Search

People may perform diverse gestures affected by various mental and physical factors when speaking the same sentences. This inherent one-to-many relationship makes co-speech gesture generation from audio particularly challenging. Conventional CNNs/RNNs assume one-to-one mapping, and thus tend to predict the average of all possible target motions, easily resulting in plain/boring motions during inference. So we propose to explicitly model the one-to-many audio-to-motion mapping by splitting the cross-modal latent code into shared code and motion-specific code. The shared code is expected to be responsible for the motion component that is more correlated to the audio while the motion-specific code is expected to capture diverse motion information that is more independent of the audio. However, splitting the latent code into two parts poses extra training difficulties. Several crucial training losses/strategies, including relaxed motion loss, bicycle constraint, and diversity loss, are designed to better train the VAE. Experiments on both 3D and 2D motion datasets verify that our method generates more realistic and diverse motions than previous state-of-the-art methods, quantitatively and qualitatively. Besides, our formulation is compatible with discrete cosine transformation (DCT) modeling and other popular backbones (i.e., RNN, Transformer). As for motion losses and quantitative motion evaluation, we find structured losses/metrics (e.g. STFT) that consider temporal and/or spatial context complement the most commonly used point-wise losses (e.g. PCK), resulting in better motion dynamics and more nuanced motion details. Finally, we demonstrate that our method can be readily used to generate motion sequences with user-specified motion clips on the timeline.

RGBT tracking seeks to leverage both visible and thermal infrared images to enhance the robustness of target tracking. This method makes up for the limitations of single-sensor tracking. The RGB and thermal infrared images complement each other effectively, enabling the tracker to operate seamlessly in complex environments day and night. However, RGBT tracking still faces challenges, such as thermal crossover, occlusion (OCC), illumination changes, rapid target movement, and background clutter. These problems hinder the performance improvement of RGBT trackers. Researchers have proposed numerous RGBT tracking methods in response to these challenges over time. This article aims to comprehensively review and summarize the latest research progress in RGBT tracking. First, we categorize the existing tracing methods according to the RGBT tracing framework. Then, we introduce the datasets/benchmarks used to evaluate RGBT tracking methods. Subsequently, we present the results of representative tracking methods on multiple testing benchmarks, providing a more intuitive presentation of current research advances in the field of RGBT tracking. Finally, research directions are discussed to further advance the development of RGBT tracking.

Arbitrary image stylization by neural networks has become a popular topic, and video stylization is attracting more attention as an extension of image stylization. However, when image stylization methods are applied to videos, unsatisfactory results that suffer from severe flickering effects appear. In this article, we conducted a detailed and comprehensive analysis of the cause of such flickering effects. Systematic comparisons among typical neural style transfer approaches show that the feature migration modules for state-of-the-art (SOTA) learning systems are ill-conditioned and could lead to a channelwise misalignment between the input content representations and the generated frames. Unlike traditional methods that relieve the misalignment via additional optical flow constraints or regularization modules, we focus on keeping the temporal consistency by aligning each output frame with the input frame. To this end, we propose a simple yet efficient multichannel correlation network (MCCNet), to ensure that output frames are directly aligned with inputs in the hidden feature space while maintaining the desired style patterns. An inner channel similarity loss is adopted to eliminate side effects caused by the absence of nonlinear operations such as softmax for strict alignment. Furthermore, to improve the performance of MCCNet under complex light conditions, we introduce an illumination loss during training. Qualitative and quantitative evaluations demonstrate that MCCNet performs well in arbitrary video and image style transfer tasks. Code is available at https://github.com/kongxiuxiu/MCCNetV2.

