partially observed action sequence. We design two criteria for a rational generation: 1) the instruction of subsequent action keeps the semantic consistency with the observed sequence; 2) the generation sequence is satisfied with the distribution of the sequence of real data. Moreover, we design an uncertainty module to decide the instructional action class for the generation network. AASE predicts instructional actions with uncertainty learning and evolves different instructional actions by generating the subsequent skeletons, which find the most probable action to represent the partially observed action by learning the way of perceiving the tendency of the ongoing action. We conduct experiments on seven widely used action datasets: NTU-60, NTU-120, UCF101, UT-Interaction, BIT, PKU-MMD and HMDB51, and our experimental results clearly demonstrate that our method achieves very competitive performance with state-of-the-art. [ABSTRACT FROM AUTHOR]

In this paper, we present a hybrid semantic affinity learning method (HSA) to capture and leverage the dependencies of categories for 3D semantic segmentation. Unlike existing methods that only use the cross-entropy loss to perform one-to-one supervision and ignore the semantic relations between points, our approach aims to learn the label dependencies between 3D points from a hybrid perspective. From a global view, we introduce the structural correlations among different classes to provide global priors for point features. Specifically, we fuse word embeddings of labels and scene-level features as category nodes, which are processed via a graph convolutional network (GCN) to produce the sample-adapted global priors. These priors are then combined with point features to enhance the rationality of semantic predictions. From a local view, we propose the concept of local affinity to effectively model the intra-class and inter-class semantic similarities for adjacent neighborhoods, making the predictions more discriminative. Experimental results show that our method consistently improves the performance of state-of-the-art models across indoor (S3DIS, ScanNet), outdoor (SemanticKITTI), and synthetic (ShapeNet) datasets. [ABSTRACT FROM AUTHOR]

In this paper, we mainly investigate the average sampling problem for the homogeneous random fields in a reproducing kernel subspace of mixed Lebesgue space. Based on the counterpart sampling result for the deterministic signals in the same space, a mean square convergence result for recovering the homogeneous random fields by the iterative reconstruction algorithm is obtained. [ABSTRACT FROM AUTHOR]

In this paper, a lower bounded of sequences on classical sampling theorem for random signals is given and is used to extend the applicable range of frame. The convergence property of sampling series, the estimate of truncation error both in mean square sense and for almost sure results on sampling theorem for multidimensional random signals from asymmetrical average sampling are analyzed. Using the new results on frames recently, the famous Shannon sampling theorem on multidimensional random signals has been extended with a new idea. [ABSTRACT FROM AUTHOR]

Human parsing, especially in the wild, has attracted a lot of attention due to its great potential in many real-world applications. The Pyramid Spatial Parsing (PSP) module has shown superior performances in scene and human parsing tasks. However, the basic AvgPool operation in PSP equally aggregates spatial clues of a local region, and thus mixes up influences of different human parts presented in this region. It results in failures in capturing useful contexts relevant to parsing different parts. To address this problem, a suitable mechanism to collect spatial clues aligning with different human parts is proposed in this paper. We employ a Gather-Excite (GE) operation, a replacement of the AvgPool-Upsample operation in a pyramidical structure, to accurately reflect relevant human parts of various scales. The GE operation contains two steps: the gather operation that adaptively aggregates spatial clues to relevant human parts, and the excite operation that generates new feature maps with the gathered contextual information. This results in a novel Pyramidical Gather-Excite Context (PGEC) module to solve the multi-scale problem and parse person at various scales. The PGEC module is composed of multiple GE operations with different spatial extents and aggregates local and global spatial clues for better modeling multi-scale contextual information in parallel. Moreover, we integrate the PGEC module with fine-grained details, edge preserving module and deep supervision to formulate a novel PGEC Network (PGECNet) for human parsing. The proposed PGECNet has achieved state-of-the-art performance on four single-person human parsing datasets (i.e., LIP, PPSS, ATR and Fashion Clothing) and two multi-person human parsing datasets (i.e., PASCAL-Person-Part and CIHP). The experimental results show that the proposed PGEC is superior to the PSP and ASPP modules especially in single-human parsing task. The source code is publicly available at https://github.com/31sy/PGECNet. [ABSTRACT FROM AUTHOR]

