Search

The intellectual property of deep networks can be easily “stolen” by surrogate model attack. There has been significant progress in protecting the model IP in classification tasks. However, little attention has been devoted to the protection of image processing models. By utilizing consistent invisible spatial watermarks, the work (Zhang et al. 2020) first considered model watermarking for deep image processing networks and demonstrated its efficacy in many downstream tasks. Its success depends on the hypothesis that if a consistent watermark exists in all prediction outputs, that watermark will be learned into the attacker's surrogate model. However, when the attacker uses common data augmentation attacks (e.g., rotate, crop, and resize) during surrogate model training, it will fail because the underlying watermark consistency is destroyed. To mitigate this issue, we propose a new watermarking methodology, “structure consistency”, based on which a new deep structure-aligned model watermarking algorithm is designed. Specifically, the embedded watermarks are designed to be aligned with physically consistent image structures, such as edges or semantic regions. Experiments demonstrate that our method is more robust than the baseline in resisting data augmentation attacks. Besides that, we test the generalization ability and robustness of our method to a broader range of adaptive attacks.

Transformer based methods have achieved great success in image inpainting recently. However, we find that these solutions regard each pixel as a token, thus suffering from an information loss issue from two aspects: 1) They downsample the input image into much lower resolutions for efficiency consideration. 2) They quantize $256^{3}$2563 RGB values to a small number (such as 512) of quantized color values. The indices of quantized pixels are used as tokens for the inputs and prediction targets of the transformer. To mitigate these issues, we propose a new transformer based framework called “PUT”. Specifically, to avoid input downsampling while maintaining computation efficiency, we design a patch-based auto-encoder P-VQVAE. The encoder converts the masked image into non-overlapped patch tokens and the decoder recovers the masked regions from the inpainted tokens while keeping the unmasked regions unchanged. To eliminate the information loss caused by input quantization, an Un-quantized Transformer is applied. It directly takes features from the P-VQVAE encoder as input without any quantization and only regards the quantized tokens as prediction targets. Furthermore, to make the inpainting process more controllable, we introduce semantic and structural conditions as extra guidance. Extensive experiments show that our method greatly outperforms existing transformer based methods on image fidelity and achieves much higher diversity and better fidelity than state-of-the-art pluralistic inpainting methods on complex large-scale datasets (e.g., ImageNet).

Deep Neural Network classifiers are vulnerable to adversarial attacks, where an imperceptible perturbation could result in misclassification. However, the vulnerability of DNN-based image ranking systems remains under-explored. In this paper, we propose two attacks against deep ranking systems, i.e., Candidate Attack and Query Attack, that can raise or lower the rank of chosen candidates by adversarial perturbations. Specifically, the expected ranking order is first represented as a set of inequalities. Then a triplet-like objective function is designed to obtain the optimal perturbation. Conversely, an anti-collapse triplet defense is proposed to improve the ranking model robustness against all proposed attacks, where the model learns to prevent the adversarial attack from pulling the positive and negative samples close to each other. To comprehensively measure the empirical adversarial robustness of a ranking model with our defense, we propose an empirical robustness score, which involves a set of representative attacks against ranking models. Our adversarial ranking attacks and defenses are evaluated on MNIST, Fashion-MNIST, CUB200-2011, CARS196, and Stanford Online Products datasets. Experimental results demonstrate that our attacks can effectively compromise a typical deep ranking system. Nevertheless, our defense can significantly improve the ranking system's robustness and simultaneously mitigate a wide range of attacks.

Image matting is a fundamental and challenging problem in computer vision and graphics. Most existing matting methods leverage a user-supplied trimap as an auxiliary input to produce good alpha matte. However, obtaining high-quality trimap itself is arduous. Recently, some hint-free methods have emerged, however, the matting quality is still far behind the trimap-based methods. The main reason is that, some hints for removing semantic ambiguity and improving matting quality are essential. Apparently, there is a trade-off between interaction cost and matting quality. To balance performance and user-friendliness, we propose an improved deep image matting framework which is trimap-free and only needs sparse user click or scribble interaction to minimize the needed auxiliary constraints while still allowing interactivity. Moreover, we introduce uncertainty estimation that predicts which parts need polishing and conduct uncertainty-guided refinement. To trade off runtime against refinement quality, users can also choose different refinement modes. Experimental results show that our method performs better than existing trimap-free methods and comparably to state-of-the-art trimap-based methods with minimal user effort. Finally, we demonstrate the extensibility of our framework to video human matting without any structure modification, by adding optical flow-based sparse hint propagation and temporal consistency regularization imposed on the single frame.

