Traffic signs recognition using dynamic-scale CNN

Real-time systems are often designed using a set of periodic tasks. Task periods are usually set by the system requirements, but deadlines and computation times can be modified in order to improve system performance. Sensitivity analysis in real-time systems has focused on changes in task computation times using fixed priority analysis. Only a few studies deal with the modification of deadlines in dynamic priority scheduling. The aim of this work is to provide a sensitivity analysis for task deadlines in the context of dynamic-priority, pre-emptive, uniprocessor scheduling. In this paper, we present a deadline minimisation method that achieves the maximum reduction. As undertaken in other studies concerning computation times, we also define and calculate the critical scaling factor for task deadlines. Our proposal is evaluated and compared with other works in terms of jitter. The deadline minimisation can be used to strongly reduce jitter of control tasks, in a real-time control application.

Today, more and more distributed computer applications are being modeled and constructed using real‐time principles and concepts. In 1989, the Object Management Group (OMG) formed a Real‐Time Special Interest Group (RT SIG) with the goal of extending the Common Object Request Broker Architecture (CORBA) standard to include real‐time specifications. This group's most recent efforts have focused on the requirements of dynamic distributed real‐time systems.One open problem in this area is resource access synchronization for tasks employing dynamic priority scheduling. This paper presents two resource synchronization protocols that the authors have developed which meet the requirements of dynamic distributed real‐time systems as specified by Dynamic Scheduling Real‐Time CORBA (DSRT CORBA). The proposed protocols can be applied to both Earliest Deadline First (EDF) and Least Laxity First (LLF) dynamic scheduling algorithms, allow distributed nested critical sections, and avoid unnecessary runtime overhead. In order to evaluate the performance of the proposed protocols, we analyzed each protocol's schedulability. Since the schedulability of the system is affected by numerous system configuration parameters, we have designed simulation experiments to isolate and illustrate the impact of each individual system parameter. Simulation experiments show the proposed protocols have better performance than one would realize by applying a schema that utilizes dynamic priority ceiling update. Copyright © 2004 John Wiley & Sons, Ltd.

The synthesis of face sketches from real face photos and its inverse process have a wide range of applications, such as digital entertainment and assisting in the investigation of criminal cases. However, due to the significant discrepancies in texture between photos and sketches, the transformation between them is still a challenging problem. Recently, the methods based on generative adversarial nets (GANs) have shown encouraging results in the transformation between images, especially the transformation from photos to sketches. However, most of them produce different deformations or blur in the key components of the face, which affects the realism of the synthetic image. To address this challenge, we propose a novel face photo-sketch synthesis algorithm based on multi-scale feature fusion to improve the structural integrity and texture fidelity of the synthetic image. Firstly, an encoder is utilized to extract the multi-scale coding features of the input image. Then the bottom coding features are passed by a dilated convolution module and transmitted to the decoder for decoding. In the decoding process, the decoding features of different scales and the coding features of corresponding scales are concatenated in the channel dimension to obtain the multi-scale coding and decoding fusion features. Finally, the coding and decoding fusion features of different scales are further fused at the output of the decoder and the final synthesis result is generated through a convolution layer. By combining the features of different scales in the encoding and decoding processes in the channel dimension at the same time, we can maintain good image structure and texture details and generate realistic face sketch/photo images. We verify the effectiveness of the proposed method in several challenging datasets. The quantitative and qualitative evaluation shows that the proposed method is superior to other state-of-the-art methods in generating face sketches (or photos) with high visual quality.

基于动态优先级算法的RGV调度策略

Static priority scheduling is popular for traffic scheduling in ATM switches because it is less costly than dynamic priority scheduling while being sensitive to the delay constraints of connections. We study delay computation and priority assignment problems in an ATM network with static priority scheduling. Given an ATM network with arbitrary topology, it is possible that the traffic on it may become unstable (i.e., packet delays become unbounded) due to the potential cyclic dependency of the traffic. An unstable network is definitely unacceptable for many delay sensitive applications. We start by formally deriving a simple condition under which the network is guaranteed to be stable. We then develop a numerical method to compute worst case end to end delays in an ATM network with arbitrary topology. Convergence of the method is formally proved and a closed form for the computing error is obtained. Despite its advantages, static priority scheduling remains sensitive to proper priority assignment. We describe two simple priority assignment methods, which we show to outperform other commonly used methods.