Thermal infrared (TIR) target tracking task is not affected by illumination changes and can be tracked at night, on rainy days, foggy days, and other extreme weather; so it is widely used in auxiliary driving, unmanned aerial vehicle reconnaissance, video surveillance, and other scenes. However, the TIR target tracking task also presents some challenges, such as intensity change, occlusion, deformation, similarity interference, and so on. These challenges significantly affect the performance of the TIR target tracking methods. To resolve these challenges in the TIR target tracking scenarios, numerous tracking methods have appeared in recent years. The purpose of this article is to give a comprehensive review and summary of the research status of the TIR target tracking methods. We first classify the TIR target tracking methods according to their frameworks and briefly summarize the advantages and disadvantages of different tracking methods, which can better understand the current research progress of the TIR target tracking methods. Next, the public datasets/benchmarks for performance testing of the TIR target tracking methods are introduced. Subsequently, we demonstrate the tracking results of several representative tracking methods on some datasets/benchmarks to more intuitively show the progress of the TIR target tracking methods made in the current research. Finally, we discussed the future research direction in an attempt to promote the better development of the TIR target-tracking tasks.

While deep-learning-based tracking methods have achieved substantial progress, they entail large-scale and high-quality annotated data for sufficient training. To eliminate expensive and exhaustive annotation, we study self-supervised (SS) learning for visual tracking. In this work, we develop the crop-transform-paste operation, which is able to synthesize sufficient training data by simulating various appearance variations during tracking, including appearance variations of objects and background interference. Since the target state is known in all synthesized data, existing deep trackers can be trained in routine ways using the synthesized data without human annotation. The proposed target-aware data-synthesis method adapts existing tracking approaches within a SS learning framework without algorithmic changes. Thus, the proposed SS learning mechanism can be seamlessly integrated into existing tracking frameworks to perform training. Extensive experiments show that our method: 1) achieves favorable performance against supervised (Su) learning schemes under the cases with limited annotations; 2) helps deal with various tracking challenges such as object deformation, occlusion (OCC), or background clutter (BC) due to its manipulability; 3) performs favorably against the state-of-the-art unsupervised tracking methods; and 4) boosts the performance of various state-of-the-art Su learning frameworks, including SiamRPN++, DiMP, and TransT.

Unlike visual object tracking, thermal infrared (TIR) object tracking methods can track the target of interest in poor visibility such as rain, snow, and fog, or even in total darkness. This feature brings a wide range of application prospects for TIR object-tracking methods. However, this field lacks a unified and large-scale training and evaluation benchmark, which has severely hindered its development. To this end, we present a large-scale and high-diversity unified TIR single object tracking benchmark, called LSOTB-TIR, which consists of a tracking evaluation dataset and a general training dataset with a total of 1416 TIR sequences and more than 643 K frames. We annotate the bounding box of objects in every frame of all sequences and generate over 770 K bounding boxes in total. To the best of our knowledge, LSOTB-TIR is the largest and most diverse TIR object tracking benchmark to date. We spilt the evaluation dataset into a short-term tracking subset and a long-term tracking subset to evaluate trackers using different paradigms. What’s more, to evaluate a tracker on different attributes, we also define four scenario attributes and 12 challenge attributes in the short-term tracking evaluation subset. By releasing LSOTB-TIR, we encourage the community to develop deep learning-based TIR trackers and evaluate them fairly and comprehensively. We evaluate and analyze 40 trackers on LSOTB-TIR to provide a series of baselines and give some insights and future research directions in TIR object tracking. Furthermore, we retrain several representative deep trackers on LSOTB-TIR, and their results demonstrate that the proposed training dataset significantly improves the performance of deep TIR trackers. Codes and dataset are available at https://github.com/QiaoLiuHit/LSOTB-TIR.