Background: Nutcracker syndrome is a disease characterized by complex symptoms, making its diagnosis challenging and often delayed, often resulting in a painful experience for the patients. Objective: This study aimed to investigate the pathogenesis of nutcracker syndrome through the perspective of hemodynamics by simulating blood flow with varying compression degrees of the left renal vein. Methods: 3D patient-specific vascular models of the abdominal aorta, superior mesenteric artery and left renal vein were constructed based on CT images of patients suspected of having nutcracker syndrome. A hemodynamic simulation was then conducted using computational fluid dynamics to identify the correlation between alterations in hemodynamic parameters and varying degrees of compression. Results: The study indicated the presence of an evident gradient in velocity distribution over the left renal vein with relatively high degrees of stenosis (α ≤ 50°), with maximum velocity in the central region of the stenosis. Additionally, when the compression degree of the left renal vein increases, the pressure distribution of the left renal vein presents an increasing number of gradient layers. Furthermore, the wall shear stress shows a correlation with the variation of blood flow velocity, i.e., the increase of wall shear stress correlates with the acceleration of the blood flow velocity. Conclusions: Using computational fluid dynamics as a non-invasive instrument to obtain the hemodynamic characteristics of nutcracker syndrome is feasible and could provide insights into the pathological mechanisms of the nutcracker syndrome supporting clinicians in diagnosis.  Graphical abstract:

Representing multimodal behaviors is a critical challenge for pedestrian trajectory prediction. Previous methods commonly represent this multimodality with multiple latent variables repeatedly sampled from a latent space, encountering difficulties in interpretable trajectory prediction. Moreover, the latent space is usually built by encoding global interaction into future trajectory, which inevitably introduces superfluous interactions and thus leads to performance reduction. To tackle these issues, we propose a novel Interpretable Multimodality Predictor (IMP) for pedestrian trajectory prediction, whose core is to represent a specific mode by its mean location. We model the distribution of mean location as a Gaussian Mixture Model (GMM) conditioned on sparse spatio-temporal features, and sample multiple mean locations from the decoupled components of GMM to encourage multimodality. Our IMP brings four-fold benefits: 1) Interpretable prediction to provide semantics about the motion behavior of a specific mode; 2) Friendly visualization to present multimodal behaviors; 3) Well theoretical feasibility to estimate the distribution of mean locations supported by the central-limit theorem; 4) Effective sparse spatio-temporal features to reduce superfluous interactions and model temporal continuity of interaction. Extensive experiments validate that our IMP not only outperforms state-of-the-art methods but also can achieve a controllable prediction by customizing the corresponding mean location.

Effectively tackling the problem of temporal action localization (TAL) necessitates a visual representation that jointly pursues two confounding goals, i.e., fine-grained discrimination for temporal localization and sufficient visual invariance for action classification. We address this challenge by enriching the local, global and multi-scale contexts in the popular two-stage temporal localization framework. Our proposed model, dubbed ContextLoc++, can be divided into three sub-networks: L-Net, G-Net, and M-Net. L-Net enriches the local context via fine-grained modeling of snippet-level features, which is formulated as a query-and-retrieval process. Furthermore, the spatial and temporal snippet-level features, functioning as keys and values, are fused by temporal gating. G-Net enriches the global context via higher-level modeling of the video-level representation. In addition, we introduce a novel context adaptation module to adapt the global context to different proposals. M-Net further fuses the local and global contexts with multi-scale proposal features. Specially, proposal-level features from multi-scale video snippets can focus on different action characteristics. Short-term snippets with fewer frames pay attention to the action details while long-term snippets with more frames focus on the action variations. Experiments on the THUMOS14 and ActivityNet v1.3 datasets validate the efficacy of our method against existing state-of-the-art TAL algorithms.