This article involves the study of a multiple finite source queueing model with a single server and dynamic, nonpreemptive priority service discipline. The input to the queue is comprised of customers from multiple finite sources. The time which the customers spend at the corresponding sources are exponentially distributed. The service times of the customers can follow exponential, Erlang, or hyperexponential probability density function, with the same mean regardless of the class. Using an extension of mean value analysis, a recursive algorithm is developed to obtain approximate values of the mean waiting time in queues for each priority class. The mean number of waiting customers and the server utilization of each priority class can be obtained using the result of this recursive algorithm and Little's formula. Numerical examples are presented to illustrate the methodology. The algorithm developed in this article is validated using simulation.

Feasibility tests for hard real-time systems provide information about the schedulability of a set of tasks. However, this information is a yes or no answer whether the task set achieves the test or not. From the system design point of view, it would be useful to have more information, for example, how much can one vary some task parameters, such as computation time, without jeopardizing the system feasibility. The aim of the work is to provide a method to determine how much a task can increase its computation time, maintaining the system feasibility under a dynamic priority scheduling. This extra time can be determined not only in all the task activations, but in n of a window of m task invocations. This is what we call a window-constrained execution time system. In control applications, this information can be used to execute supervision activities, such as model updating which is not required to be executed in all the periods, or to determine new controller parameters for the current operating conditions. In fault tolerance, this information allows us to recover n faults in m activations.

Cyberbullying in Online Social Networks (OSNs) has grown to be a serious problem among teenagers. While a considerable amount of research has been conducted focusing on designing highly accurate classifiers to automatically detect cyberbullying instances in OSNs, two key practical issues remain to be worked upon, namely scalability of a cyberbullying detection system and timeliness of raising alerts whenever cyberbullying occurs. These two issues form the motivation of our work. We propose a multi-stage cyberbullying detection solution that drastically reduces the classification time and the time to raise alerts. The proposed system is highly scalable without sacrificing accuracy and highly responsive in raising alerts. The design is comprised of two novel components, a dynamic priority scheduler and an incremental classification mechanism. We have implemented this solution, and using data obtained from Vine, we conducted a thorough performance evaluation to demonstrate the utility and scalability of each of these components. We show that our complete solution is significantly more scalable and responsive than the current state of the art.

In this thesis, we address the issue of efficiently providing a range of per-flow QoS guarantees in the Internet paradigm. After a review of the relevant work on QoS guarantees, we propose a Flow-state-dependent Dynamic Priority Scheduling (FDPS) algorithm that provides Service Curves assurance. In this scheduling scheme, admitted QoS-based flows are protected yet allowing best effort traffic to co-exist in a multiservice network environment. The proposed FDPS scheme performs packet monitoring, marking, scheduling and discarding. With its fine granularity in packet marking, individual packets are forwarded and scheduled in a controlled and orderly manner. By means of mathematical analysis, we show that the FDPS algorithm realizes the arrival and service curves required in Service Curves and thus allows the theory of Service Curves to be applicable in a real network for per flow QoS provisioning. We also present an example for the admission control policing associated with FDPS. Compared with existing scheduling algorithms for QoS provisioning, FDPS achieves a tight end-to-end delay for QoS flows in a multi-service packet switching network. The complexity of this algorithm is independent of the number of flow, facilitating high scalability and possible implementation. Experimental results based on NS-2 validate the theoretical analysis and demonstrate the superiority of the proposed scheme.

As the acquisition of very high resolution (VHR) images becomes easier, the complex characteristics of VHR images pose new challenges to traditional machine learning semantic segmentation methods. As an excellent convolutional neural network (CNN) structure, U-Net does not require manual intervention, and its high-precision features are widely used in image interpretation. However, as an end-to-end fully convolutional network, U-Net has not explored enough information from the full scale, and there is still room for improvement. In this study, we constructed an effective network module: residual module under a multisensory field (RMMF) to extract multiscale features of target and an attention mechanism to optimize feature information. RMMF uses parallel convolutional layers to learn features of different scales in the network and adds shortcut connections between stacked layers to construct residual blocks, combining low-level detailed information with high-level semantic information. RMMF is universal and extensible. The convolutional layer in the U-Net network is replaced with RMMF to improve the network structure. Additionally, the multiscale convolutional network was tested using RMMF on the Gaofen-2 data set and Potsdam data sets. Experiments show that compared to other technologies, this method has better performance in airborne and spaceborne images.