The current debate on volcanic-plutonic connection is centered on whether eficient liquid-crystal segregation dominates the evolution of a mushy reservoir to produce evolved, crystal-poor rhyolite and cumulate leftover. However, magma recharge may remarkably influence the evolution of a mushy reservoir and obscure the evidence of liquid-crystal segregation. This complexity poses a challenge to exploring the connection of volcanic-plutonic rocks. This study investigates the Qinzhou Bay granitic complex (~250–248 Ma) from South China, which contains crystal-poor (<19 vol%) peraluminous rhyolites and subsequent crystal-rich (28–54 vol%) porphyries. Although the rhyolite and porphyry units have a close spatio-temporal link, they do not share a fractionation trend and similar whole-rock Sr-Nd-O isotopic compositions; thus, a direct connection is not evidenced. We further present textural analyses, mineral and melt inclusion compositions, thermobarometry (the combination of Ti-in-zircon thermometer and Ti-in-quartz thermobarometer), and thermodynamic modeling to examine the alternative interpretations, i.e., the two units may have intrinsically independent origins or the connection of the two units has been obscured. For the rhyolite unit, thermobarometric results reveal a polybaric storage system consisting of middle (>600 ± 80 MPa) and upper (~150 ± 40 and ~60 ± 20 MPa) crustal reservoirs. Variations in quartz Fe content and chlorine-rich, metaluminous melt inclusions suggest that magma hybridization with less-evolved metaluminous magmas occurred at both crustal levels. In particular, the elevated Fe contents in the quartz population that crystallized at the shallowest level (~60 ± 20 MPa) suggest that recharge magmas were directly injected into the shallowest reservoir. Deviation of the whole-rock composition from the liquid evolution trend recorded in melt inclusions suggests a combined efect of magma mixing and crystal-melt segregation processes in upper crustal reservoirs. Thermodynamic modeling and mass balance calculations suggest that the whole-rock composition of the rhyolite could be reproduced by mixing between regionally exposed dacites and segregated melts at crystallinities of 50–60% (using parental magma represented by the least-evolved melt inclusion). For the porphyry unit, thermobarometric results reveal magma storage at middle (more than 450 ± 40 to 550 ± 40 MPa) and upper (110 ± 20 to 140 ± 20 MPa) crustal levels. The small-scale oscillatory zonation of plagioclase, the pervasive resorption of quartz and alkali feldspar, and the presence of peraluminous microgranular enclaves in the porphyries suggest a recharge event of metasediment-sourced magmas, triggering reactivation and convection of the reservoir. Autoclastic and overgrowth textures of quartz, plagioclase, and alkali feldspar phenocrysts and development of columnar jointing suggest that the reactivated porphyritic magmas ascended and emplaced at ultrashallow levels (~30 ± 10 MPa).

Existing methods for video-based person re- identification (ReID) mainly learn the appearance feature of a given pedestrian via a feature extractor and a feature aggregator. However, the appearance models would fail to learn a large inter-class variance when different pedestrians have similar appearances. Considering that different pedestrians have different walking postures and body proportions, we propose to learn the discriminative pose feature beyond the appearance feature for video retrieval. Specifically, we implement a two-branch architecture to separately learn the appearance feature and pose feature, and then concatenate them together for inference. To learn the pose feature, we first detect the pedestrian pose in each frame through an off-the-shelf pose detector, and construct a temporal graph using the pose sequence. We then exploit a recurrent graph convolutional network (RGCN) to learn the node embeddings of the temporal pose graph, which devises a global information propagation mechanism to simultaneously achieve the neighborhood aggregation of intra-frame nodes and message passing among inter-frame graphs. Finally, we propose a dual-attention method (DAM) consisting of node-attention and time-attention to obtain the temporal graph representation from the node embeddings, where the self-attention mechanism is employed to learn the importance of each node and each frame. We verify the proposed method on three video-based ReID datasets, i.e., Mars, DukeMTMC and iLIDS-VID, whose experimental results demonstrate that the learned pose feature can effectively improve the performance of existing appearance models.