Weakly-supervised temporal action localization (W-TAL) aims to classify and localize all action instances in untrimmed videos under only video-level supervision. Without frame-level annotations, it is challenging for W-TAL methods to clearly distinguish actions and background, which severely degrades the action boundary localization and action proposal scoring. In this paper, we present an adaptive two-stream consensus network (A-TSCN) to address this problem. Our A-TSCN features an iterative refinement training scheme: a frame-level pseudo ground truth is generated and iteratively updated from a late-fusion activation sequence, and used to provide frame-level supervision for improved model training. Besides, we introduce an adaptive attention normalization loss, which adaptively selects action and background snippets according to video attention distribution. By differentiating the attention values of the selected action snippets and background snippets, it forces the predicted attention to act as a binary selection and promotes the precise localization of action boundaries. Furthermore, we propose a video-level and a snippet-level uncertainty estimator, and they can mitigate the adverse effect caused by learning from noisy pseudo ground truth. Experiments conducted on the THUMOS14, ActivityNet v1.2, ActivityNet v1.3, and HACS datasets show that our A-TSCN outperforms current state-of-the-art methods, and even achieves comparable performance with several fully-supervised methods.

Semantic image synthesis, translating semantic layouts to photo-realistic images, is a one-to-many mapping problem. Though impressive progress has been recently made, diverse semantic synthesis that can efficiently produce semantic-level or even instance-level multimodal results, still remains a challenge. In this article, we propose a novel diverse semantic image synthesis framework from the perspective of semantic class distributions, which naturally supports diverse generation at both semantics and instance level. We achieve this by modeling class-level conditional modulation parameters as continuous probability distributions instead of discrete values, and sampling per-instance modulation parameters through instance-adaptive stochastic sampling that is consistent across the network. Moreover, we propose prior noise remapping, through linear perturbation parameters encoded from paired references, to facilitate supervised training and exemplar-based instance style control at test time. To further extend the user interaction function of the proposed method, we also introduce sketches into the network. In addition, specially designed generator modules, Progressive Growing Module and Multi-Scale Refinement Module, can be used as a general module to improve the performance of complex scene generation. Extensive experiments on multiple datasets show that our method can achieve superior diversity and comparable quality compared to state-of-the-art methods. Codes are available at https://github.com/tzt101/INADE.git.

An improved slotted end connection was proposed by setting a thick-walled rectangular hollow section (RHS) at both ends of the member. A total of eight conventional and improved slotted connection specimens were designed to study the performance of the improved slotted connection under axial tensile loading, with the investigated parameter being the fillet length (Lw). The test results reveal that the conventional connections suffer from shear tear-out failure or net section crack failure, while the improved connections all undergo gross section cracks as Lwvaries. The load-bearing capacity of the improved connection is 68.3%, 13.7%, 12.2%, and 10.1% higher than that of the corresponding conventional connection, respectively. According to the parametric finite element analysis, the local stress concentration around the slot can be effectively reduced by setting a thick-walled section, indicating that the proposed connection was acceptable in terms of axial tensile load. Furthermore, the formula for calculating the load-bearing capacity of the improved connection was derived from that of the conventional connection and was verified by test results.

In order to improve the energy efficiency of shot blasting impact descaling, a three-dimensional finite element impact descaling model was established. Based on the finite element model, the cracking behavior of the scale layer on hot rolled strip from different impacts angles was simulated. The results of finite element calculation and theoretical analysis show that: (1)Under the premise of constant velocity, the descaling area increases with the increase of impact angle, but the increasing rate tends to be moderate. (2)The depth of the impact tunnel and the residual compressive stress surface (-200 MPa) increase as the impact angle goes bigger. The ideal range of impact angle for shot blasting descaling should be 60°-75°.

