Showing papers by "Fan Yang published in 2015"
Posted Content•
TL;DR: In this article, the authors show that for instance-level image retrieval, lower layers often perform better than the last layers in convolutional neural networks, and adopt VLAD encoding to encode features into a single vector for each image.
Abstract: Deep convolutional neural networks have been successfully applied to image classification tasks. When these same networks have been applied to image retrieval, the assumption has been made that the last layers would give the best performance, as they do in classification. We show that for instance-level image retrieval, lower layers often perform better than the last layers in convolutional neural networks. We present an approach for extracting convolutional features from different layers of the networks, and adopt VLAD encoding to encode features into a single vector for each image. We investigate the effect of different layers and scales of input images on the performance of convolutional features using the recent deep networks OxfordNet and GoogLeNet. Experiments demonstrate that intermediate layers or higher layers with finer scales produce better results for image retrieval, compared to the last layer. When using compressed 128-D VLAD descriptors, our method obtains state-of-the-art results and outperforms other VLAD and CNN based approaches on two out of three test datasets. Our work provides guidance for transferring deep networks trained on image classification to image retrieval tasks.
329 citations
07 Jun 2015
TL;DR: This work provides guidance for transferring deep networks trained on image classification to image retrieval tasks, and presents an approach for extracting convolutional features from different layers of the networks, and adopts VLAD encoding to encode features into a single vector for each image.
Abstract: Deep convolutional neural networks have been successfully applied to image classification tasks. When these same networks have been applied to image retrieval, the assumption has been made that the last layers would give the best performance, as they do in classification. We show that for instance-level image retrieval, lower layers often perform better than the last layers in convolutional neural networks. We present an approach for extracting convolutional features from different layers of the networks, and adopt VLAD encoding to encode features into a single vector for each image. We investigate the effect of different layers and scales of input images on the performance of convolutional features using the recent deep networks OxfordNet and GoogLeNet. Experiments demonstrate that intermediate layers or higher layers with finer scales produce better results for image retrieval, compared to the last layer. When using compressed 128-D VLAD descriptors, our method obtains state-of-the-art results and outperforms other VLAD and CNN based approaches on two out of three test datasets. Our work provides guidance for transferring deep networks trained on image classification to image retrieval tasks.
290 citations
07 Dec 2015
TL;DR: A novel Multi-Task Learning with Low Rank Attribute Embedding (MTL-LORAE) framework for person re-identification that outperforms existing approaches by a large margin and produces state-of-the-art results.
Abstract: We propose a novel Multi-Task Learning with Low Rank Attribute Embedding (MTL-LORAE) framework for person re-identification. Re-identifications from multiple cameras are regarded as related tasks to exploit shared information to improve re-identification accuracy. Both low level features and semantic/data-driven attributes are utilized. Since attributes are generally correlated, we introduce a low rank attribute embedding into the MTL formulation to embed original binary attributes to a continuous attribute space, where incorrect and incomplete attributes are rectified and recovered to better describe people. The learning objective function consists of a quadratic loss regarding class labels and an attribute embedding error, which is solved by an alternating optimization procedure. Experiments on three person re-identification datasets have demonstrated that MTL-LORAE outperforms existing approaches by a large margin and produces state-of-the-art results.
226 citations
TL;DR: In this paper, a facile controlled synthesis method of Bi3+ self-doped NaBiO3 nanosheets was developed by using acidic hydrolysis of BiO3·2H2O in HNO3 solutions.
Abstract: A facile controlled synthesis method of Bi3+ self doped NaBiO3 nanosheets was developed by using acidic hydrolysis of NaBiO3·2H2O in HNO3 solutions. It was found that the hydrolysis of NaBiO3·2H2O in HNO3 solutions at low concentrations (0 420 nm). The pseudo-zero order reaction kinetic constant for the degradation of rhodamine B (20 μmol L−1) over the Bi3+ self doped NaBiO3 was 0.68 μmol L−1 min−1, being 3.1 times that of NaBiO3·2H2O. The much enhanced photocatalytic activity of the Bi3+ self doped NaBiO3 nanosheet photocatalyst was attributed to its improved absorption of visible light and separation of photo-generated h+/e− pairs.
