摘要 :
This paper reviews the AIM 2020 challenge on extreme image inpainting. This report focuses on proposed solutions and results for two different tracks on extreme image inpainting: classical image inpainting and semantically guided ...
展开
This paper reviews the AIM 2020 challenge on extreme image inpainting. This report focuses on proposed solutions and results for two different tracks on extreme image inpainting: classical image inpainting and semantically guided image inpainting. The goal of track 1 is to inpaint large part of the image with no supervision. Similarly, the goal of track 2 is to inpaint the image by having access to the entire semantic segmentation map of the input. The challenge had 88 and 74 participants, respectively. 11 and 6 teams competed in the final phase of the challenge, respectively. This report gauges current solutions and set a benchmark for future extreme image inpainting methods.
收起
摘要 :
Collaborative filtering (CF) has become one of the most popular and widely used methods in recommender systems, but its performance degrades sharply in practice due to the sparsity and bias of the real-world user feedback data. In...
展开
Collaborative filtering (CF) has become one of the most popular and widely used methods in recommender systems, but its performance degrades sharply in practice due to the sparsity and bias of the real-world user feedback data. In this paper, we propose a novel counter-factual data augmentation framework AD-AUG to mitigate the impact of the imperfect training data and empower CF models. The key idea of AD-AUG is to answer the counterfactual question: "what would be a user's feedback if his previous purchase history had been different?". Our framework is composed of an augmenter model and a recommender model. The augmenter model aims to generate counterfactual user feedback based on the observed ones, while the recommender leverages the original and counterfactual user feedback data to provide the final recommendation. In particular, we design two adversarial learning-based methods from both "bottom-up" data-oriented and "top-down" model-oriented perspectives for counterfactual learning. Extensive experiments on three real-world datasets show that the AD-AUG can greatly enhance a wide range of CF models, demonstrating our framework's effectiveness and generality.
收起
摘要 :
This study investigates the influence of leakage through the apex seal, fuel type and recess size on the performance of a Rotary Engine. A commercial CFD software, Fluent, was applied and a two-dimensional model was constructed to...
展开
This study investigates the influence of leakage through the apex seal, fuel type and recess size on the performance of a Rotary Engine. A commercial CFD software, Fluent, was applied and a two-dimensional model was constructed to predict the engine performance. Air and fuel flow into the chamber by two separate ducts in this model were used to simulate fuel-air mixing. Three different apex seal clearances (0mm, 0.4mm, 0.5mm) were simulated. The computations with two types of fuel, CH_4 and C_8H_(18), were performed and put in result comparison in this study. The recess sizes were based on three compression ratios, 8.33, 9.55 and 10.18. To simplify rotor mesh construction in the model with leakage, a porous region was generated to model this quite small gap. The generation of a porous region can greatly reduce mesh sizes and stabilize the numerical iterations. In the leakage analysis, the result showed that the formation of clockwise and counterclockwise vortices was induced by the rotor rotation when the air and fuel flow is charged into the intake chamber. In the compression chamber, the vortex close to the burning chamber makes fuel/air mixing better when the clearance increases. But in the opposite, the larger clearance is, the less amount of fuel is into the chamber due to the leakage. Also with leakage, the high pressure fluid in the combustion chamber went through the seal clearance into the neighbor compression chamber, and that drives the flame moving toward the compression chamber. Better fuel/air mixing was achieved and reacted, and thus the burning zone extended. Although the burning area increased and a better mixing was achieved in accordance with increasing leakage, lower pressure in the working chamber caused the indicated mean effective pressure to drop and then the performance of the engine decreased. Comparing with the numerical result of different fuels, C_8H_(18) had a better combustion in the chamber and produced higher power. Decreasing the recess size, which increased the compression ratio, was beneficial to building up the pressure in the working chamber before ignition, and thus higher pressure in the burning stage resulted in higher power output.
收起
摘要 :
The problem of prediction of resistance to flow in open channels has attracted the attention of hydraulic engineers for a long time. Over the last few decades, the Nikuradse's concept of an equivalent sand grain roughness size has...
