摘要
基于Johoson等的图像风格转换网络模型,在保证网络性能的前提下,在原有的网络结构上,通过运用更高效的网络计算方法对原有残差网络进行优化。实验结果表明,改进后的方法在几乎不降低图像质量的前提下,一定程度上克服了图像风格迁移模型存储量大、计算代价高、计算资源消耗大、难以移植到移动端的问题。
In this study,we propose an efficient network computing method based on Johoson’s image style transformation network model to optimize the original residual network for ensuring suitable network performance.The conducted experiments prove that the proposed method can solve the following problems:high storage and calculation cost associated with the image style transformation network model;massive consumption of the computing resources;and difficulty with respect to the transplantation to a mobile terminal without reducing the image quality.
作者
裴斐
刘进锋
李崤河
Pei Fei;Liu Jinfeng;Li Xiaohe(College of Information Engineering,Ningxia University,Yinchuan,Ningxia 750021,China)
出处
《激光与光电子学进展》
CSCD
北大核心
2020年第6期219-225,共7页
Laser & Optoelectronics Progress
基金
国家自然科学基金(61762073)
宁夏高等学校科学研究项目(NGY2015044)。
关键词
图像处理
图像风格迁移
卷积神经网络
深度残差网络
模型压缩
image processing
image style transformation
convolutional neural network
deep residual network
model compression