High-speed maglev trains will play an important role in the high-speed transportation system in the near future. However, under the conditions of strong magnetic fields and continuous operation, the actuators of the high-speed maglev train suspension system are prone to lose partial effectiveness, which makes the suspension control problem challenging. In addition, most existing fault-tolerant control (FTC) methods for suspension systems require linearization around the equilibrium points during the controller design or stability analysis. Therefore, from a practical perspective, this study presents a novel nonlinear FTC strategy with adaptive compensation for high-speed maglev train suspension systems. First, a nonlinear dynamic model of the suspension system based on join-structure is established and the actuator failures are described. Then, a nonlinear fault-tolerant suspension control law with an adaptive update law is designed to achieve stable suspension against partial actuator failure. The Lyapunov theory and extended Barbalat lemma are utilized to rigorously prove the closed-loop asymptotic stability even if there is partial actuator failure, without any approximation to the original nonlinear dynamics. Finally, hardware experimental results are included to demonstrate the effectiveness of the proposed approach.

In the field of maritime surveillance, the global navigation satellite system (GNSS)-based passive radar has proven its potential for moving target detection (MTD), localization, and velocity estimation. The next stage is to investigate the possibility of obtaining the radar image of the moving ship for target recognition. However, the limited signal power budget of GNSS prevents the conventional inverse synthetic aperture radar technique that is based on target rotational motion and short observation time for GNSS-based passive radar imaging moving target. In this article, a two-stage imaging processing method relying on the target translational motion over a long observation time is proposed. The first stage confirms the presence of the target by a long-time MTD processing technique. In the second stage, based on the analysis of the Doppler history of the target signal in the slow-time domain, short-time Fourier transform and modified random sample consensus are combined to robustly estimate target velocity with reduced computation complexity. To obtain the focused bistatic image, azimuth compression is conducted by using the estimated target velocity. Finally, an image fusion operation is implemented to combine the bistatic images achievable from multiple satellites so that a multistatic image with high quality can be created. The effectiveness of the proposed method is confirmed by the real experimental results of three cargo ships illuminated by several satellites.

Thermal infrared (TIR) target tracking is susceptible to occlusion and similarity interference, which obviously affects the tracking results. To resolve this problem, we develop an Aligned Spatial-Temporal Memory network-based Tracking method (ASTMT) for the TIR target tracking task. Specifically, we model the scene information in the TIR target tracking scenario using the spatial-temporal memory network, which can effectively store the scene information and decrease the interference of similarity interference that is beneficial to the target. In addition, we use an aligned matching module to correct the parameters of the spatial-temporal memory network model, which can effectively alleviate the impact of occlusion on the target estimation, hence boosting the tracking accuracy even further. Through ablation study experiments, we have demonstrated that the spatial-temporal memory network and the aligned matching module in the proposed ASTMT tracker are exceptionally successful. Our ASTMT tracking method performs well on the PTB-TIR and LSOTB-TIR benchmarks contrasted with other tracking methods.

The interactions between tension twins and dislocations of ZK60 alloy under dynamic compression were investigated by electron backscattering diffraction and transmission electron microscopy. The samples were divided into two groups, where group-1 experienced primary deformation with different strains (4, 6, 9, and 12%) and group-2 experienced secondary deformation based on the primary deformation (4% + 8%, 6% + 6%, 9% + 3%) to reach the final deformation of 12%. The results reveal that the tension twins are the dominant deformation mechanism during the early stages of deformation, and the volume fraction of tension twins and dislocation density increase with the increase in deformation. Moreover, the twin boundaries and dislocations generated during the primary deformation become the resistance to the dislocations motion during the secondary deformation.

Lycopene is the main pigment in red-flesh citrus fruits, and its formation is a research hotspot. To explore the basis of lycopene accumulation in red-flesh mutants, we profiled the terpenoid metabolites. Compared with their respective wild types, Cara Cara (Cara) [and Red-Anliu (R-An)] oranges showed increased carotenoid and limonoid aglycone contents and decreased contents of abscisic acid (ABA) catabolites, monoterpenoid volatiles, and sesquiterpenoid volatiles. Cara contained less than half of the amount of ABA glucose ester (ABAGE), the main ABA derivative in oranges. Parallel lower transcript levels of NCEDand ABA glucosyltransferasein Cara were detected at the mature green stage. These results document the changes in terpenoid profiles in Cara and show that the red flesh of citrus color mutants is related to weak ABA catabolism, especially ABAGE, and decreased transcript levels of two genes encoding uridine diphosphate (UDP)-glycosyltransferases that catalyze ABAGE biosynthesis.