102 citations
TL;DR: A concentration gradient controlled (CGC) kinetics model is proposed to explain the combined effects of lead-free soldering on Cu substrates and it is found that a larger Cu concentration gradient results in smaller Cu6Sn5 grains and more consumption of Cu substrate.
Abstract: Size effect model on kinetics of interfacial reaction between Sn-xAg-yCu solders and Cu substrate
99 citations
TL;DR: Clinical characteristics are better predictors of HRQOL in ESRD patients than socio-demographics in Singapore, and dialysis modality has no impact on the health utility of those patients.
Abstract: This study aimed to identify factors associated with the health-related quality of life (HRQOL) of multiethnic Asian end-stage renal disease (ESRD) patients treated with dialysis. The role of dialysis modality was also explored. Data used in this study were from two cross-sectional surveys of Singaporean ESRD patients on haemodialysis (HD) or peritoneal dialysis (PD). In both surveys, participants were assessed using the kidney disease quality of life (KDQOL) instrument and questions assessing socio-demographic characteristics. Clinical data including co-morbidity (measured by Charlson comorbidity index [CCI]), albumin level, haemoglobin level, and dialysis-related variables (e.g. dialysis vintage and dialysis adequacy) were retrieved from medical records. The 36-item KDQOL (KDQOL-36) was used to generate three summary scores (physical component summary [PCS], mental component summary [MCS] and kidney disease component summary [KDCS]) and two health utility scores (Short Form 6-dimension [SF-6D] and EuroQol 5-dimension [EQ-5D]). Linear regression analysis was performed to examine the association of factors with each of the HRQOL scale scores. Five hundred and two patients were included in the study (mean age 57.1 years; male 52.4 %; HD 236, PD 266). Mean [standard deviation (SD)] PCS, MCS and KDCS scores were 37.9 (9.7), 46.4 (10.8) and 57.6 (18.1), respectively. Mean (SD) health utility score was 0.66 (0.12) for SF-6D and 0.60 (0.21) for EQ-5D. In multivariate regression analysis, factors found to be significantly associated with better HRQOL included: young ( 60 years), low CCI (<5), high albumin (≥37 g/l) and high haemoglobin (≥11 g/dl) with PCS; long dialysis vintage (≥3.5 years) with MCS; old age, Malay ethnicity and PD modality with KDCS; low CCI, high albumin and high haemoglobin with EQ-5D and high albumin with SF-6D. Clinical characteristics are better predictors of HRQOL in ESRD patients than socio-demographics in Singapore. Dialysis modality has no impact on the health utility of those patients.
73 citations
TL;DR: In this article, a novel heterostructure of TiO2 modified Co3O4 acicular nanowire arrays has been fabricated, which demonstrates a good performance for ethanol detection at working temperatures as low as 160 °C.
Abstract: A novel heterostructure of TiO2 modified Co3O4 (TiO2/Co3O4) acicular nanowire (NW) arrays has been fabricated in this study, which demonstrates a good performance for ethanol detection at working temperatures as low as 160 °C. Co3O4 NW arrays were first grown on an Al2O3 substrate patterned with an Ag/Pd electrode by a hydrothermal method, and then TiO2 nanoparticles were decorated on the surface of Co3O4 NW arrays by using pulsed laser deposition (PLD). It is found that after decoration of TiO2, the TiO2/Co3O4 NW array sensor exhibits a much higher response to ethanol (Rg/Ra = 65, Rg is the sensor resistance measured in a mixture of target gases and Ra is the resistance measured in air) compared with the pristine Co3O4 NW sensor (Rg/Ra = 25). Importantly, the TiO2/Co3O4 sensor has shown a detection limit as low as 10 ppm, and a good reproducibility. The reason for the enhanced sensing properties of TiO2/Co3O4 is considered to be due to the formation of a p–n junction between the p-type Co3O4 and n-type TiO2.