A range of indolizine smoothly underwent visible-light-induced intermolecular [3+2] annulations with internal alkynes to afford pyrrolo[2,1,5-cd]indolizine in good to excellent yields with high regioselectivity. Through this cascade reaction, a series of fluoroactive fused indolizines with a large π-system were conveniently synthesized. The usage of visible light as energy source with air as a stoichiometric oxidant under simple conditions makes this process attractive and practical.

Action detection and recognition are popular subjects of research in the field of computer vision. The task of action detection can be regarded as the sum of action location and recognition. Action features described by using information concerning the human skeleton have the advantages of robustness against external factors and requiring a small amount of calculation. This study proposes a skeleton-based action analysis model based on a recurrent neural network framework. The model learns action features by modelling static and dynamic features of skeleton joints and the importance of different video frames by introducing an attention module. For action location, conditional random field loss function is introduced to establish the context dependency of output labels. In the aspect of action recognition, the hierarchical training mechanism with triple loss models action features at coarse-grained and fine-grained levels. The authors’ proposed method delivers state-of-the-art results on action location and recognition tasks.

We investigate new types of gap solitons in a periodic parity-time (PT)-symmetric lattice with fractional-order diffraction. Both the fundamental and dipole solitons in the first and second gaps are discussed. It is found that fractional-order diffraction can not only stabilize low-power dipole PT solitons in the first gap under focusing nonlinearity, but also help to get stable dipole PT solitons in the second gap under defocusing nonlinearity. Additionally, the influence of the strength ${w_i}$w_i of the gain–loss component on the properties of solitons is also analyzed. It is shown that increasing ${w_i}$w_i is unfavorable to the stability of fractional fundamental solitons, especially for the second gap, while for fractional dipole solitons, the increase of ${w_i}$w_i may lead to their destabilization in the first gap, but stabilization in the second gap.

Two copper-rich open-framework sulfides, K4Cu8Ge3S12(1) and Rb4Cu8Ge3S12(2), have been synthesized under solvothermal conditions. Compounds 1and 2are isostructural and contain icosahedral [Cu8S12]16−clusters as basic building blocks. These clusters are primitive cubic packed and connect to one another by discrete Ge4+ions to generate 3D copper-rich Cu−Ge−S framework and form 3D channels along ⟨100⟩ directions where the alkali metal cations reside. These two open-framework sulfides crystallize in cubic perovskite structure.

The high-elastic Cu-20.0Ni-5.0Sn-0.25Zn-0.22Mn alloy was designed and prepared using a 830°C/2h+850°C/2h dual-stage homogenization annealing process. The true stress-strain curves of well annealed Cu-20.0Ni-5.0Sn-0.25Zn-0.22Mn alloy were plotted as a function of compressive temperature and strain rate. The results showed that the stage division of thermal compression deformation is temperature dependent, which involves work hardening, dynamic recovery and recrystallization stages. The maximum value of the true stress increases as the strain rate gets larger, but decreases as the deformation temperature rises. The high temperature compression deformation process of the alloy is a thermal activated process, and the corresponding constitutive equation of the true stress-strain is established.

Nanoscale zero-valent irons (nZVIs) are easy to agglomerate and deactivate, which limited the practical applications in the subsurface for in situ remediation of contaminated groundwater due to poor radius of influence. Here, we utilized the dual support effect of coupling surface-modified carbon black with a surfactant (carboxymethyl cellulose (CMC) or sodium dodecyl sulfate (SDS)) to manipulate the charge, electron, and geometric structure of the interface, the obtained Fe/CB/CMC (wt%=1:1:2) and Fe/CB/SDS (wt%=1:2:1) suspensions exhibited superior removal capacity toward As(III) with capacities of 80.7 and 90.7 mg g-1, respectively. In a clay medium, 25.2% and 18.0% of nZVIs in Fe/CB/CMC-1:1:2 and Fe/CB/SDS-1:2:1 can break through the column while approximately 100% are retained in bare nZVIs under the same conditions. In addition, input velocity and ionic strength demonstrate a positive correlation with the retention of nZVIs in the Fe/CB/CMC-112 and a slight impact on the deposition of nZVIs in the Fe/CB/SDS-121 due to the rheology and charge differences between CMC and SDS. This study demonstrates the feasibility of combining anionic surfactants and carbon supports to enhance the reactivity and mobility of nZVIs for in situ remediation.