Robust principal component analysis (RPCA) has recently seen ubiquitous activity for dimensionality reduction in image processing, visualization and pattern recognition. Conventional RPCA methods model the low-rank component as regularizing each singular value equally. However, in numerous modern applications, each singular value has different physical meaning and should be treated differently. This is one of the main reasons why RPCA techniques cannot work well in dealing with many realistic problems. To solve this problem, a novel hierarchical Bayesian RPCA model with adaptive singular value penalty is proposed. This model enforces the low-rank constraint by introducing an adaptive penalty function on the singular values of the low-rank component. In particular, we impose a hierarchical Exponent-Gamma prior on the singular values of the low-rank component and the Beta-Bernoulli prior on sparsity indicators. The variational Bayesian framework and the Markov chain Monte Carlo-based Bayesian inference are considered for inferring the posteriors of all latent variables involved in low-rank and sparse components. Numerical experiments demonstrate the competitive performance of the proposed model on synthetic and real data. [ABSTRACT FROM AUTHOR]

Histogram maximum likelihood estimators of semi-parametric space–time self-exciting point process models via expectation–maximization algorithm can be biased when the background process is inhomogeneous. We explore an alternative estimation method based on the variable bandwidth kernel density estimation (KDE) and EM algorithm. The proposed estimation method involves expanding the semi-parametric models by incorporating an inhomogeneous background process in space and time and applying the variable bandwidth KDE to estimate the background intensity function. Using an example, we show how the variable bandwidth KDE can be estimated this way. Two simulation examples based on residual analysis are designed to evaluate and validate the ability of our methods to recover the background intensity function and parametric triggering intensity function. [ABSTRACT FROM AUTHOR]

Some works in modified compressive sensing (CS) show that reconstruction of sparse signals can obtain better results than traditional CS using the partially known support. In this paper, we extend the idea of these works to the null space tuning algorithm with hard thresholding, feedbacks (NST + HT + FB) and derive sufficient conditions for robust sparse signal recovery. The theoretical analysis shows that including prior information of partially known support relaxes the preconditioned restricted isometry property condition comparing with the NST + HT + FB. Numerical experiments demonstrate that the modification improves the performance of the NST+HT+FB, thereby requiring fewer samples to obtain an approximate reconstruction. Meanwhile, a systemic comparison with different methods based on partially known support is shown. [ABSTRACT FROM AUTHOR]

Clothing attribute recognition, especially in unconstrained street images, is a challenging task for multimedia. Existing methods for multi-task clothing attribute prediction often ignore the relation between specific attributes and positions. However, the attribute response is always location-sensitive, i.e., different spatial locations have various contributions to attributes. Inspired by the locality of clothing attributes, in this paper, we introduce the attention mechanism to incorporate the impact of positions for clothing attribute prediction with only image-level annotations. However, the performance improvement is limited if we directly use the traditional spatial attention model for each task since it does not take the influence from other tasks into account. Instead, we propose a novel task-aware attention mechanism, which estimates the importance of each position across different tasks. We first evaluate a task attention network with an end-to-end multi-task clothing attribute learning architecture on the shop domain. And then, we employ curriculum learning strategy, which transfers the well-trained shop domain attribute knowledge to the street domain attribute prediction. Experiments are conducted on three clothing benchmarks, i.e., cross-domain clothing attribute dataset, woman clothing dataset, and man clothing dataset. The performance of attribute prediction demonstrates the superiority of the proposed task-aware attention mechanism over several state-of-the-art methods both in shop and street domains. [ABSTRACT FROM AUTHOR]

