An Improved Cartoon Character Avatar Generation Algorithm Based on GAN

Aimed at solving the problems of poor diversity of generated results and generally blurred images in traditional image generation methods, an improved cartoon character avatar generation algorithm WGAN-GP based on GAN is proposed. During the process, this method is used for generating image. By using transposed convolutional layers and inserting BN layers in the middle of each convolutional layer to define activation function as Leaky_ReLU as well as using Wasserstein distance instead of JS divergence as the adversarial loss function of the network and choosing RMSProp as the optimizer, improvements of our method are realized. The experimental results show that, the improved model solves the problems that the network is difficult to converge and the model is easy to collapse, and improves its generation diversity. Meanwhile, the details and texture of the generated images are clear and close to the sample set, which proves the effectiveness of the proposed algorithm.