展开
The problem of prediction of resistance to flow in open channels has attracted the attention of hydraulic engineers for a long time. Over the last few decades, the Nikuradse's concept of an equivalent sand grain roughness size has been successfully used in a wide variety of problems involving hydraulic resistance to flow. However, when it comes to practical implications such as vegetated flows, where the relative spacing, size and shape of the roughness elements are particularly important features, then the use of an equivalent roughness size becomes problematic because of the non uniformity of the boundary roughness. To overcome this problem, many researchers have investigated extensively using artificial roughnesses on the channel boundary, to predict the friction factor in rigid boundary open channel flow. By drawing analogies between skin friction and form drag, this study proposes a new friction factor function based on the roughness density. This new formula has been verified using data acquired from several flume experiments, carried out for flow fields with vegetation simulated by artificial cylinders. Results indicate that, for a given geometrical shape, the maximum friction occurs at a particular roughness density, when the bed is fully covered by large eddies. After this the flow resistance starts to decrease irrespective of the increase in roughness density. This variation should be quite similar in all the cases of geometrical shapes of roughness elements. Further this method provides a useful basis for practicing engineers to calculate the friction factor in vegetated flows, flow around bridge piers, etc. It also assists in the determination of when the maximum friction would occur, developing a significant step towards the better understanding of the resistance mechanism.
收起
摘要 :
The problem of prediction of resistance to flow in open channels has attracted the attention of hydraulic engineers for a long time. Over the last few decades, the Nikuradse's concept of an equivalent sand grain roughness size has...
展开
The problem of prediction of resistance to flow in open channels has attracted the attention of hydraulic engineers for a long time. Over the last few decades, the Nikuradse's concept of an equivalent sand grain roughness size has been successfully used in a wide variety of problems involving hydraulic resistance to flow. However, when it comes to practical implications such as vegetated flows, where the relative spacing, size and shape of the roughness elements are particularly important features, then the use of an equivalent roughness size becomes problematic because of the non uniformity of the boundary roughness. To overcome this problem, many researchers have investigated extensively using artificial roughnesses on the channel boundary, to predict the friction factor in rigid boundary open channel flow. By drawing analogies between skin friction and form drag, this study proposes a new friction factor function based on the roughness density. This new formula has been verified using data acquired from several flume experiments, carried out for flow fields with vegetation simulated by artificial cylinders. Results indicate that, for a given geometrical shape, the maximum friction occurs at a particular roughness density, when the bed is fully covered by large eddies. After this the flow resistance starts to decrease irrespective of the increase in roughness density. This variation should be quite similar in all the cases of geometrical shapes of roughness elements. Further this method provides a useful basis for practicing engineers to calculate the friction factor in vegetated flows, flow around bridge piers, etc. It also assists in the determination of when the maximum friction would occur, developing a significant step towards the better understanding of the resistance mechanism.
收起
摘要 :
Complete and accurate road network information is an important basis in the detection of EHV transmission lines, and regular updates of road distribution near transmission lines are necessary and meaningful. However, no relevant r...
展开
Complete and accurate road network information is an important basis in the detection of EHV transmission lines, and regular updates of road distribution near transmission lines are necessary and meaningful. However, no relevant research has been found for this application area, and coupled with the fact that roads themselves are significantly challenging, extracting roads with good connectivity and integrity in remote sensing images remains a problem to be solved. Therefore, in this paper, we develop a new end-to-end road extraction network. Multiple Attention Networks (MANet). Specifically, by fusing convolutional and self-attentive approaches, we focus on global contextual features to obtain an effective feature map. In addition, the Strip Multi-scale Channel Attention (SMCA) module is specifically designed for the line features of roads, focusing on extracting row and column features, while the Edge-aware Module (EAM) is used to extract connected and complete roads, aided by edge information. Meanwhile, in order to enhance the practicality of the study, a Mengxi Transmission Line Road Dataset was constructed independently following the processing process of remote sensing images in industrial production. By conducting relevant quantitative and qualitative experiments on this dataset and the publicly available CHN6-CUG dataset, it is fully verified that the method in this paper is superior to other advanced methods and can still extract roads with strong connectivity in complex backgrounds, which has good potential and outstanding advantages in practical applications.
收起
摘要 :
LSTM's special gate structure and memory unit make it suitable for solving problems that are related to time series. It has excellent performance in the fields of machine translation and reasoning. However, LSTM also has some shor...