Lignin is a promising raw material for membrane fabrication due to its abundant oxygen-containing functional groups, branched molecular structure, and high abundance in the botanic field. In this report, polydopamine (PDA) coated carbon nanotubes (PCNTs) are synthesized. Then, a PCNT-incorporated lignin-based loose nanofiltration membrane is fabricated using quaternized lignin (QL) as a raw material, triphosgene (TP) as a crosslinker, and PCNTs as nanoadditive by interfacial polymerization. The results show that the PCNTs uniformly distribute in the QL-based separation layer providing quick diffusion paths for water molecules and salt ions. The coated PDA layer improves the chemical compatibility between the PCNTs and the separation layer. The incorporated PCNTs optimize the membrane microstructure and elevate the pure water permeance to 45.5 LMH bar-1. The membrane exhibits high rejections to various dyes (> 90%) and high salt permeation (> 90%), demonstrating an excellent dye/salt separation performance. Moreover, it has satisfactory capabilities for resisting dye adsorption and fouling, and its separation capability also has good tolerances to the acid, alkali, and active chlorine treatments. The quaternary ammonium ions and hydrophilic surface endow the membrane with favorable bactericidal property (99.3%) and bacteria-adhesion resistance. Besides, TP is a better crosslinker than trimesoyl chloride for optimizing the membrane microstructure and crosslinking degree in terms of dye/salt separation.

A high-performance lignin-based loose nanofiltration membrane is fabricated using quaternized lignin as a novel raw material and triphosgene as a crosslinker. It has high pure water permeance (33.8 L m−2h−1bar−1), dye rejections, and low salt rejections during the separation of dye/salt mixtures. The quaternary ammonium functional groups endow the membrane with excellent antibacterial properties. It also has satisfactory dye-fouling resistance and separation-performance stability. Triphosgene plays an important role in optimizing the membrane structure and enhancing permeability.A high-performance lignin-based loose nanofiltration membrane is fabricated using quaternized lignin as a novel raw material and triphosgene as a crosslinker. It has high pure water permeance, dye rejections, and low salt rejections during the separation of dye/salt mixtures. The quaternary ammonium functional groups endow the membrane with excellent antibacterial properties.

Arteriovenous grafts (AVGs) have emerged as the preferred option for constructing hemodialysis access in numerous patients. Clinical trials have demonstrated that decellularized vascular graft exhibits superior patency and excellent biocompatibility compared to polymer materials; however, it still faces challenges such as intimal hyperplasia and luminal dilation. The absence of suitable animal models hinders our ability to describe and explain the pathological phenomena above and in vivoadaptation process of decellularized vascular graft at the molecular level. In this study, we first collected clinical samples from patients who underwent the construction of dialysis access using allogeneic decellularized vascular graft, and evaluated their histological features and immune cell infiltration status 5 years post-transplantation. Prior to the surgery, we assessed the patency and intimal hyperplasia of the decellularized vascular graft using non-invasive ultrasound. Subsequently, in order to investigate the in vivoadaptation of decellularized vascular grafts in an animal model, we attempted to construct an AVG model using decellularized vascular grafts in a small animal model. We employed a physical–chemical–biological approach to decellularize the rat carotid artery, and histological evaluation demonstrated the successful removal of cellular and antigenic components while preserving extracellular matrix constituents such as elastic fibers and collagen fibers. Based on these results, we designed and constructed the first allogeneic decellularized rat carotid artery AVG model, which exhibited excellent patency and closely resembled clinical characteristics. Using this animal model, we provided a preliminary description of the histological features and partial immune cell infiltration in decellularized vascular grafts at various time points, including Day 7, Day 21, Day 42, and up to one-year post-implantation. These findings establish a foundation for further investigation into the in vivoadaptation process of decellularized vascular grafts in small animal model.

