Anime Avatar Generation Using C-DCGAN

With the rapid development of animation culture in today's society,a large number of a:nimation characters based on personal preferences need to be quickly created.Nowadays,Generative adversarial networks(GAN)can solve this problem.GAN and its derivative models have many explorations in the field of image generation.Different GAN models can accomplish different tasks in a targeted manner.In this work,C-DCGAN is selected to explore the generation of anime character avatars under specific conditions.The conditional-DCGAN(C-DCGAN)is a subtle combination of GAN and DCGAN,which combines the advantages and features of these two models.Consider:ing the properties of C-DCGAN,we want to find out the best generation model for this task by exploring different training modes and model structure of C-DCGAN.At the same time,in order to generate high quality images,a new labeled data set is created.We use the techniques of web crawling,face detection and k-means clustering to collect and preprocess data sets.This high-quality data set helps the model complete training and generate images that achieve the desired goals.