基于Tensorflow.js的DIY图像风格迁移系统开发  

Development of DIY image style migration system based on Tensorflow.js

作　　者：陈晓渊 李云飞 陈维曦 谢枭鹏 向毅 Chen Xiaoyuan;Li Yunfei;Chen Weixi;Xie Xiaopeng;Xiang Yi(School of Intelligent Technology and Engineering,Chongqing University of Science and Technology,Chongqing 401331,China)

机构地区：[1]重庆科技学院智能技术与工程学院,重庆401331

出　　处：《无线互联科技》2021年第21期47-48,共2页Wireless Internet Technology

基　　金：重庆市教改重大项目.项目编号:201016;重庆市大学生创新创业训练计划项目.项目编号:S202011551024;重庆市研究生科研创新项目.项目编号:CYS20364。

摘　　要：针对目前各类美图软件美化效果单一、风格雷同的问题,文章开发一套基于快速风格迁移的DIY图像风格迁移系统。系统通过COCOS游戏引擎和JS等工具完成网页的UI设计与数据的获取处理,利用深度学习技术,使用Google公司的Tensorflow.js开源框架搭建风格迁移网络,并完成模型迁移及对JS获取到的风格图像、内容图像数据的转换,实现将用户选定图片的风格迁移到指定的图片上。同时,系统支持用户指定再训练轮次并对模型重新离线训练,调整风格迁移效果。Aiming at the problems of single beautification effect and similar styles of various Meitu softwares,a set of DIY image style migration system based on rapid style migration was developed.Use COCOS game engine and JS tools to complete the UI design and data acquisition and processing of web pages,use deep learning technology,use Google’s Tensorflow.js open source framework to build a style migration network,and complete model migration and style images acquired by JS.The conversion of content image data realizes the transfer of the style of the picture selected by the user to the specified picture.At the same time,the system supports users to specify retraining rounds and retrain the model offline to adjust the style transfer effect.

关 键 词：深度学习 Tensorflow.js 风格迁移 离线训练 DIY 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]