Wuhan was the epicentre of the COVID-19 outbreak in China. We aimed to determine the seroprevalence and kinetics of anti-SARS-CoV-2 antibodies at population level in Wuhan to inform the development of vaccination strategies.

H2/N2synthesis gas is of great importance for the ammonia industry, and its production process is both complex and energy consuming. Nitrous oxide (N2O) is a potent greenhouse gas (GHG) that also depletes the stratospheric ozone. Here, we propose a novel concept of a membrane reactor in which ammonia synthesis gas production, N2O decomposition, and water splitting are combined. In this process, oxygen produced from N2O decomposition and water splitting is removed by a dense oxygen permeable membrane, which then reacts with oxygen-consuming gas. This promising membrane reactor process has been successfully demonstrated by experiment. By optimizing the N2O and H2O flow rate, a complete N2O decomposition is achieved, and a synthesis gas with a molar ratio of H2/N2= 3:1 is produced at the same time. At 850 °C, a H2production rate of 14.85 mL/min and a N2production rate of 5.07 mL/min are derived from the membrane reactor with 100% N2O conversion and 29.60% H2O conversion. The reactor also exhibits excellent stability over a 300 h long-term test. This study will give some constructive guides to the subsequent work on process intensification through the membrane reactor and may lead to the development of green chemical production.

Current unsupervised feature selection methods cannot well select the effective features from the corrupted data. To this end, we propose a robust unsupervised feature selection method under the robust principal component analysis (PCA) reconstruction criterion, which is named the adaptive weighted sparse PCA (AW-SPCA). In the proposed method, both the regularization term and the reconstruction error term are constrained by the $\ell _{2,1}$ -norm: the $\ell _{2,1}$ -norm regularization term plays a role in the feature selection, while the $\ell _{2,1}$ -norm reconstruction error term plays a role in the robust reconstruction. The proposed method is in a convex formulation, and the selected features by it can be used for robust reconstruction and clustering. Experimental results demonstrate that the proposed method can obtain better reconstruction and clustering performance, especially for the corrupted data. [ABSTRACT FROM AUTHOR]

Human gesture is one of the most important ingredients of context information. In this paper, a novel gesture recognition framework based on tri-axis accelerometer mounted on a cell phone is proposed. Since the length of acceleration signals is variable according to different gestures and different subjects' input speed, most of recognition algorithms cannot be used. To solve this problem, we propose a novel resampling method by combining decimation and interpolation. After that, we propose 1D Gabor coefficients of acceleration signals as features. However, the dimensionality of Gabor feature space used in gesture recognition is very high. We adopt Boosting and a two-stage method PCA plus LDA to select and compress the Gabor feature. The classifier we used is fast multi-class support vector machine. The average recognition results of 17 complex gestures using the proposed Gabor feature are effective. The performance of experimental results shows that gesture-based interaction can be used as a novel human computer interaction for consumer electronics and mobile devices.

The insect larvae of Apriona germari, Apriona swainsoniand other closely related species are related to the insects that bore through the stems of the Caesalpinia decapetala. Because of therich nutritional as well as medicinal value of edible A. germari, it is often used to nourish the body, enhance immunity, and to treat ailments such as diarrhea. The soluble protein content of A. germariwas determined using Kjeldahl and Soxhlet extraction methods. The crude fat content and the content of various fatty acids was determined by the normalization method, normalization method refers to that after the fat in the sample is extracted by hydrolysis ether solution, it is saponified and methylated under alkaline conditions to generate fatty acids and methyl esters, which are analyzed by capillary column gas chromatography and quantitatively measured by area normalization method. Amino acid content determined using an amino acid analyzer after acid hydrolysis showed 17 kinds of amino acids. A diarrhea mouse model was established by injecting a suspension of Escherichia coliinto the abdominal cavity of mice. The mice were then fed with A. germarifodders at dose concentrations of 12%, 8%, 4% and 0%. The number of E. coliwas determined from the intestinal tracts of mice; the activities of antioxidant enzymes and digestive enzymes in the serum were determined. Eleven kinds of fatty acids were detected in A. germari; unsaturated fatty acids accounted for 71.87% of the total fatty acids. A. germariis an edible resource insect with high nutritional value. Mice fed with A. germarishowed an increase in the activity of antioxidant enzymes and digestive enzymes in the serum, which helped in the effective inhibition and reduction in the total number of intestinal bacteria and E. coli. The results indicate that A. germarihas high nutritional value; A. germarialso has a certain therapeutic effect on diarrhea caused by E. coliin mice.

