The invariant sets and the solutions of the 1+2-dimensional generalized thin film equation are discussed. It is shown that there exists a class of solutions to the equations, which are invariant with respect to the se...The invariant sets and the solutions of the 1+2-dimensional generalized thin film equation are discussed. It is shown that there exists a class of solutions to the equations, which are invariant with respect to the set $$E_0 = \{ u:u_x = v_x F(u),u_y = v_y F(u)\} ,$$ where v is a smooth function of variables x, y and F is a smooth function of u. This extends the results of Galaktionov (2001) and for the 1+1-dimensional nonlinear evolution equations.展开更多
为解决角度变化下的人脸检测中存在参数量大及角度幅度变量小的问题,提出区域渐进校准网络用于任意平面角度的人脸检测,通过级联网络结构降低角度变化、提升网络运行速度。采用区域生成网络产生高质量的候选区域,构造渐进校准网络,逐步...为解决角度变化下的人脸检测中存在参数量大及角度幅度变量小的问题,提出区域渐进校准网络用于任意平面角度的人脸检测,通过级联网络结构降低角度变化、提升网络运行速度。采用区域生成网络产生高质量的候选区域,构造渐进校准网络,逐步缩小面部平面角度变化范围,同时由粗到细地对候选区域执行面部检测。其中,特征提取的中间层融合参数量较少时,更好地表示了面部特征,调整锚的设置解决小尺度面部问题。在角度增强的FDDB(face detection data set and benchmark)数据集与WIDER FACE数据集上的实验结果表明,提出的方法分别取得了89.1%与90.4%的平均召回率,准确度高于快速区域卷积神经网络(Faster RCNN),且运行速度更快。在实际项目中使用该算法,验证了该方法的有效性及可行性。展开更多
基金This work was supported by the National Natural Science Foundation of China (Grant No. 10671156)the Program for New Century Excellent Talents in Universities (Grant No. NCET-04-0968)
文摘The invariant sets and the solutions of the 1+2-dimensional generalized thin film equation are discussed. It is shown that there exists a class of solutions to the equations, which are invariant with respect to the set $$E_0 = \{ u:u_x = v_x F(u),u_y = v_y F(u)\} ,$$ where v is a smooth function of variables x, y and F is a smooth function of u. This extends the results of Galaktionov (2001) and for the 1+1-dimensional nonlinear evolution equations.
文摘为解决角度变化下的人脸检测中存在参数量大及角度幅度变量小的问题,提出区域渐进校准网络用于任意平面角度的人脸检测,通过级联网络结构降低角度变化、提升网络运行速度。采用区域生成网络产生高质量的候选区域,构造渐进校准网络,逐步缩小面部平面角度变化范围,同时由粗到细地对候选区域执行面部检测。其中,特征提取的中间层融合参数量较少时,更好地表示了面部特征,调整锚的设置解决小尺度面部问题。在角度增强的FDDB(face detection data set and benchmark)数据集与WIDER FACE数据集上的实验结果表明,提出的方法分别取得了89.1%与90.4%的平均召回率,准确度高于快速区域卷积神经网络(Faster RCNN),且运行速度更快。在实际项目中使用该算法,验证了该方法的有效性及可行性。