In the real-world face recognition problems, the collected query set images often suffer serious disturbances. To address the problem, we propose an image set face recognition method based on extended low rank recovery and collaborative representation. By exploiting a Frobenius norm term, an extended low rank representation model is firstly developed to remove all possible disturbances from the query set and reconstruct the rank-one query set. To improve the computational efficiency, a compact and discriminative dictionary is learned from the large gallery set, and the closed form solutions for both the dictionary atom and the coding coefficient are straightway derived. The final classification is performed by using any frame in the reconstructed query set instead of using the whole set, which can further improve the running efficiency. Extensive experiments are conducted on the benchmark Honda/USCD and Youtube Celebrities database to verify that the proposed method outperforms significantly the state-of-the-art methods in terms of robustness and efficiency. [ABSTRACT FROM AUTHOR]

Quantitative understanding of human activity is very important as many social and economic trends are driven by human actions. We propose a novel stochastic process, the Multi-state Markov Cascading Non-homogeneous Poisson Process (M2CNPP), to analyze human e-mail communication involving both periodic patterns and bursts phenomenon. The model parameters are estimated using the Generalized Expectation Maximization (GEM) algorithm while the hidden states are treated as missing values. The empirical results demonstrate that the proposed model adequately captures the major temporal cascading features as well as the periodic patterns in e-mail communication. [ABSTRACT FROM AUTHOR]

This paper presents an effective method for marine sewage detection from a remote-sensing image. It is inspired by the Grab-Cut mechanism that iterative estimation and incomplete labeling allow a considerably reduced degree of user interaction for a given quality of result. By establishing the relationship between the color feature and the object seeds, we first model object and background with Gaussian mixture model, respectively, followed by iteratively updating the parameter of model to decline the energy function. To improve the computation efficiency, we propose to extend the region of interest as background. The proposed method accounts for not only the effect of color feature, but also the geographical information. The experimental results demonstrate that the proposed method is more reliable in marine sewage detection compared to other state-of-the-art methods. [ABSTRACT FROM AUTHOR]

The concept of structured sparse coding noise is introduced to exploit the spatial correlations and nonlocal constraint of the local structure. Then the model of nonlocally centralized simultaneous sparse coding(NC-SSC) is proposed for reconstructing the original image, and an algorithm is proposed to transform the simultaneous sparse coding into reweighted low-rank approximation. Experimental results on image denoisng, deblurring and super-resolution demonstrate the advantage of the proposed NC-SSC method over the state-of-the-art image restoration methods. [ABSTRACT FROM AUTHOR]

Salient object detection is very useful in many computer vision applications such as image segmentation, content-based image editing and object recognition. In this paper, we present a salient object detection algorithm by using color spatial distribution (CSD) and minimum spanning tree weight (MSTW). We first use a segmentation algorithm to decompose an image into superpixel-level elements, then use these elements as nodes to construct a minimum spanning tree (MST), each connected edge weight is the mean color difference between two nodes. CSD of each element can be computed by integrating color, spatial distance and MSTW. Note that if the color of one element is the most widely distributed over the entire image, it should have the biggest CSD value, we regard this element as a background node (BG Node). Then we use the MSTW between other element and BG node to generate a MSTW map. The superpixel-level saliency map can be obtained by combining the CSD map and MSTW map. Finally, we use a guided filter to get the pixel-level saliency map. Experimental results on two databases demonstrate that our proposed method outperforms other previous state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

In this paper, a discriminative structured dictionary learning algorithm is presented. To enhance the dictionary's discriminative power, the reconstruction error, classification error and inhomogeneous representation error are integrated into the objective function. The proposed approach learns a single structured dictionary and a linear classifier jointly. The learned dictionary encourages the samples from the same class to have similar sparse codes, and the samples from different classes to have dissimilar sparse codes. The solution to the objective function is achieved by employing a feature-sign search algorithm and Lagrange dual method. Experimental results on three public databases demonstrate that the proposed approach outperforms several recently proposed dictionary learning techniques for classification. [ABSTRACT FROM AUTHOR]

