Scene Style Conversion Algorithm of AI Digital Host: A Deep Learning Approach

Abstract

In the recent years, image diversity generation algorithms based on deep learning have gradually mined feature information that can describe the essential content of images through hierarchical learning, which has become a current research hotspot. In particular, the emergence of GANs has made the qualitative leap in the speed and quality of image data generation due to its good autonomous generation ability and ingenious weakly supervised learning mode. In this study, the focus is deep learning based scene style conversion algorithm of the AI digital host. The proposed model has 2 aspects: (1) Efficient face modelling. In this step, the face features are extracted and combined with deep neural networks to obtain the efficient representations, then, the rough stitching of point cloud data is applied to construct the different perspectives the faces. (2) Efficient style conversion algorithm. This study designs an improved GAN algorithm to create new image conversion scheme to complete the design of the AI host. In the experiment section, the mainstream evaluation methods are adopted to test the performance.