This paper targets on the problem of set to recognition, which learns metric between two image sets. Images in each belong same identity. Since images a can be complementary, they hopefully lead higher accuracy practical applications. However, quality sample cannot guaranteed, and samples with poor will hurt metric. In this paper, aware network (QAN) is proposed confront problem, where automatically learned although such information not explicitly provided training stage. The has branches,...

Evidence theory, also called belief function provides an efficient tool to represent and combine uncertain information for pattern classification. combination can be interpreted, in some applications, as classifier fusion. The sources of evidence corresponding multiple classifiers usually exhibit different classification qualities, they are often discounted using weights before combination. In order achieve the best possible fusion performance, a new credal redistribution (CBR) method is...

Pavement crack detection from images is a challenging problem due to intensity inhomogeneity, topology complexity, low contrast, and noisy texture background. Traditional learning-based approaches have difficulties in obtaining representative training samples. We propose new unsupervised multi-scale fusion (MFCD) algorithm that does not require data. First, we develop windowed minimal path-based method extract the candidate cracks image at each scale. Second, find correspondences across...

Distillation-based learning boosts the performance of miniaturized neural network based on hypothesis that representation a teacher model can be used as structured and relatively weak supervision, thus would easily learned by model. However, we find converged heavy is still strong constraint for training small student model, which leads to higher lower bound congruence loss. In this work, consider knowledge distillation from perspective curriculum teacher's routing. Instead supervising with...

Deep learning has made significant achievements in many application areas. To train and test models more efficiently, enterprise developers submit run their deep programs on a shared, multi-tenant platform. However, some of the fail after long execution time due to code/script defects, which reduces development productivity wastes expensive resources such as GPU, storage, network I/O.

Image matching is a primary technology for optical and synthetic aperture radar (SAR) image fusion but often shows limited performance due to the highly nonlinear differences between SAR modalities. Recently, deep neural networks (DNNs) have been investigated effectively extract features tasks, where DNNs are trained based on elaborated design of loss functions low value expected obtain better performance. In this letter, we first theoretically demonstrate that when state-of-the-art function...

Convolutional neural networks (CNNs) have been widely applied in the context of ship detection synthetic aperture radar (SAR) images, but performance is still not ideal scenarios with clutter interference. Mining frequency-domain information to suppress sea SAR has attracted wide attention. However, existing methods do process adaptively, which results degradation performance. To overcome this problem, article proposes a novel deep learning network called YOLO-FA. YOLO-FA contains proposed...

Abstract Inspired by the remarkable performance of SnSe‐based compounds in thermoelectrics, a strontium‐tin‐selenium (SrSnSe 2 ) compound is theoretically designed, observing anisotropic Rashba spin‐orbital splitting and strong four‐phonon scattering behavior. Through comprehensive analyses elastic constants, phonon dispersion, ab initio molecular dynamics calculations, mechanical, dynamic, thermal stability SrSnSe demonstrated. Electronic calculations reveal that band decomposed into two...

This paper targets on the problem of set to recognition, which learns metric between two image sets. Images in each belong same identity. Since images a can be complementary, they hopefully lead higher accuracy practical applications. However, quality sample cannot guaranteed, and samples with poor will hurt metric. In this paper, aware network (QAN) is proposed confront problem, where automatically learned although such information not explicitly provided training stage. The has branches,...

Change detection in heterogeneous remote sensing images is crucial for disaster damage assessment. Recent methods use homogenous transformation, which transforms the optical and synthetic aperture radar (SAR) into same feature space, to achieve change detection. Such transformations mainly operate on low-level space may corrupt semantic content, deteriorating performance of To solve this problem, article presents a new homogeneous transformation model termed deep fusion (DHFF) based image...

The emerging volumetric videos offer a fully immersive, six degrees of freedom (6DoF) viewing experience, at the cost extremely high bandwidth demand. In this paper, we design, implement, and evaluate Vues, an edge-assisted transcoding system that delivers high-quality with low requirement, decoding overhead, quality experience (QoE) on mobile devices. Through IRB-approved user study, build first-of-its-kind QoE model to quantify impact various factors introduced by content into 2D videos....

In this paper, we propose a new paradigm, named Historical Object Prediction (HoP) for multi-view 3D detection to leverage temporal information more effectively. The HoP approach is straightforward: given the current times-tamp t, generate pseudo Bird's-Eye View (BEV) feature of timestamp t-k from its adjacent frames and utilize predict object set at t-k. Our motivated by observation that enforcing detector capture both spatial location motion objects occurring historical timestamps can lead...

A novel filter for nonlinear and non-Gaussian systems is proposed in this paper. The unscented Kalman designed to give a preliminary estimation of the state. An additional RBF-network added UKF innovation term compensate non-Gaussianity whole system. Renyi's entropy introduced parameters are updated using minimum criterion at each time step. It has been shown that algorithm high accuracy because can characterize all randomness residual while only cares mean covariance. proved with properly...

Non-volitional discontinuation of motion, namely bradykinesia, is a common motor symptom among patients with Parkinson's disease (PD). Evaluating bradykinesia severity an important part clinical examinations on PD in both diagnosis and monitoring phases. However, subjective evaluations from different clinicians often show low consistency. The research works that explore objective quantification are mostly based highly-integrated sensors. Although these sensor-based methods demonstrate...

Homogeneous regions, which are smooth areas that lack blur clues to discriminate if they focused or non-focused. Therefore, bring a great challenge achieve high accurate multi-focus image fusion (MFIF). Fortunately, we observe depth maps highly related focus and defocus, containing preponderance of discriminative power locate homogeneous regions. This offers the potential provide additional cues assist MFIF task. Taking into consideration, in this paper, propose new depth-distilled...

Here we explore two related but important tasks based on the recently released REalistic Single Image DEhazing (RESIDE) benchmark dataset: (i) single image dehazing as a low-level restoration problem; and (ii) high-level visual understanding (e.g., object detection) of hazy images. For first task, investigated variety loss functions show that perception-driven significantly improves performance. In second provide multiple solutions including using advanced modules in dehazing-detection...

Regularization method is an effective tool for synthetic aperture radar (SAR) image despeckling. Design of the regularization terms describing priors plays a vital role in this kind method. In article, new combinational model speckle reduction (CRM-SR) proposed, which term elaborately designed to contain both fractional-order total variation (FrTV) and nonlocal low rank (NLR) regularization. The inherits advantages FrTV NLR improves performance SAR despeckling and, therefore, better...

In this paper, we propose a two-level block matching pursuit (TLBMP) algorithm based on probabilistic graph model for polarimetric through-wall radar imaging (TWRI). typical L-band to X-band TWRI, indoor targets assume spatial extent and occupy clustered pixels. When sensing is used obtain independent observations, images of can be enhanced within the joint sparsity framework. Toward objective, TLBMP devised exploit both property pattern multiple images. Simulations experimental results data...

Space-borne synthetic aperture radar (SAR) and optical sensors are important tools for building damage detection. Fusion of SAR images improves detection performance. However, when the resolutions two different kinds differ, performance existing pixel-level fusion methods deteriorates significantly due to interpolation-induced distortion. To solve this problem, paper presents a new superpixel-based belief (SBBF) model The superpixels on identified by segmentation pre-earthquake image perform...