The research on the robust principal component analysis has been attracting much attention recently. Generally, the model assumes sparse noise and characterizes the error term by the λ-norm. However, the sparse noise has clustering effect in practice so using a certain λ-norm simply is not appropriate for modeling. In this paper, we propose a novel method based on sparse Bayesian learning principles and Markov random fields. The method is proved to be very effective for low-rank matrix recovery and contiguous outliers detection, by enforcing the low-rank constraint in a matrix factorization formulation and incorporating the contiguity prior as a sparsity constraint. The experiments on both synthetic data and some practical computer vision applications show that the novel method proposed in this paper is competitive when compared with other state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Product variation reduction is critical to improve process efficiency and product quality, especially for multistage machining process (MMP). However, due to the variation accumulation and propagation, it becomes quite difficult to predict and reduce product variation for MMP. While the method of statistical process control can be used to control product quality, it is used mainly to monitor the process change rather than to analyze the cause of product variation. In this paper, based on a differential description of the contact kinematics of locators and part surfaces, and the geometric constraints equation defined by the locating scheme, an improved analytical variation propagation model for MMP is presented. In which the influence of both locator position and machining error on part quality is considered while, in traditional model, it usually focuses on datum error and fixture error. Coordinate transformation theory is used to reflect the generation and transmission laws of error in the establishment of the model. The concept of deviation matrix is heavily applied to establish an explicit mapping between the geometric deviation of part and the process error sources. In each machining stage, the part deviation is formulized as three separated components corresponding to three different kinds of error sources, which can be further applied to fault identification and design optimization for complicated machining process. An example part for MMP is given out to validate the effectiveness of the methodology. The experiment results show that the model prediction and the actual measurement match well. This paper provides a method to predict part deviation under the influence of fixture error, datum error and machining error, and it enriches the way of quality prediction for MMP. [ABSTRACT FROM AUTHOR]

The classical Shannon sampling theorem is suitable for reconstructing a band-limited signal from its sampled values taken at regular instances with equal step by using the well-known sinc function. However, due to the inertia of the measurement apparatus, it is impossible to measure the value of a signal precisely at such discrete time. In practice, only unsymmetrically local averages of signal near the regular instances can be measured and used as the inputs for a signal reconstruction method. In addition, when implemented in hardware, the traditional sinc function cannot be directly used for signal reconstruction. We propose using the Taylor expansion of sinc function to reconstruct signal sampled from unsymmetrically local averages and give the upper bound of the reconstruction error (i.e., truncation error). The convergency of the reconstruction method is also presented. [ABSTRACT FROM PUBLISHER]

We present a technique to recover and refine the depth map from a single image captured by a conventional camera in this paper. Our method builds on the universal imaging principle: only scene at the focus distance will converge to a single sharp point on imaging sensor but other scene will yield different blur effects varying with its distance from the camera lens. We first estimate depth values at edge locations via spectrum contrast and then recover the full depth map using a depth matting optimization method. Due to the fact that some blur textures such as soft shadows or blur patterns will produce ambiguity results during the procedure of depth estimation, we use a total variation-based image smoothing method to smooth the original image, a smoothed image with detailed texture being suppressed can be generated. Taking this smoothed image as reference image, a guided filter is used to refine the final depth map. [ABSTRACT FROM PUBLISHER]

We present a technique to recover and refine the depth map from a single image captured by a conventional camera in this paper. Our method builds on the universal imaging principle: only scene at the focus distance will converge to a single sharp point on imaging sensor but other scene will yield different blur effects varying with its distance from the camera lens. We first estimate depth values at edge locations via spectrum contrast and then recover the full depth map using a depth matting optimization method. Due to the fact that some blur textures such as soft shadows or blur patterns will produce ambiguity results during the procedure of depth estimation, we use a total variation-based image smoothing method to smooth the original image, a smoothed image with detailed texture being suppressed can be generated. Taking this smoothed image as reference image, a guided filter is used to refine the final depth map. [ABSTRACT FROM AUTHOR]

The lown-rank tensor recovery problem is an interesting extension of thecompressed sensing. This problem consists of finding a tensor of minimumn-rank subject to linear equality constraints and has been proposed in many areas such as data mining, machine learning and computer vision. In this paper, operator splitting technique and convex relaxation technique are adapted to transform the lown-rank tensor recovery problem into a convex, unconstrained optimization problem, in which the objective function is the sum of a convex smooth function with Lipschitz continuous gradient and a convex function on a set of matrices. Furthermore, in order to solve the unconstrained nonsmooth convex optimization problem, an accelerated proximal gradient algorithm is proposed. Then, some computational techniques are used to improve the algorithm. At the end of this paper, some preliminary numerical results demonstrate the potential value and application of the tensor as well as the efficiency of the proposed algorithm. [ABSTRACT FROM PUBLISHER]