In this work, a novel process integration scheme for p-MTJ devices’ passivation and contacting was proposed. The method can efficiently protect the ferromagnetic metals and the magnesium oxide which are the key building block of p-MTJs, and effectively make electrical contact with the interconnect metals for p-MTJs. The scheme consists of passivation of p-MTJs with dual dielectrics - silicon nitride and silicon oxide, followed by planarization and selective wet etch. The proposed integration scheme was successfully demonstrated with 80 nm size p-MTJ devices.

Grey Wolf Optimiser (GWO) is the recently proposed meta-heuristic algorithm inspired by grey wolves. The original version of GWO has been proposed to solve problems in continuous search spaces. However, there are many optimisation problems in discrete binary search spaces such as feature selection and dimensionality reduction. We applied eight transfer functions in our BGWO and compared their performance. The test functions whose minimum points are gotten at 0 are selected as bench functions but also those functions their minimum value are gotten at 1.

It has been reported that 10 to 15% of young normogonadotrophic women show suboptimal response to standard long protocols. Letrozole (LE), an aromatase inhibitor, was shown to improve ovarian sensitivity to follicle stimulating hormone (FSH) and follicular response to gonadotrophin treatment in poor ovarian response patients. We reasoned that it might be possible to utilize LE in young normogonadotrophic patients with unexpected hypo-response in standard gonadotropin-releasing hormone agonist long protocol. A total of 652 patients defined as normogonadotrophic patients with unexpected hypo-response were divided into 2 groups, the +LE group and the +Gn group. +LE group: A fixed daily dose of 2.5 mg of LE was added on day 8 of stimulation. +Gn group: A fixed daily dose of 75 U of human menopausal gonadotrophin was added on day 8 of stimulation. The primary outcome measures were the number of oocytes obtained, fertilization rate, days of stimulation, and total FSH dosage. The secondary outcome measures were the implantation rate and ongoing pregnancy rate. There were no significant differences in the clinical and hormonal characteristics between the 2 groups. A shorter duration of stimulation and a lower dosage of recombinant FSH consumption on the day of human chorionic gonadotropin administration were all observed in the +LE group. Patients who received LE therapy showed a higher number of oocytes obtained and significantly higher fertilization rates. The implantation rate and ongoing pregnancy rate were comparable in both groups. LE significantly improves the number of oocytes obtained in patients with suboptimal response to standard gonadotropin-releasing hormone agonist long protocol.

The soils in seasonal frozen regions freeze and thaw frequently, causing severe frost heave and thaw settlement problems, which bring challenges to piles of photovoltaic stents. In this paper, laboratory tests are conducted with different types of screw piles under freezing conditions, with also using smooth piles for contrast. The aim is to simulate the freezing process of screw piles according to practical working conditions based on the similarity principle. Internal thermal resistance is ignored. The change laws of temperature, displacement, as well as the influence factors of types of screw piles are analyzed. The results indicate that: with a freezing depth of 30 cm, which is half of the pile length, large-double-bladed screw piles perform the best in antijacking-up, while all-bladed screw piles perform the worst, independent of the types of soil samples tested. The fitting relationship between jacking-up displacement and freezing depth is also proposed for each type of pile. Results obtained can provide an important reference to site construction in seasonal frozen regions.

This study provides a definition for urban big data while exploring its features and applications of China's city intelligence. The differences between city intelligence in China and the “smart city” concept in other countries are compared to highlight and contrast the unique definition and model for China's city intelligence in this paper. Furthermore, this paper examines the role of urban big data in city intelligence by showing that it not only serves as the cornerstone of this trend as it also plays a core role in the diffusion of city intelligence technology and serves as an inexhaustible resource for the sustained development of city intelligence. This study also points out the challenges of shaping and developing of China's urban big data. Considering the supporting and core role that urban big data plays in city intelligence, the study then expounds on the key points of urban big data, including infrastructure support, urban governance, public services, and economic and industrial development. Finally, this study points out that the utility of city intelligence as an ideal policy tool for advancing the goals of China's urban development. In conclusion, it is imperative that China make full use of its unique advantages—including using the nation's current state of development and resources, geographical advantages, and good human relations—in subjective and objective conditions to promote the development of city intelligence through the proper application of urban big data.