The binarization step for old documents is still a challenging task even though many hand-engineered and deep learning algorithms have been offered. In this research work, we address foreground and background segmentation using a convolutional autoencoder network with 3 supporting components. The assessment of several hyper-parameters including the window size, the number of convolution layers, the kernel size, the number of filters as well as the number of encoder-decoder layers on the network is conducted. In addition, the skip connections approach is considered in the decoding procedure. Moreover, we evaluated the summation and concatenation function before the up-sampling process to reuse the previous low-level feature maps and to enrich the decoded representation. Based on several experiments, we determined that kernel size, the number of filters, and the number of encoder-decoder blocks have a little impact in term of binarization performance. While the window size and multiple convolutional layers are more impactful than other hyper-parameters. However, they require more storage and may increase computation costs. Moreover, a careful embedding of batch normalization and dropout layers also provides a contribution to handle overfitting in the deep learning model. Overall, the multiple convolutional autoencoder network with skip connection successfully enhances the binarization accuracy on old Sundanese palm leaf manuscripts compared to preceding state of the art methods.

In the unlicensed band, the notion of primary user and secondary user (to implement cognitive radio) is not explicit. By dynamic priority assignment we propose to implement cognitive radio in the unlicensed band. In time critical events, the data which is most important, has to be given the time slots. Wireless Sensor nodes in our case are considered to be mobile, and hence make it difficult to prioritize one over another. A node may be out of the reach of the cluster head or base station by the time it is allotted a time slot and hence mobility is a constraint. With the data changing dynamically and factors such as energy and mobility, which are major constraints, assigning priority to the nodes becomes difficult. In this chapter, we have discussed how Wireless Sensor Networks are able to allocate priorities to nodes in the unlicensed band with multiple parameters being posed. We have done simulations on NS-2 and have shown the implementation results.

Combinatorial explosion is a challenge for many analysis problems in the theory of hard real-time systems. One of these problems is static priority schedulability of workload models which are more expressive than the traditional periodic task model. Different classes of directed graphs have been proposed in recent years to model structures like frames, branching and loops. In contrast to dynamic priority schedulers with pseudo-polynomial time analysis methods, static priority schedulability has been shown to be intractable since it is strongly coNP-hard already for the relatively simple class of cyclic digraphs. The core of this problem is the necessity to combine different behaviors of the participating tasks. We introduce a novel iterative approach to efficiently cope with this combinatorial explosion, called combinatorial abstraction refinement. In combination with other techniques it significantly reduces exponential growth of run-time for most inputs. We apply the method to static task priorities and demonstrate that a prototype implementation outperforms the state-of-the art pseudo-polynomial analysis for dynamic priority feasibility. It further shows better scaling behavior for randomly generated problem instances. We extend the approach to non-preemptive schedulers as well as static job-type priorities where jobs of different types in the same task may be assigned different static priorities. Finally, we provide a general, abstract formulation of the algorithm, since we believe that this method can be applicable to a variety of combinatorial problems in the theory of real-time systems with certain abstraction structures.

In the unlicensed band, the notion of primary user and secondary user (To implement cognitive radio) is not explicit. By dynamic priority assignment the authors propose to implement cognitive radio in the unlicensed band. In time critical events, the data which is most important, has to be given the time slots. Wireless Sensor nodes in the authors’ case are considered to be mobile, and hence make it difficult to prioritize one over another. A node may be out of the reach of the cluster head or base station by the time it is allotted a time slot and hence mobility is a constraint. With the data changing dynamically and factors such as energy and mobility, which are major constraints, assigning priority to the nodes becomes difficult. In this paper, the authors have discussed about how Wireless Sensor Networks are able to allocate priorities to nodes in the unlicensed band with multiple parameters being posed. They have done simulations on NS-2 and have shown the implementation results.

The FPGA-based accelerators have shown great potential in improving the performance of CNN inference. However, the existing FPGA-based approaches suffer from a low Compute Unit (CU) efficiency due to a large amount of invalid computation, thus leading to a great performance degradation. In this paper, we find that no single CU can perform the best across all the convolutional layers (Conv-layers). To this end, we propose multiple sizes of compute unit (MSCU), an approach to accelerating CNN inference on FPGAs. The key idea of MSCU is to select the best combination of CUs via dynamic programming scheduling for each Conv-layer, and then assembles each Conv-layer combination into a computing solution of the given CNN to deploy on FPGAs. The experimental results show that MSCU can achieve 3.36E-03 GOPs/Slice on Xilinx Zynq ZU3EG for performance density which is 4.29x higher than other approaches.