73 citations
TL;DR: In this paper, the Pd/PdO catalysts with uniform size distribution exhibited remarkable catalytic activity during the reduction of 4-nitrophenol (4-NP) in neat water at room temperature.
Abstract: In this work, we demonstrate that the presence of PdO nanoparticles can significantly enhance the catalytic performance of Pd catalysts for the reduction of 4-nitrophenol (4-NP). Heterogeneous Pd/PdO nanoparticles supported on an oxidized multi-walled carbon nanotube (OCNTs) catalyst is prepared by a one-pot gas-liquid interfacial plasma (GLIP) method with the precursor Pd(NO3)(2)center dot 2H(2)O. The Pd/PdO catalysts with uniform size distribution exhibit remarkable catalytic activity during the reduction of 4-NP to 4-aminophenol (4-AP) in neat water at room temperature. The turnover frequency (TOF) value is up to 750 h(-1), which shows much higher catalytic activity than single Pd nanoparticles supported on OCNTs. Our results indicate that the Pd/PdO catalyst can be readily recovered and reused 10 times.
67 citations
TL;DR: In this article, a 3D composite based on reduced graphene oxide (rGO)/Fe2O3 was prepared by a one-pot hydrothermal method, and the resulting composite was found to self-assemble to form a three-dimensional network via hydro-thermal treatment.
Abstract: A novel three-dimensional (3D) composite based on reduced graphene oxide (rGO)/Fe2O3 was prepared by a one-pot hydrothermal method. Fe2O3 nanoparticles were either attached on the surface of graphene sheets or coated uniformly in the graphene sheets. The resulting composite is found to self-assemble to form a 3D network via hydrothermal treatment. Our results indicate that the as-prepared Fe2O3 nanoparticles show a porous morphology, which results in the rGO/Fe2O3 composites exhibiting excellent microwave absorbing properties in the range of 2–16 GHz and are therefore expected to be a promising candidate as microwave absorbing materials.
60 citations
TL;DR: In this paper, a large area reduced graphene oxide (rGO)/polyvinyl alcohol (PVA) film with excellent microwave absorption properties has been prepared by a simple solution processing method.
Abstract: A large-area reduced graphene oxide (rGO)/poly(vinyl alcohol) (PVA) film with excellent microwave absorption properties has been prepared by a simple solution processing method. Excellent interfacial interaction between GO and PVA has been realized due to the molecule-level dispersion. Characterization by X-ray diffraction, the Fourier transform infrared (FT-IR) spectra, X-ray photoelectron spectrum and Raman spectroscopy has confirmed the reduction process of GO in the composites. The rGO/PVA film exhibits excellent microwave absorbing properties in the range of 2–18 GHz. It is expected to be a promising candidate as a microwave absorbing material.
55 citations
TL;DR: Both EQ-5D-5L and SF-6D are valid instruments for assessing ESRD patients, however, the two preference-based measures cannot be used interchangeably and it appears that EQ-3D-3L would lead to more favorable cost-effectiveness results than SF- 6D if they are used in economic evaluations of interventions for E SRD.
Abstract: Objectives
The objective of this study was to compare the performance of the 5-level EuroQol 5-dimension (EQ-5D-5L) and the Short Form 6-dimension (SF-6D) instruments in assessing patients with end-stage renal disease (ESRD) in Singapore.
TL;DR: In this article, the authors reported that the presence of PdO nanoparticles can enhance the catalytic performance of the Pd catalyst for the Suzuki reaction in water, and they used Pd(NO3)(2)center dot 2H(2)O as a precursor.
TL;DR: Host-guest carbon dots were designed and prepared with red/near-IR dyes encapsulated as guest in the carbon nanoparticle core, and beyond the desired enhancement in optical properties, the host-guer configuration may significantly broaden the field of carbon dots.