展开
LSTM's special gate structure and memory unit make it suitable for solving problems that are related to time series. It has excellent performance in the fields of machine translation and reasoning. However, LSTM also has some shortcomings, such as low parallelism, which leads to insufficient computing speed. Some existing optimization ideas only focus on one of the software and hardware. The former mostly focuses on model accuracy, and CPU accelerated LSTM doesn't dynamically adjust to network characteristics; While the latter can be based on the LSTM model structure. Customized accelerators are often limited by the structure of LSTM and cannot fully utilize the advantages of the hardware. This paper proposed a multi-layer LSTM optimization scheme based on the idea of software and hardware collaboration. We used the pruning by row scheme to greatly reduce the number of parameters while ensuring accuracy, making it adapt to the parallel structure of the hardware. From the perspective of software, the multi-layer LSTM module was analyzed. It was concluded that some neurons in different layers could be calculated in parallel. Therefore, this paper redesigned the computational order of the multilayer LSTM so that the model guaranteed its own timing properly and it was hardware friendly at the same time. Experiments showed that our throughput increased by 10x compared with the CPU implementation. Compared with other hardware accelerators, the throughput increased by 1.2x-1.4x, and the latency and resource utilization had also been improved.
收起
摘要 :
Snake-like cable-driven redundant robots (SCDRR) have great application prospects in narrow and complex environments, thus are being widely research in recent years. However, the effect of links length distribution on workspace is...
展开
Snake-like cable-driven redundant robots (SCDRR) have great application prospects in narrow and complex environments, thus are being widely research in recent years. However, the effect of links length distribution on workspace is rarely studied. Meanwhile, the stiffness optimization of SCDRR is an important issue, due to its relatively poor stiffness performance compared with industrial robots. This paper first analyzes the relationship between workspace area, workspace volume and manipulability to illustrate the effect of links length distribution on workspace. Then stiffness optimization model with and without the variation of the Jacobian matrix can be established by statics and stiffness model. Lastly, simulations show that the optimization result between two models is not much different when cables tension are relatively low, while the simplify one has less computational complexity. The workspace analysis in this paper can provide guidance for SCDRR design. And the comparison of two proposed optimization model can provide a basis for the research of variable stiffness control.
收起
摘要 :
Snake-like cable-driven redundant robots (SCDRR) have great application prospects in narrow and complex environments, thus are being widely research in recent years. However, the effect of links length distribution on workspace is...
展开
Snake-like cable-driven redundant robots (SCDRR) have great application prospects in narrow and complex environments, thus are being widely research in recent years. However, the effect of links length distribution on workspace is rarely studied. Meanwhile, the stiffness optimization of SCDRR is an important issue, due to its relatively poor stiffness performance compared with industrial robots. This paper first analyzes the relationship between workspace area, workspace volume and manipulability to illustrate the effect of links length distribution on workspace. Then stiffness optimization model with and without the variation of the Jacobian matrix can be established by statics and stiffness model. Lastly, simulations show that the optimization result between two models is not much different when cables tension are relatively low, while the simplify one has less computational complexity. The workspace analysis in this paper can provide guidance for SCDRR design. And the comparison of two proposed optimization model can provide a basis for the research of variable stiffness control.
收起
摘要 :
This study is aimed at investigating the effects of instructional materials developed with virtual reality (VR) technology on middle school students' learning achievements and motivation towards the course of physics in China. In ...
展开
This study is aimed at investigating the effects of instructional materials developed with virtual reality (VR) technology on middle school students' learning achievements and motivation towards the course of physics in China. In this study, the authors designed a VR-based first-person perspective physics learning platform for students to learn the "velocity-time graph of uniform variable rectilinear motion". A between-subjects design was used and a total of 23 eighth graders took part in this experiment. The students were assigned to two groups on the basis of their background knowledge. The experimental group adopted VR-based learning platform for the learning activities, while the control group watched a video containing the same learning contents. The results show that students from the VR-based experimental group reported higher learning achievements compared with those from the control group. The results also show that students' levels of motivation in the experimental group were higher than those from control group. The findings of this study imply that VR technology can be utilized as an effective instructional tool for fostering students' learning of the basic physics concepts and laws in middle school.
收起