An efficient adaptive approximation demosaicking algorithm based on the sampled edge pattern was presented for mosaic images from Bayer color filter array. The proposed algorithm determined edge patterns by four nearest green values surrounding the green interpolation location. Then according to the edge patterns, different adaptive interpolation steps were applied. Simulations on 12 Kodak photos and 15 IMAX high-quality images showed that the proposed method outperformed the other four demosaicking methods (bilinear, effective color interpolation, Lu's method and Chen's method) for average color peak signal to noise ratios and maintained a relatively low complexity owing to constant color-difference interpolation step and a reasonable terminating condition of iteration. [ABSTRACT FROM AUTHOR]

The Nyquist–Shannon sampling theorem is on the reconstruction of a band-limited signal from its uniformly sampled samples. The higher the signal bandwidth gets, the more challenging the uniform sampling may become. To deal with this problem, signal reconstruction from local averages has been studied in the literature. In this paper, we obtain an improved Nyquist–Shannon sampling theorem from general local averages. In practice, the measurement apparatus gives a weighted average over an asymmetrical interval. As a special case, for local averages from symmetrical interval, we show that the sampling rate is much lower than that of a result by Gröchenig. Moreover, we obtain two exact dual frames from local averages, one of which improves a result by Sun and Zhou. At the end of this paper, as an example application of local average sampling, we consider a reconstruction algorithm: the piecewise linear approximations. [ABSTRACT FROM AUTHOR]

The WKS sampling theorem is a fundamental result in the field of telecommunication and signal processing. It is a perfect example of the synthesis of traditionally distinct disciplines in mathematics, engineering analysis, and the sciences. The theorem recovers a bandlimited signal from its samples at uniformly points, which purpose is to analyze deterministic functions, or deterministic signals. However, there are a few counterparts of this theorem for random signals so far. This article gives a new WKS sampling principle on random signals from local averages with probability 1. [ABSTRACT FROM AUTHOR]

In this paper, the explicit estimates of central moments for one kind of exponential-type operators are derived. The estimates play an essential role in studying the explicit approximation properties of this family of operators. Using the proposed method, the results of Ditzian and Totik in 1987, Guo and Qi in 2007, and Mahmudov in 2010 can be improved respectively. [ABSTRACT FROM AUTHOR]

Nonlinear m-term approximation plays an important role in machine learning, signal processing and statistical estimating. In this paper by means of a nondecreasing dominated function, a greedy adaptive compression numerical algorithm in the best m-term approximation with regard to tensor product wavelet-type basis is proposed. The algorithm provides the asymptotically optimal approximation for the class of periodic functions with mixed Besov smoothness in the L q norm. Moreover, it depends only on the expansion of function f by tensor product wavelet-type basis, but neither on q nor on any special features of f. [ABSTRACT FROM AUTHOR]

We study the approximation of the imbedding of functions from anisotropic and generalized Sobolev classes into L q ([0, 1]d) space in the quantum model of computation. Based on the quantum algorithms for approximation of finite imbedding from L to L , we develop quantum algorithms for approximating the imbedding from anisotropic Sobolev classes B( W ([0, 1] d )) to L q ([0, 1] d ) space for all 1 ⩽ q,p ⩽ ∞ and prove their optimality. Our results show that for p < q the quantum model of computation can bring a speedup roughly up to a squaring of the rate in the classical deterministic and randomized settings. [ABSTRACT FROM AUTHOR]

Szász-type operators can be constructed by a Poisson process. The purpose of this paper is to derive the converse result in connection with Szász-type operators by Steckin-Marchaud-type inequalities and new Ditzian modulus of continuity. The degree of approximation on deterministic signals is also given. [ABSTRACT FROM AUTHOR]

