Sketch-based face recognition is an interesting task in vision and multimedia research, yet it quite challenging due to the great difference between photos sketches. In this paper, we propose a novel approach for photo-sketch generation, aiming automatically transform into detail-preserving personal Unlike traditional models synthesizing sketches based on dictionary of exemplars, develop fully convolutional network learn end-to-end mapping. Our takes whole as inputs directly generates...

This paper investigates how to rapidly and accurately localize facial landmarks in unconstrained, cluttered environments rather than the well segmented face images. We present a novel Backbone-Branches Fully-Convolutional Neural Network (BB-FCN), which produces landmark response maps directly from raw images without relying on pre-process or sliding window approaches. BB-FCN contains one backbone number of network branches with each corresponding type, it operates progressive manner....

Deep models have achieved impressive performance for face hallucination tasks. However, we observe that directly feeding the hallucinated facial images into recog- nition can even degrade recognition despite much better visualization quality. In this paper, address problem by jointly learning a deep model two tasks, i.e. and recognition. particular, design an end-to-end convolution network with sub-network cascaded sub-network. The sub- are responsible producing discriminative feature...

The past decade has witnessed the rapid development of feature representation learning and distance metric learning, whereas two steps are often discussed separately. To explore their interaction, this work proposes an end-to-end framework called DARI, i.e. Distance And Representation Integration, validates effectiveness DARI in challenging task person verification. Given training images annotated with labels, we first produce a large number triplet units, each one contains three images,...

In this paper, we propose to endow deep 3D models with rotation invariance by expressing the coordinates in an intrinsic frame determined object shape itself. Key our approach is find such which should be unique identical and consistent across different instances of same category. Interestingly, principal component analysis exactly provides effective way define a frame, i.e. setting components as axes. As have direction ambiguity, there exist several frames for each object. To achieve...

Deep models are capable of fitting complex high dimensional functions while usually yielding large computation load. There is no way to speed up the inference process by classical lookup tables due high-dimensional input and limited memory size. Recently, a novel architecture (PointNet) for point clouds has demonstrated that it possible obtain complicated deep function from set 3-variable functions. In this paper, we exploit property apply table encode these This method ensures time only...

Pore features and seepage characteristics of mixed granular materials are great significance in many subjects, including engineering, for instance geotechnical petroleum extraction, hydrogeology, environmental science hydraulic engineering. In this paper, two types mathematical models proposed that consider different packing methods porosity permeability – namely, the filling model (FM) replacement (RM). On basis, incomplete coverage (overlap between large small grains during process) is...

Previous works have shown that face recognition with high accurate 3D data is more reliable and insensitive to pose illumination variations. Recently, low-cost portable acquisition techniques like ToF(Time of Flight) DoE based structured light systems enable us access easily, e.g., via a mobile phone. However, such devices only provide sparse(limited speckles in system) noisy which can not support directly. In this paper, we aim at achieving high-performance for equipped modules very...

The past decade has witnessed the rapid development of feature representation learning and distance metric learning, whereas two steps are often discussed separately. To explore their interaction, this work proposes an end-to-end framework called DARI, i.e. Distance And Representation Integration, validates effectiveness DARI in challenging task person verification. Given training images annotated with labels, we first produce a large number triplet units, each one contains three images,...

Abstract The migration, deposition and detachment of suspended particles in saturated sands are used many fields. Temperature is an important influencing factor the transport process porous media, it plays role ground‐water heat pump (GWHP) projects. This article discusses independent development a three‐dimensional percolation sandbox test device design particle migration based on Reynolds number context backfilling GWHP systems as experimental background. effects various temperatures (10,...

Cloud offloading has recently attracted a substantial amount of attention from both industry and academia. This new generation service, beyond conventional task scheduling management, utilizes the abundant computation capacity public clouds takes it as part external resource on mobile devices. In this paper, we find that cloud can potentially increase running latency offloaded tasks. Our measurement shows systems, such Gaikai, introduce up to 400ms communication between customers remote...

This paper presents a digital image processing (DIP) based finite difference method (FDM) and makes the first attempt to apply new failure process of stratified rocks from Chinese Jinping underground carves. In method, two-dimensional (2D) inhomogeneity mesostructures rock materials are identified with DIP technique. And then binarization information is used generate grids. Finally, samples under uniaxial compression condition simulated by using FDM. DIP, an segmentation algorithm on seeded...

In the past few years, we have witnessed a flourish of online social network sites (OSNs). this kind websites, users are not only information consumers, but also actively upload contents their own to OSNs. Being sharply different from conventional sites, OSNs attracted many studies recently. These studies, however, mainly focus on behaviors within OSNs, e.g., user-user interaction, and distribution time spend certain etc. There much fewer characteristics traffic patterns upon Internet...

Identifying the same individual across different scenes is an important yet difficult task in intelligent video surveillance. Its main difficulty lies how to preserve similarity of person against large appearance and structure variation while discriminating individuals. In this paper, we present a scalable distance driven feature learning framework based on deep neural network for re-identification, demonstrate its effectiveness handle existing challenges. Specifically, given training images...

Abstract To identify the superconvergent areas of gradient solutions is a worthwhile part recovery in finite element method. Compared with triangular and quadrilateral elements, position determination for polygonal much more difficult because non‐polynomial form interpolation functions. In this article, superconvergence points C 0 discovered, its existence proved theoretically. By dividing canonical into sub‐triangles, set provisional elements generated to replace existing polygon element....