Consumption of boiled crayfish may lead to Haff disease (HD), which is considered to result from an unidentified toxin, although the etiology is still obscure. Profiling of the N-glycome in HD would assist in deciphering the underlying molecular mechanism of the disease, whereas HD-associated glycosylation has never been explored. Herein, we enrolled 90 serum samples with HD patients and healthy controls from the Wuhan Center for Disease Control & Prevention between 2019 and 2020. N-glycome profiles of both serum and serum-derived immunoglobulin G (IgG) in HD were characterized by means of high-throughput-based orthogonal mass spectrometry. It was observed that HD is associated with an increase in the core fucosylation and mono-galactosylation of total serum glycoproteins. The serum level of IgG was found to serve as a good indicator for HD patients. In addition, differential galactosylation and sialylation of IgG were strongly correlated with HD. It was notable that the changes in the galactosylation and sialylation of IgG1 and IgG2 were subclass specific. Interestingly, altered sialylation and galactosylation of IgG2 or IgG3/4 strongly correlated with clinical markers for HD. Our study reveals the association of differential IgG N-glycosylation with HD, providing new insight into the etiology of this rare disease.

The Siamese tracking framework has attracted much attention due to its scalability and efficiency in recent years. However, it is less effective in recognizing arbitrary targets with various variations, especially in complex scenarios with background distractors and illumination variations. To this end, we propose a Siamese Residual Network to formulate the characteristics of a specific given target for visual tracking, consisting of an identity branch and a residual branch. The identity branch consists of a generic offline-trained similarity-matching network, which distinguishes the target from the background at the class level. To complement the identity branch for handling complex scenarios and dramatic target appearance variations, we develop a residual branch learned from the samples of exact target states and online distractors under the guidance of the identity branch. These two branches representing arbitrary targets with both class-level and sample-level features achieve accurate target localizations under complicated scenarios. In addition, we propose an adaptive KL-based scheme for updating the residual branch effectively by avoiding overfitting to a long-retained target appearance. Extensive experimental results on OTB-2013, OTB-2015, VOT2016, VOT-2018, VOT-2019, Temple-Color-128, and LaSOT show that the proposed method performs against state-of-the-art trackers. [ABSTRACT FROM AUTHOR]

With the fast development of social networks, privacy protection on social network data has already attracted more and more research attention. If an adversary has collected enough information by intentionally querying a social network, the privacy of social network data might be successfully inferred based on the previous query results, which is seldom investigated by most existing approaches. In this paper, we first formulate the correlated query results as the non-zero prior knowledge, then propose a novel differential privacy approach to enhance privacy of social network data from being inferred. The corresponding ϵ-difference is re-formulated as well as the calculation of the sensitivity between two “neighboring” social graphs. Particularly, the calculation of sensitivity is proposed for three widely adopted queries on social networks, i.e., node degree distribution query, cut set query and the shortest path query. To evaluate our approach, rigorous experiments have been performed on three real world data sets. The promising experimental results have demonstrated that the proposed approach is superior to the state-of-the-art privacy protection approaches with respect to data privacy and data utility. [ABSTRACT FROM AUTHOR]