A large number of beacon nodes will be applied to the weighted centroid localization algorithm, but too much hardware facilities will affect the efficiency of the algorithm. Using the cosine theorem, On the basis of the original beacon nodes, add some virtual static beacon nodes to participate in orientation, It can reduce the hardware cost required for positioning. In addition, the math on and analysis of the localization algorithm, by derivative principle explaining the weight coefficient and RSSI ranging influence on localization algorithm. Simulation experiments show that the weighted centroid algorithm based on VSBN needed fewer beacon nodes, high positioning accuracy.

Against path optimization problem for wireless sensor network, this paper proposes a path optimization strategy for wireless sensor network based on improved shuffled frog leaping algorithm. The shuffled frog leaping algorithm was used as wireless sensor network path optimization main frame, gauss mutation and opposition-based learning were used to overcome the defects of easily trapping into local optimum and low accuracy computation. Simulation results show that the route optimization mechanism can effectively prolongs the network lifetime,reduces energy consumption, and improves the overall network performance.

This paper introduces the basic principles and calculation methods for the correlation dimension and Kolmogorov entropy. By calculating the correlation dimension and Kolmogorov entropy when the gear is under different working conditions, we can analyze the inherent relationship between the two in depicting of the running condition of the gearbox. The result shows that,the correlation dimension and Kolmogorov entropy have a good consistency in the description of working status of gearbox. This conclusion not only provides a good basis for the gearbox running condition judgment and fault diagnosing, but can also provide the experimental basis for the chaotic characteristic parameters selection in state monitoring and fault diagnosing.

To inspect and forecast the failure of the gearbox equipment, the dynamic models of tooth surface flake are established. The thesis explores the gearbox nonlinear dynamic characteristic caused by tooth surface flake, and researches the variant joggling stiffness caused by flake failure,analyzes the primary resonance of the flake failure. By the singularity analysis, the fault causes the bifurcation mode of the gear system, which shows effect of the generation and evolution of the fault on the dynamic response of the system. Using the experimental device,the theoretical models are tested and verified.

This paper is made after the dynamic test on Jing hang Channel Bridge in Wujiang, Jiangsu Province. Based on the test on this bridge, we get dynamic characteristics of the bridge span structure. We will analysis the vibration generated from the steel and concrete box girders of the bridge while the car is running on it. Based on the different characteristics of vibration from the two kinds of girders, we can analysis the influence of this hybrid girder on driving comfort.

Objective To investigate the expression and roles of p38 mitogen-activated protein kinase (p38 MAPK) in LPS-induced acute lung injury (ALI) in mice. Methods The ALI mice models were set up by intraperineal injection of lipopolysaccharide (LPS). The expressions of p38 MAPK in lung tissues were detected by immunohistochemistry and Western-blot. Results The positive expressions of p38 MAPK distribute mainly in infiltrative inflammatory cells, epithelial cells and endothelial cells. And the level of expression of phosphated p38 MAPK in ALI group were higher obviously than that in the control group, and it reached a peak after two hours. Conclusion p38 MAPK signaling pathway was triggered by ALI induced by endotoxin.

Magnetic fluid is the key core of magnetic fluid sensor. Here magnetic model of magnetic fluid is analyzed. The magnetic curve of magnetic fluid is divided into 3 parts and the liner area is presented to make magnetic fluid sensor more sensitively. The magnetic force of magnetic fluid in magnetic field is calculated and analyzed. The magnetic permeability of magnetic fluid in sensor could be treated as a constant in the linear area of magnetization curve. And the magnetic force of magnetic fluid in magnetic field could be ignored if the length to diameter ratio of the solenoid is large enough.