Recommendation of Clip Templates Based on Cross-Modal Retrieval

Abstract

Nowadays, the use of video editing software has increased dramatically. However, there is a problem of insufficient intelligence in the recommendation of clip templates in these software. Therefore, this paper addresses this problem and devotes to combining machine learning algorithms and deep learning to achieve optimization of video clip template recommendation, and proposes the design of a clip template recommendation system based on cross-modal retrieval technology. Firstly, the Requests module is used to crawl some data from Baidu images and NetEase cloud music websites and store them persistently as components of user templates to make the templates diverse and meet the needs of more users. Secondly, the algorithm network construction based on PyTorch framework was completed to realize background replacement and music matching, improve the template matching mechanism for users, and generate videos from images; finally, the Android Studio platform was used to develop the APP for Android system, and the Web server was built to realize the data interaction between the client side and the server side, so that users can easily use the APP to get functional experience.