Abstract: Carbon dots, generally small carbon nanoparticles with various forms of surface passivation, have achieved the performance level of semiconductor quantum dots in the green spectral region, but their absorption and fluorescence in red/near-IR are relatively weaker. Conceptually similar to endofullerenes, host-guest carbon dots were designed and prepared with red/near-IR dyes encapsulated as guest in the carbon nanoparticle core. Beyond the desired enhancement in optical properties, the host-guest configuration may significantly broaden the field of carbon dots.
TL;DR: Evidence of this meta-analysis strongly supports the sex dimorphism in the prognosis of trauma patients and further work should be done to decipher potential mechanism.
Abstract: Background This study aims to assess the influence of sex on outcomes among trauma patients, including injury severity, medical resource utility, complications, and mortality. Methods A systematic review of the literature was conducted by internet search. Data were extracted from selected studies and analyzed using Stata to compare outcomes between male and female injured patients. Results Eventually, 19 studies met our inclusion criteria with 100,566 men and 39,762 women. Pooled data revealed that male sex was associated with increased risk of mortality, hospital length of stay, and higher incidence of complications. No difference was detected in injury severity between male and female patients. Conclusion Evidence of this meta-analysis strongly supports the sex dimorphism in the prognosis of trauma patients and further work should be done to decipher potential mechanism.
TL;DR: In this paper, a modification of few-layer graphene grown by chemical vapour deposition via nitrogen plasma ion irradiation and its application as counter electrodes in bifacial dye-sensitized solar cells (DSSCs) was reported.
TL;DR: The results indicate the significance of shyness-sensitivity for adjustment and the role of adjustment in the development of shys-s sensitivity in today's urban Chinese society.
Abstract: This study examined reciprocal contributions between shyness-sensitivity and social, school, and psychological adjustment in urban Chinese children. Longitudinal data were collected once a year from Grade 3 to Grade 6 (ages 9-12 years) for 1,171 children from multiple sources. Shyness-sensitivity positively contributed to social, school, and psychological difficulties over time, with the most consistent effects on peer preference and loneliness. Social and school adjustment negatively contributed to the development of shyness-sensitivity. The initial levels of shyness-sensitivity and social and school adjustment moderated the growth of each other, mainly as a resource-potentiating factor. The results indicate the significance of shyness-sensitivity for adjustment and the role of adjustment in the development of shyness-sensitivity in today's urban Chinese society.
TL;DR: Subgroup analysis showed that PD may be similar to or better than HD in survival outcomes among young patients (≤65 years old) without diabetes or cardiovascular disease, although survival outcomes may not differ between the two dialysis modalities in young and healthier patients.
Abstract: Background
Studies comparing patient survival of hemodialysis (HD) and peritoneal dialysis (PD) have yielded conflicting results and no such study was from South-East Asia. This study aimed to compare the survival outcomes of patients with end-stage renal disease (ESRD) who started dialysis with HD and PD in Singapore.
TL;DR: In this paper, a 3D nitrogen-doped graphene nano-architecture was synthesized by using catalytic carbonization of aromatic polyimide (PI) on magnesium oxide (MgO) nanowires.
Abstract: Three-dimensional (3D) nanocarbon architectures have attracted great interest in materials science and nanotechnology. In this work, we report the synthesis of a 3D nitrogen-doped graphene nanoarchitecture by using catalytic carbonization of aromatic polyimide (PI) on magnesium oxide (MgO) nanowires. The produced porous graphene nanofibers are crosslinked into an integrated monolithic structure, thereby offering continuous electron conductivity and efficient charge transport. Moreover, nitrogen doping was achieved during the in situ carbonization, enhancing the electronic and interfacial properties of the porous graphene nanofibers. The obtained monolithic frameworks were directly used as binder-free electrodes for lithium batteries, and yield remarkable electrochemical performances, such as high reversible capacity, high rate capability, and superb cycling stability. The monolithic structures can be used as a material platform for the preparation of other functional composites, extending their applications beyond energy storage.
05 Jan 2015
TL;DR: A re-ranking algorithm for image retrieval by fusing multi-feature information to exploit the underlying relationships among images and significantly improves performance over baseline methods and outperforms many state-of-the-art retrieval methods.
