From the view of information flow, a super-network equilibrium optimization model is proposed to compute the solution of the operation architecture which is made up of a perceptive level, a command level and a firepow...From the view of information flow, a super-network equilibrium optimization model is proposed to compute the solution of the operation architecture which is made up of a perceptive level, a command level and a firepower level. Firstly, the optimized conditions of the perceptive level, command level and firepower level are analyzed respectively based on the demand of information relation,and then the information supply-and-demand equilibrium model of the operation architecture super-network is established. Secondly,a variational inequality transformation(VIT) model for equilibrium optimization of the operation architecture is given. Thirdly, the contraction projection algorithm for solving the operation architecture super-network equilibrium optimization model with fuzzy demands is designed. Finally, numerical examples are given to prove the validity and rationality of the proposed method, and the influence of fuzzy demands on the super-network equilibrium solution of operation architecture is discussed.展开更多
目的实时渲染图形程序(如游戏、虚拟现实等)对高分辨率和高刷新率的要求越来越高,因此,针对渲染图像的实时超分辨率技术在实时渲染中非常必要。然而,现有的视频超分算法和实时渲染处于不同的数据处理管线之中,这导致其难以直接应用到实...目的实时渲染图形程序(如游戏、虚拟现实等)对高分辨率和高刷新率的要求越来越高,因此,针对渲染图像的实时超分辨率技术在实时渲染中非常必要。然而,现有的视频超分算法和实时渲染处于不同的数据处理管线之中,这导致其难以直接应用到实时渲染管线里。方法对此,提出了一个基于帧循环结构的实时神经超采样方法。充分利用实时渲染管线中生成的低分辨场景几何数据,以提升超采样网络对于三维空间信息的感知力;将帧循环框架结合到超采样方法中,通过引入先前帧重建结果的特征来改善当前帧的重建结果,从而实现时间尺度上的稳定性;将重加权网络和注意力网络置于特征提取模块中,以提升提取到的特征的有效性。此外,本文还提出了一个面向神经超采样的实时渲染流程,该流程能够将超采样网络部署至图形计算管线之上,并与实时渲染管线相结合。结果与同样能够实时且效果较好的基准方法面向实时渲染的神经超采样(neural super-sampling for real-time rendering,NSRR)比较,本文方法在速度少许提升的前提下,图像质量指标峰值信噪比(peak signal to noise ratio,PSNR)平均提升了0.4 dB,并在部署到实时渲染管线后,通过轻量化裁剪继续保持实时性且部分场景效果仍然优于非实时的部署后NSRR;在网络模块的消融实验中也证明了各个子模块对于神经超采样任务的有效性。结论本文提出的神经超采样网络模型与搭建的神经超采样渲染流程,在取得更好效果的同时具有一定的实用价值。展开更多
基金supported by the National Natural Science Foundation of China (71771216,71701209)Shaanxi Natural Science Foundation (2019 JQ-250)。
文摘From the view of information flow, a super-network equilibrium optimization model is proposed to compute the solution of the operation architecture which is made up of a perceptive level, a command level and a firepower level. Firstly, the optimized conditions of the perceptive level, command level and firepower level are analyzed respectively based on the demand of information relation,and then the information supply-and-demand equilibrium model of the operation architecture super-network is established. Secondly,a variational inequality transformation(VIT) model for equilibrium optimization of the operation architecture is given. Thirdly, the contraction projection algorithm for solving the operation architecture super-network equilibrium optimization model with fuzzy demands is designed. Finally, numerical examples are given to prove the validity and rationality of the proposed method, and the influence of fuzzy demands on the super-network equilibrium solution of operation architecture is discussed.
文摘目的实时渲染图形程序(如游戏、虚拟现实等)对高分辨率和高刷新率的要求越来越高,因此,针对渲染图像的实时超分辨率技术在实时渲染中非常必要。然而,现有的视频超分算法和实时渲染处于不同的数据处理管线之中,这导致其难以直接应用到实时渲染管线里。方法对此,提出了一个基于帧循环结构的实时神经超采样方法。充分利用实时渲染管线中生成的低分辨场景几何数据,以提升超采样网络对于三维空间信息的感知力;将帧循环框架结合到超采样方法中,通过引入先前帧重建结果的特征来改善当前帧的重建结果,从而实现时间尺度上的稳定性;将重加权网络和注意力网络置于特征提取模块中,以提升提取到的特征的有效性。此外,本文还提出了一个面向神经超采样的实时渲染流程,该流程能够将超采样网络部署至图形计算管线之上,并与实时渲染管线相结合。结果与同样能够实时且效果较好的基准方法面向实时渲染的神经超采样(neural super-sampling for real-time rendering,NSRR)比较,本文方法在速度少许提升的前提下,图像质量指标峰值信噪比(peak signal to noise ratio,PSNR)平均提升了0.4 dB,并在部署到实时渲染管线后,通过轻量化裁剪继续保持实时性且部分场景效果仍然优于非实时的部署后NSRR;在网络模块的消融实验中也证明了各个子模块对于神经超采样任务的有效性。结论本文提出的神经超采样网络模型与搭建的神经超采样渲染流程,在取得更好效果的同时具有一定的实用价值。