The paper presents a novel hybrid network for large-scale action recognition from multiple modalities. The network is built upon the proposed weighted dynamic images. It effectively leverages the strengths of the emerging Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) based approaches to specifically address the challenges that occur in large-scale action recognition and are not fully dealt with by the state-of-the-art methods. Specifically, the proposed hybrid network consists of a CNN based component and an RNN based component. Features extracted by the two components are fused through canonical correlation analysis and then fed to a linear Support Vector Machine (SVM) for classification. The proposed network achieved state-of-the-art results on the ChaLearn LAP IsoGD, NTU RGB+D and Multi-modal & Multi-view & Interactive ( M 2 I ) datasets and outperformed existing methods by a large margin (over 10 percentage points in some cases). [ABSTRACT FROM AUTHOR]

The current researches mainly adopt "Guide to the expression of uncertainty in measurement (GUM)" to calculate the profile error. However, GUM can only be applied in the linear models. The standard GUM is not appropriate to calculate the uncertainty of profile error because the mathematical model of profile error is strongly non-linear. An improved second-order GUM method (GUMM) is proposed to calculate the uncertainty. At the same time, the uncertainties in different coordinate axes directions are calculated as the measuring points uncertainties. In addition, the correlations between variables could not be ignored while calculating the uncertainty. A k-factor conversion method is proposed to calculate the converge factor due to the unknown and asymmetrical distribution of the output quantity. Subsequently, the adaptive Monte Carlo method (AMCM) is used to evaluate whether the second-order GUMM is better. Two practical examples are listed and the conclusion is drawn by comparing and discussing the second-order GUMM and AMCM. The results show that the difference between the improved second-order GUM and the AMCM is smaller than the difference between the standard GUM and the AMCM. The improved second-order GUMM is more precise in consideration of the nonlinear mathematical model of profile error. [ABSTRACT FROM AUTHOR]

Downsampling input images is a simple trick to speed up visual object-detection algorithms, especially on robotic vision and applied mobile vision systems. However, this trick comes with a significant decline in accuracy. In this paper, dual-resolution dual-path Convolutional Neural Networks (CNNs), named DualNets, are proposed to bump up the accuracy of those detection applications. In contrast to previous methods that simply downsample the input images, DualNets explicitly take dual inputs in different resolutions and extract complementary visual features from these using dual CNN paths. The two paths in a DualNet are a backbone path and an auxiliary path that accepts larger inputs and then rapidly downsamples them to relatively small feature maps. With the help of the carefully designed auxiliary CNN paths in DualNets, auxiliary features are extracted from the larger input with controllable computation. Auxiliary features are then fused with the backbone features using a proposed progressive residual fusion strategy to enrich feature representation.This architecture, as the feature extractor, is further integrated with the Single Shot Detector (SSD) to accomplish latency-sensitive visual object-detection tasks. We evaluate the resulting detection pipeline on Pascal VOC and MS COCO benchmarks. Results show that the proposed DualNets can raise the accuracy of those CNN detection applications that are sensitive to computation payloads. [ABSTRACT FROM AUTHOR]

To eliminate the numerical oscillations appearing in the first-order symmetric smoothed particle hydrodynamics (FO-SSPH) method for simulating transient heat conduction problems with discontinuous initial distribution, this paper presents a second-order symmetric smoothed particle hydrodynamics (SO-SSPH) method. Numerical properties of both SO-SSPH and FO-SSPH are analyzed, including truncation error, numerical accuracy, convergence rate, and stability. Experimental results show that for transient heat conduction with initial smooth distribution, both FO-SSPH and SO-SSPH can achieve second-order convergence, which is consistent with the theoretical analysis. However, for one- and two-dimensional conduction with initial discontinuity, the FO-SSPH method suffers from serious unphysical oscillations, which do not disappear over time, and hence it only achieves first-order convergence; while the present SO-SSPH method can avoid unphysical oscillations and has second-order convergence rate. Therefore, the SO-SSPH method is a feasible tool for solving transient heat conduction problems with both smooth and discontinuous distributions, and it is easy to be extended to high dimensional cases. [ABSTRACT FROM AUTHOR]