Abstract: We present a re-ranking algorithm for image retrieval by fusing multi-feature information. We utilize pair wise similarity scores between images to exploit the underlying relationships among images. The initial ranked list for a query from each feature is represented as an undirected graph, where edge strength comes from feature-specific image similarity. Graphs from multiple features are combined by a mixture Markov model. In addition, we utilize a probabilistic model based on the statistics of similarity scores of similar and dissimilar image pairs to determine the weight for each graph. The weight for a feature is query specific, where the ranked lists of different queries receive different weights. Our approach for calculating weights is data-driven and does not require any learning. A diffusion process is then applied to the fused graph to reduce noise and achieve better retrieval performance. Experiments demonstrate that our approach significantly improves performance over baseline methods and outperforms many state-of-the-art retrieval methods.
TL;DR: In this paper, a simple approach for fabricating flexible, free-standing, catalytic films composed of graphene oxide/carbon nanotube-Au composites by using a one-pot chemical reduction route was reported.
Abstract: We report a simple approach for fabricating flexible, free-standing, catalytic films composed of graphene oxide/carbon nanotube–Au (GO/CNT–Au) composites by using a one-pot chemical reduction route. The catalyst film was characterized by scanning electron microscopy, transmission electron microscopy, X-ray diffraction, X-ray photoelectron spectroscopy, thermogravimetric analysis and Raman spectroscopy to investigate the morphology, crystalline structure of the composites. The layered structure of the catalyst film is porous, flexible and exhibits excellent catalytic property in the reduction of 4-nitrophenol (4-NP) to 4-aminophenol (4-AP) reaction. This study demonstrate here for simple yet effective synthesis of catalyst film with a number of advantages such as good strength, high stability, easy operation, and good reusability with minimal loss of activity.
01 Oct 2015
TL;DR: A proposed algorithm for dynamically analyzing facial expressions using robust descriptors in order to compose a novel feature selection as well as an effective classification process for depression severity assessment is described.
Abstract: Depression is one of the most common mental disorders affecting millions of people worldwide. Developing adjunct tools aiding depression assessment is expected to impact overall health outcomes and treatment cost reduction. To this end, platforms designed for automatic and non-invasive depression assessment could help in detecting signs of the disease on a regular basis, without requiring the physical presence of a mental health professional. Despite the different approaches that can be found in the literature, both in terms of methods and algorithms, a fully satisfactory system for the automatic assessment of depression severity has not been presented as yet. This paper describes a proposed algorithm for dynamically analyzing facial expressions using robust descriptors in order to compose a novel feature selection as well as an effective classification process. Additionally a preliminary evaluation of the system is presented, by applying local curvelet binary patterns in three orthogonal planes for depression severity assessment.
TL;DR: The developed SFE method with online CNTs trapping followed by GC-MS analysis has been demonstrated to be an efficient way for quantitative analysis of trace-level PAHs and their nitro-, oxy-, and alkyl-derivatives in soil samples.
TL;DR: In this article, a bifacial dye-sensitized solar cells (DSSCs) with the ability to utilize incidental light from both front and rear-sides have drawn massive attention from researchers in recent years.
TL;DR: A series of yttrium-doped lanthanum zirconate (La1−xYx)2Zr2O7 bulk samples were prepared and their ionic conductivities were studied by impedance spectroscopy.
TL;DR: A Stein kernel on a Riemannian manifold instead of Euclidean metrics, combined with sparse representation, is employed for polarimetric SAR land cover classification, named Stein-sparse representation-based classification (SRC).
