;Since the development of technalogy , network becomes more ubiquitous and advanced rapidly . Meanwhile , people are growing their demands and expectations for higher quality, and this trend extends to images with high resolution. To deal with the effective compression of massive data , VVC adopts various techniques , such as QTMT and Rate Distortion Optimal . However , these precision processes also result in high complexity in coding calculations . Then , our work aims at combin ing popular machine learning and deep learning , applying them to VVC inter prediction . At the beginning , we use machine learning method Sup port Vector Method on Coding Units(CUs) partition , and then employ the deep learning method Convolutional Neural Network for further refine ment . Finally, integrating the CU-PU Decision algorithm and using it to determine the final partition for the groups defined by SVM-CNN allows simple blocks to skip the time-consuming Rate-Distortion Optimization (RDO). After the correct partitioning mentioned above, the experimental results show an average BDBR gain of -2.03%, with a total time-saving of 49.59% compared to VVC.
