超分辨率重建在视频的传输和显示中起着重要的作用。为了既保证重建视频的清晰度,又面向用户实时显示,提出了一种采用精简卷积神经网络的快速视频超分辨率重建方法。所提的精简卷积神经网络体现在以下三点:首先,考虑到输入的尺寸大小会...超分辨率重建在视频的传输和显示中起着重要的作用。为了既保证重建视频的清晰度,又面向用户实时显示,提出了一种采用精简卷积神经网络的快速视频超分辨率重建方法。所提的精简卷积神经网络体现在以下三点:首先,考虑到输入的尺寸大小会直接影响网络的运算速度,所提网络省去传统方法的预插值过程,直接对多个低分辨率输入视频帧提取特征,并进行多维特征通道融合。接着,为了避免网络中产生零梯度而丢失视频的重要信息,采用参数线性纠正单元(Parametric Rectified Linear Unit,PReLU)作为激活函数,并采用尺寸更小的滤波器调整网络结构以进行多层映射。最后,在网络末端添加反卷积层上采样得到重建视频。实验结果显示,所提方法相比有代表性的方法在PSNR和SSIM指标上分别平均提升了0.32dB和0.016,同时在GPU下达到平均41帧/秒的重建速度。结果表明所提方法可快速重建质量更优的视频。展开更多
In this paper,the role of constant optimal forcing(COF) in correcting forecast models was numerically studied using the well-known Lorenz 63 model.The results show that when we only consider model error caused by para...In this paper,the role of constant optimal forcing(COF) in correcting forecast models was numerically studied using the well-known Lorenz 63 model.The results show that when we only consider model error caused by parameter error,which also changes with the development of state variables in a numerical model,the impact of such model error on forecast uncertainties can be offset by superimposing COF on the tendency equations in the numerical model.The COF can also offset the impact of model error caused by stochastic processes.In reality,the forecast results of numerical models are simultaneously influenced by parameter uncertainty and stochastic process as well as their interactions.Our results indicate that COF is also able to significantly offset the impact of such hybrid model error on forecast results.In summary,although the variation in the model error due to physical process is time-dependent,the superimposition of COF on the numerical model is an effective approach to reducing the influence of model error on forecast results.Therefore,the COF method may be an effective approach to correcting numerical models and thus improving the forecast capability of models.展开更多
文摘超分辨率重建在视频的传输和显示中起着重要的作用。为了既保证重建视频的清晰度,又面向用户实时显示,提出了一种采用精简卷积神经网络的快速视频超分辨率重建方法。所提的精简卷积神经网络体现在以下三点:首先,考虑到输入的尺寸大小会直接影响网络的运算速度,所提网络省去传统方法的预插值过程,直接对多个低分辨率输入视频帧提取特征,并进行多维特征通道融合。接着,为了避免网络中产生零梯度而丢失视频的重要信息,采用参数线性纠正单元(Parametric Rectified Linear Unit,PReLU)作为激活函数,并采用尺寸更小的滤波器调整网络结构以进行多层映射。最后,在网络末端添加反卷积层上采样得到重建视频。实验结果显示,所提方法相比有代表性的方法在PSNR和SSIM指标上分别平均提升了0.32dB和0.016,同时在GPU下达到平均41帧/秒的重建速度。结果表明所提方法可快速重建质量更优的视频。
基金sponsored by the National Basic Research Program of China(Grant No.2012CB955202)the Knowledge Innovation Program of the Chinese Academy of Sciences(Grant No.KZCX2-YW-QN203)the National Natural Science Foundation of China(Grant No.41176013)
文摘In this paper,the role of constant optimal forcing(COF) in correcting forecast models was numerically studied using the well-known Lorenz 63 model.The results show that when we only consider model error caused by parameter error,which also changes with the development of state variables in a numerical model,the impact of such model error on forecast uncertainties can be offset by superimposing COF on the tendency equations in the numerical model.The COF can also offset the impact of model error caused by stochastic processes.In reality,the forecast results of numerical models are simultaneously influenced by parameter uncertainty and stochastic process as well as their interactions.Our results indicate that COF is also able to significantly offset the impact of such hybrid model error on forecast results.In summary,although the variation in the model error due to physical process is time-dependent,the superimposition of COF on the numerical model is an effective approach to reducing the influence of model error on forecast results.Therefore,the COF method may be an effective approach to correcting numerical models and thus improving the forecast capability of models.