Abstract: Normally, polarimetric SAR classification is a high-dimensional nonlinear mapping problem. In the realm of pattern recognition, sparse representation is a very efficacious and powerful approach. As classical descriptors of polarimetric SAR, covariance and coherency matrices are Hermitian semidefinite and form a Riemannian manifold. Conventional Euclidean metrics are not suitable for a Riemannian manifold, and hence, normal sparse representation classification cannot be applied to polarimetric SAR directly. This paper proposes a new land cover classification approach for polarimetric SAR. There are two principal novelties in this paper. First, a Stein kernel on a Riemannian manifold instead of Euclidean metrics, combined with sparse representation, is employed for polarimetric SAR land cover classification. This approach is named Stein-sparse representation-based classification (SRC). Second, using simultaneous sparse representation and reasonable assumptions of the correlation of representation among different frequency bands, Stein-SRC is generalized to simultaneous Stein-SRC for multi-frequency polarimetric SAR classification. These classifiers are assessed using polarimetric SAR images from the Airborne Synthetic Aperture Radar (AIRSAR) sensor of the Jet Propulsion Laboratory (JPL) and the Electromagnetics Institute Synthetic Aperture Radar (EMISAR) sensor of the Technical University of Denmark (DTU). Experiments on single-band and multi-band data both show that these approaches acquire more accurate classification results in comparison to many conventional and advanced classifiers.
TL;DR: In this article, a Ni-Y/SWCNTN was used to obtain ultra-deep hydrodesulfurization (HDS) on the fluid catalytic cracking (FCC) gasoline and diesel.
TL;DR: A center extraction method is proposed for improving the accuracy of the laser stripe center extraction based on image evaluation of Gaussian fitting structural similarity and analysis of the multiple source factors and measurement experiments for a large-scale aviation composite component are carried out.
Abstract: Large-scale triangulation scanning measurement systems are widely used to measure the three-dimensional profile of large-scale components and parts. The accuracy and speed of the laser stripe center extraction are essential for guaranteeing the accuracy and efficiency of the measuring system. However, in the process of large-scale measurement, multiple factors can cause deviation of the laser stripe center, including the spatial light intensity distribution, material reflectivity characteristics, and spatial transmission characteristics. A center extraction method is proposed for improving the accuracy of the laser stripe center extraction based on image evaluation of Gaussian fitting structural similarity and analysis of the multiple source factors. First, according to the features of the gray distribution of the laser stripe, evaluation of the Gaussian fitting structural similarity is estimated to provide a threshold value for center compensation. Then using the relationships between the gray distribution of the laser stripe and the multiple source factors, a compensation method of center extraction is presented. Finally, measurement experiments for a large-scale aviation composite component are carried out. The experimental results for this specific implementation verify the feasibility of the proposed center extraction method and the improved accuracy for large-scale triangulation scanning measurements.
24 May 2015
TL;DR: A CMOS rectifier with unbalanced source-coupled pairs and auxiliary capacitors is utilized to constitute the reverse received signal strength indicator (reverse-RSSI) architecture as proposed power detector with operating frequency from 300 MHz to 10 GHz.
Abstract: A wide band radio frequency (RF) root-mean-square (RMS) power detector (PD) is presented in this paper A CMOS rectifier with unbalanced source-coupled pairs and auxiliary capacitors is utilized to constitute the reverse received signal strength indicator (reverse-RSSI) architecture as proposed power detector with operating frequency from 300 MHz to 10 GHz The auxiliary capacitors are introduced to improve linearity at high input power dramatically The power detector can be connected to power amplifier without directional coupler due to the capacitor attenuation array Simulation results show that the maximum detection power is as high as +30 dBm and dynamic range reaches more than 42 dB with ±1 dB error The proposed power detector is implemented in a standard 180nm CMOS process with 0113 mm2 core area The supply voltage is 33 V, and its static power consumption is 055 mW
TL;DR: There was a relationship between kinematic changes, representing the motor planning strategies, and physical frailty in these aged adults, and these changes should be taken into account in clinical practice.
Abstract: Objective
The purpose of this work was to analyze and compare the movement kinematics of sit-to-stand (STS) and back-to-sit (BTS) transfers between frail aged adults and young subjects, as well as to determine the relationship between kinematic changes and functional capacities.
TL;DR: In this paper, a simplified one-step hydrothermal approach was used to prepare the graphene-snO 2 -amorphous carbon composites which have a fixed SnO 2 content and various graphene to amorphous carbon (Gr/C) mass ratio as anode for Li-ion batteries.