基于DeblurGANv2的图像去模糊技术的轻量化应用  

Lightweight application of image deblurring technology based on DeblurGANv2

作　　者：黄萍 管丽鹏 朱惠娟[1] HUANG Ping;GUAN Li-peng;ZHU Hui-juan(Nanjing University of Science and Technology Zijin College,Nanjing 210023,China)

机构地区：[1]南京理工大学紫金学院,南京210023

出　　处：《信息技术》2024年第10期49-55,共7页Information Technology

基　　金：江苏省大学生创新创业训练项目成果(20211365-4011H);南京理工大学紫金学院教育教学改革与研究课题成果(20220103001);南京理工大学紫金学院精品课程成果(03101021)。

摘　　要：随着深度学习算法的发展,人们开始研究使用深度学习模型来解决图像模糊的问题。性能优异的DeblurGANv2在图像去模糊应用领域取得了不错的效果,但由于算法的复杂性,对模型运行设备的性能要求也会更高。为了能够将DeblurGANv2模型移植到Android终端上应用,在研究了基于DeblurGANv2模型图像去模糊技术在电脑端应用基础上,用Pytorch对DeblurGANv2模型进行转化,缩小了模型的体积和运算时间,将模型大小由233MB压缩为13.2MB,使模型最终可以在Android系统上运行,运行时间较电脑端模型也有所下降,为其他深度学习算法向移动端或者低性能设备上的部署提供了思路和解决方案。With the development of deep learning algorithms,people begin to study the use of deep learning models to solve the problem of image blurring.DeblurGANv2 with excellent performance has achieved good results in the field of image deblurring applications,but due to the complexity of the algorithm,the performance requirements for model running devices are also higher.In order to be able to transplant the DeblurGANv2 model to Android terminals for application,on the basis of studying the application of image deblurring technology based on the DeblurGANv2 model on the computer terminal,Python is used to convert the DeblurGANv2 model,reducing the size and computing time of the model,compressing the model size from 233MB to 13.2MB,making the model ultimately able to run on the Android system,and the running time is also reduced compared to the computer side model.The application provides ideas and solutions for the deployment of other deep learning algorithms to mobile terminals or low-performance devices.

关 键 词：图像处理 图像去模糊 深度学习 对抗生成网络 移动应用开发 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]